Off course not there are no problems. It`s just a faulty windows install. Or some incompatibility issues. Don`t get fooled. The problems mentioned can never cause what he is showing, they are way to small.
For AMD Users (AM4/AM5): Disabling USB selective suspend setting in the power plan options really made an immediate, noticeable difference in both LatencyMon and real-world performance. Try it out.
@@vmafarah9473 I only have a 9900k/Z390 as my newest Intel platform but it probably does help. Unfortunately the ultimate limiting factor are Nvidia's drivers. There's just no way to control the problem that they have.
Why not just use an NVIDIA LDAT to measure latency? Also would be really interested to see how the 5800x3D and 7800x3D as well as 11th gen intel compare in these types of testing.
11th Gen was so similar to 10th Gen people were almost upset by it. There’s fewer cores and a slight ipc and single core uplift. This does improve performance in a few games but has it lose in some games. I would assume the latency difference would be negligible between them on Windows 10.
@@quintrapnell3605I think the upset was more that it was on the same process node and that the 11900k had no reason to exist with the 8 core limit. In case u weren't aware the 11900k had about double the single threaded uplift the 10900k had over the 9900k and that was with the 10900k getting an extra 300mhz while the 11900k got none. 11th gen was much higher ipc than 10th gen. That's also why it only had 8 cores, the cores had much more transistors than comet lake cores bcuz they actually are signicantly different, so when they were forced to backport the cores to 14nm, they could only fit 8 of them on the die.
would love to see the further testing with more architectures :) I'm especially curious how AMD compares (Zen 2/3 mainly, though I don't know how different the various AMD gens are)
I'm really curious if Dr. Ian Cutress would chime in, as he is a brilliant person, and he might have some contacts within Intel that could shed a bit of light without breaking any NDAs or such. You nailed my question ahead of time. Does this happen with AMD CPUs as well? :)
It could be the e-cores. They are all clumped together on one side of the die, so even if the CPU wanted to start using a P-core it could have latency going from an e-core over to a p-core. Maybe see if disabling the e-cores makes a difference
@@lophilip ah, I see. I must've missed that. Next I need to look at the size of the big cores. The 13900k simply has a physically larger die than the 10900k. So maybe extra latency is added there. But maybe it's just from the e-cores. Which is why we need to look at the p-cores specifically. If not that, I'd say it's probably just some architectural issue that we can't really solve
Adding more cache to the frontside bus would help future CPUs with the little core big core model. We’re talking like a 100MB+ of L3 cache and 300MB+ of L4 cache.
In its essence, it's probably a (Microsoft Windows) kernel or some software issue. To be a real hardware issue, it would either break things or it just doesn't work or support something. Code optimization isn't an easy thing at these low levels without knowing and fully understanding the bigger picture.
Increased latency can definitely be explained by increased distance between different parts of a system, i.e. in this case presumably the CPU and the I/O controller hub. Differences in hardware design can reduce performance in specific situations without breaking anything.
@@VVayVVard Yes, true if you look at it purely electrical and from a signal level. It's like considering an expensive or smaller form factor motherboard board. Mostly these issues are "noticeable/measurable" on a system level between chipset and CPU and various IO. Not on the CPU itself, they are either somewhere between specification or not (true you can have little benefits in better silicon quality) but, to say the translation and interpretation of these lightning fast and tinny differences (noticeable in the OS) in a pure hardware issue ... you put a lot of faith in multiple perfect software overhead layers to perform flawless. But hypothetical from a theory perspective, you are right. Now I don't remember, but I thought the hypothesis here was a new form of branch prediction between gens, no? So that dives into low-level x86 instructions and code, in my opinion, not latency cost by physics.
Sounds like you may be interested in getting an Optane drive. They are on a clearance since earlier this year, due to end of production. I used Optane 905p for photo editing, and Lightroom and Photoshop became a lot snappier. Access latency is 10 us vs 100 us of NAND.
I was just talking about this "feeling" with a friend. Between an i9-9900KF and r9-5900. The Intel chip just feels snappy when multitasking and doing a bunch of small ops rapid fire. Kind of like doing a drag race with a diesel versus gas. Intel would be the gas and AMD the diesel. (Just for this comparison on these two chips) doing 2 runs, one with no weight, One with a 5,000 lb trailer behind it. The 9900kf feels instant right off the line but then bogs down on larger things. The 5900 feels slower off the line but pulls ahead quickly with the larger ops.
This is interesting, not the first time I hear that. Obviously 9900K is snappier cause it's monolithic with lower latencies, and slower in big apps because it has less cores. (I guess 10900k would be a better comparison here). But microtasks seem more important for a good pc feeling. Have a look at my channel with tuned i9-9900K, there'll soon be a video comparing 4.7ghz 3200cl16 49ns vs 5ghz tuned 4266cl16 33ns. The difference is Mind-Blowing. It is also as good as tuned 5900X in games.
This kinda reminded me a bit of when I used to have a 3300X and later changed to the 5900X, for the good ol' drag n drop files the 3300X was superbly snappy compared to my 5900X, all due of having the single CCD. Still it's not something as notorious as you have showed here with the 10th vs 13th gen Intel, but I do agree on the moment you changed hardware, there's things that you used to do that are pretty basic, that feel "off"
Would be even more interesting to go further back down the rabbit hole with X58 and Win 7. I expect you'll find similar results. For years I thought I was tripping with my intel systems, the best of which was my old X58 920 D0 banger at 4.4, closely followed by the Sandy Bridge 5ghz 2500k both lightning fast in day to day snappiness. Don't seem to find the same issues on Ryzen 5 series honestly.
Nvidia Drivers seem to have big issues with latency on some systems, my 4090+5800X3D shows more latency in Latency Mon if I use both monitors (LG C1 OLED and Asus PG279QZ) than when I use just one of them (doesn't matter which one). I also tested a 5900X with both a 3080 and a 6800XT, more latency with the 3080. Maybe something to look into?
@@ilovehotdogs125790 I can't tell really, never had it. But it probably won't take dual rank 4000+ b-die and it makes a big difference in snappiness. So the answer is no I suppose, maybe 8700K :)
@@HeartOfAdel yeah maybe the 8700k, no idea how good the imc is on that chip. The 10900k can do 4400mhz dual rank 32gb. So if the ram speed matters more than the ringbus size then maybe the 10900k is still lower latency.
@ilovehotdogs125790 8700k certainly has a hard time going above dual rank 4000-4200, but 9900ks and 10900k can, not worth it though for the voltage requirement usually. I think all of them will feel about the same.
@@MJ-uk6lu yeah but is it measurably better? Because even on chips like the 10700k, it having 2 less cores didn't help the ringbus since it was just 8 cores with 2 dead cores. So if the lower core chips just have the same die with some disabled cores instead of having a smaller ringbus then will that even help?
I wish I had time to try these tests on various Linux distros. Latency has always been a software variable on those OSes. Just last night, a system update took my AMD 5950X from a sluggish UI experience to smooth. The only other variable, was that I disabled the TPM in BIOS but I doubt this had anything to do with it.
what version (old and new) specifically? what distro? did kernel update too? this might be useful for future reference... I'm also an AMD user but I ditched Linux last year...
@@FeelingShred Pop_OS...It wasn't any "version" - just a routine system update which could have included any number of changes that solved the issue. Using Linux is a mater of principle to me. There are no issues that will ever make me switch back to an evil OS like windows or mac. My motto is, if Linux can't do it, it can be done in a VM. If that doesn't work either, then it doesn't need to be done.
thanks for the info it is well deserved to see it is an architect issue then an entire system issue. And the reason why other enthusiasts have not talk about it because they do not know the information that you already have been told by CPU specialist. Always stay busy and keep giving us good hardware content.
Please test at least one 5000 series AMD CPU on part 3 (5600 and 5800X3D are the most popular). Personally I want to see these CPUs 5600G, 5600, 5800X3D, 7600, 7600X3D. Monolithic (5600G) vs chiplet (5600) design latency probably not too ineresting based on gaming 0,1% lows but who knows maybe there is an advantage for the monolithic chips.
4:30 - How exactly are you measuring input lag to the milliscond with 120fps? The camera would only have ~10ms resolution. You'd need to go to 1000fps for 1ms resolution..
THANK YOU I though i was going NUTS i upgrade from a 10850k to a 12900k about a year ago and when i was moving multiple files in davinci resolve it felt like the 12th gen was slower. Thank you Brian for this Video.
I have a 10th gen laptop for work given to me by work. Although it's only an i3 it's the snappiest PC I've ever used. I'm well impressed and now know my gaming PC is going to be a 10th gen if I go Intel.
Glad these are not the types of problems I have to deal with, given my cheap general purpose build... Huanazhi X99TF ($50-discounted) Xeon E5-2696v3 (TU and UV: -50/-50mV)-used for $46 48GB DDR4 ($80-used) RTX2080Ti ($230-used) PSU 850W Gold (used, $50) 1TB NVME m.2 ($120)
I have my i7-12700K PC that destroys the PCs used on the video for system latency, i built it for music production and emulation. The i7 cost me 330 dollars, 32GB of DDR5 4800 for 250, MSI Z690 Pro A DDR5 for 150, two 1 TB M2 SSDs 120$, i'm looking for a 8GB RX580 for 100 bucks right now (crazy people still selling garbage GT1030s around the same price and they are slower than the Intel UHD 770 i have right now running Crysis).
11700K undervolted @4Ghz all-core on an ASRock Extreme Z590 is my choice for general use. (While having a virtual LAN party running in the background.)
@@puiuus core and ring both x40, 1.15V fixed core, 0.95 fixed sa, 1.10 vio, but 1.40 dram (126gb) Learning infrastructure as code and throw myself several virtual LAN parties. Above 4Ghz isn't worth the extra heat for me. Attempting everything on just 1 computer though will use remote desktop on cheap notebooks/laptops like an open-box ASUS E410 from Best Buy.
The Northbridge / SA is still located directly on the CPU silicon. I'm not saying your impression of additional latency is incorrect but your technical explanation of why it's happening is dead wrong. It is all still a monolithic die. If you don't believe me, look at an open source die shot of a 13900K. You're comparing 2 different chipsets as well as 2 different generations of memory (DDR4 & DDR5) between DDR4 and DDR5 and how the memory controller handles data. It has absolutely nothing to do with Raptor Lake moving the northbridge (which it did not). Whomever you spoke to at Computex either misunderstood you or something got lost in translation.
This is fascinating! I went from a 10900K -13900k and although benchmarks prove the new system is much faster the very day experience is different. For music production and photoshop the 10900k was snappier and a better experience!
C states in your BIOS allow your CPU to shut off which takes quite a while to come back on (anti-snappiness). Simple fix, just turn off the C3 C6 C7 wait states so the CPU stays in C1 all the time and you will likely regain the responsiveness you paid for. Hyperthreading (HT) and Efficiency cores are not your friend if you're in a hurry. Buy a "Kill A Watt Electricity Usage Monitor" to measure the wattage used by your computer with different BIOS settings.
@@H53. The effects of the C states are workload-dependent, i.e., how the application runs, how it is used, how it is compiled, etc. That's a lot of syllables to say turning off C states will always help improve "snappiness" and never hurt performance. This holds true for all CPUs with power management features. A $20 kill-a-watt meter will track electricity usage and provide empirical data to show how much (or not) electricity a computer uses, and the effects of changing the C state settings can easily be seen. I turned off the C states and bought a new display and my net electricity use is lower and my computer is always snappy. Life's no bargain buying a fast CPU only to force it to be janky all the time.
Thread Director adds latency and occasionally directs latency sensitive threads on the e-cores instead of the p-cores. It is also super aggressive on moving threads to e-cores from p-cores in the name of “efficiency” so it can park the p-cores faster. I had a 12600 non k. It was quick and handled Cyberpunk 2077 very well. Upgraded to the 13700k and the only thing I noticed is alt tabbing out of games is insanely fast and acts as if the game isn’t running.
Part 3 should be comparing with Ryzen I guess. Ryzen used to be unresponsive due to slow CCD/CCX and IF but improved over the years with cache, especially with 3D cache.
@@christopherjames9843 Definetly got some slowdowns I didn't have before, but not sure if it's some hardware/software issue I have, or something related to the latency stuff.
As someone very sensitive to latency I found the most common cause is power management and downclocking, disabling low power states will definitely improve this. Same goes for GPUs, I remember my RX550 used to cause mouse movement stutters on idle desktop when the card was repeatedly downclocking and upclocking. However I do appreciate my 13600k idling at
Yeah, wanted to test out latencymon on my 5800X3D and i got like 700nano seconds instantly, until I randomly had task manager open, got low scores, and it clicked. Turned on "high perf." power profile and got sub 100 nanosecond scores.
I genuinely miss my i7 7700k vs my 7950x3D. No parking, no asymmetrical cores, no annoying fluctuating clock frequencies, just a pure and simple processor that WORKED. I hate where things are going in the desktop space. We are regressing because these companies can't keep putting out faster hardware due to Moore's law dying. It's bad times ahead, friends.
@@tobiaspabst55244600 cl16 dual rank? If it is, you better downclock that thing for slower degradation... I've only seen a couple of dudes reaching that on dual rank on apex but it requires 1.35-1.45V SA/IO usually. If it's single rank, 4200-4300mhz will be faster in games with less voltage requirements. Myself rocking 9900K with dr 4266cl16 33.8ns, that cpu is very snappy and fast in games. I have a comparison incoming between tuning vs 3200cl16, crazy difference. That's about 5700x3d in gaming.
Holy Hell...I knew it everyone told me it was all in my head. About 6 months ago I gave my brother by 10850k 32mhz cl17,19,19 system components as I went to 13700k ddr4 4266mhz cl15,16,16. And i 1st noticed on startup once i had same startup programs being loaded as the 10850k system.(wish i had a record of the startup time). I had issues when dragging and dropping many multiple large GTA5 mod files as i usually have 4 or 5 different copies at once. Becae a huge pain would cause me to make stupid mistakes like droping files were i did not want to. Got so bad that i took mobo and cpu back to Parkville Microcenter but since it was 34 days since purchase i could not return but exchange. But i did not figure 2 chips or mobo could be faulty so i exchanged for the same(AHHHHH). And of course there was no change. The guys at microcenter said it was all in my head exept for the manager that told me he had noticed it and about a half dozen other customers complained about it as well. Great video as always Brian!! Thanks again!! I shoulda kept the damn 10th gen system, screw my little bro LOl LOL LOL. I demand reperations from INTEL....Good luck right???
And the i5 12400, if it's an H0 stepping, doesn't just have the P-Cores disabled at a physical level but actually uses a smaller die that only physically has 6 P-Cores (and no E-Cores) to begin with; I thought that might be an advantage too (shorter bus with less stops, etc) vs having a full die but with the E-cores all disabled.
When hardware was a constraint they made software actually run well. Imagine if the conservative programming of the past was applied to today's software on current hardware. Everything would be instantaneous.
It`s a nice CPU but really not snappier than a new one. There are other problems happening here. He`s playing us, the explanations are ridicilously unscientific. A windows reinstall will do miracles.
@@brugj03he hasn't isolated the ring bus by turning off the E cores, didn't even mention it. E cores are for 'Cinebench workloads'. If he isn't using those, turn em off.
In RAM overclocking communities, this isn't super obscure information. I don't know the why, all I know is some architectures are just better than others. I know DDR4 is a contributing factor too because even though it doesn't have nearly as much bandwidth potential, it tends to do better latency-wise. The amount that DDR5 factors into it will probably be less and less as DDR5 matures though. AMD also generally does worse than Intel in terms of latency (Not as in their architecture to architecture variance, but they simply tend to have slightly worse latency results than Intel). The problem is RAM overclocking is fairly niche, especially outside of enthusiast overclocking communities. So not many people are aware of this.
@@username8644 I agree on the first part but not completely on the second. I think it can help to understand the architecture you're dealing with but I don't think it's required. There are some examples that come to mind, like knowing that 12th/13th gen IMCs are largely random and not binned better the higher you go compared to most other modern CPUs, or that locked 12th/13th gen CPUs have a locked SA voltage. Other than that I can't think of much. I'm open to being wrong though and i'm no expert by any means.
@@labombaromba Ram architecture, not CPU. You need to understand all the timings and latency, and how they affect one another. There are literally like a hundred things to know and keep track of if you are serious about ram overclocking. Edit: compared to CPU or GPU overclocking where you really don't need to know much.
@@username8644 Ohhh okay I get you now. You're absolutely right on that. There are so many different variables compared to CPU and GPU overclocking it's crazy.
DDR5 can have worse latency with its first request command compared to DDR4 with higher TCL but anything else including total system latency on DDR5 is superior. DDR5 is designed to have more banks and bank groups, longer data burst lengths and 2 independent 32 bit addressable sub channels. A lower aida64 latency number doesn't always mean better latency when RAM has many different timing commands.
I'd love to see the comparison with AMD and 11th gen. I believe people noticed Ryzen had better mouse latency when it came out because USB was built into the CPU die.
@@alikiyani420 as it seems, the 11th Gen is still Monolithic microarchitecture, so it should NOT be affected by this issue. Even more interestingly: according to the Bard AI ... the last monolithic consumer CPU from Intel is: i7-12700KF !!! .......imagine pairing this one with DDR5-6000 CL30 RAM! :O :O :O ...... I guess You would get *nearly* the same responsiveness as Ryzen 7800X3D !
I don't know what problems or weird bugs you have for your systems, but i have a 12700K and DPC latency is lower than your 10th gen, 12th gen and 13th gen systems including the i9s. I use DAWs, lightyears more CPU intensive than Premiere Pro.
Here is what I see. As an owner of 13900k, I tune my CPU, but don't overclock it. It has a random frame time spike and it is consistent, but yet random to what is going on in games. It actually shows in other people's benchmarks with 4090 and 4080, I see it too. LatencyMon completely hard locks my system. I tried different ram 2 motherboards, then my previous one would just say that my PC is not meant for real time recording in LatencyMon. I have drop outs in my professional audio equipment with 13900k, but amazing fps in games, with an occasional single stutter. Latency in LatencyMon is high, until i turn off High Precision Event Timer in Device Manager, but the stutter does not go away. In fact I saw frame to frame latency get worse with HPET disabled. LatencyMon still breaks my system on both of my Apex Encore Z790 motherboards. 9900k has none of those issues. Something in the E-Core/P-Core/Internal Memory Controller of 13900k and the way Windows 11 scheduler talks to it. I saw none of this random stutter with 7950x Ryzen 9. I speculate difference in core frequencies does make a difference, even though I have SpeedStep and SpeedShift turned off. I have 7950x coming to test it, so we shall see if it exhibits same issue. By the way, I tried two 13900k CPUs, same issue. Head to my channel to find a way how to remove 95% of stutters on 13th, 14th, and 12th gen Intel cpus. I found it this week. Note that new Arrowhead CPU by intel will not have hyperthreading, and I understand why now. I stumbled on partial solution to my stutters. Stutter happened in Ready or Not as you first spawn in, I assume it loads everything to L2/L3 cache, but it's very visible.
Perhaps you shoud have tested the 13th gen with DDR4 RAM like the 10th gen. Not only would that even the playing field a bit, but also potentially remove another unintented difference in results? Because what if the difference or the latency issue is caused by (or at least it plays a part in it) the DDR5 RAM? I mean it shouldn't.. but at this point I have no idea whats possible and what could cause an issue and what couldn't, so why not just even that playing field by having both platforms be on DDR4?
I have a i7-12700 with DDR4 and have never faced those issues ... He should have also tested with a completely different hardware setup for the 13th ... Because probably that is the root cause of his problems ...
I was wondering the same thing I went with DDR4 3600 CL14 (bdie) dual rank dual channel 4 sticks with my 13700k just because it was an interesting low latency combo
@@Horendus123 are you sure it was low latency? Have you run the same set of tests he did in the video? I'm asking this because I'll have to make a purchase decision soon.
@@Dhruv-qw7jf am I sure the RAM I have is low latency? Yes it is objectively considered low latency and best DDR4 has to offer. Have I run a battery of tests like in the video. No.
This is an EXTREMELY interesting topic, not that niche actually, especially these days where people are BORED with new generations not giving the expected performance uplift. I truely think you could have gone a bit further because your findings deserve a closer look to not let things unexplained. Let's not forget that Intel's 12th and 13th gens are heavily tuned for idle efficiency, running on the e-cores, these CPU can draw as little as 6W (as for my 12700k) in idle, so it is not that surprising to observe a somewhat higher latency on these chips in some specific situations. Some ideas for a part 3 : * Disable e-cores and run the benchmarks again * Compare and show the operating frequencies of the CPUs * Disable any power saving features and C-states in the bios to see if it has an impact on latency Your content is catching my eye more and more these days, and I think you are making the right choice not trying to show how many FPS a 4090 can pull out of a 13900k in 1080p. People don't really need performance databases, they need to be shown if they are gonna SEE a performance uplift from a hardware upgrade. This information is actually hard to find. Have a great day. From France :)
Max DPC latency on my 12700K (5 watts iddle, not 6) is 50 microseconds, average is 20 microseconds. It's an MSI Z690 Pro A DDR5 motherboard, original 2021 BIOS. What higher latency? I do stuff with digital audio workstations and emulators, the 10900K is slow AF, 2015 Skylake cores. Literally is not that different from laptop CPUs at the time for singlethread speeds. Adobe software has stability problems. My 12700K is an AVX512 batch from 2021, i disabled the e-cores, enabled AVX512, the OS felt slower, and i lost fps on Switch emulation that supposedly supports those instructions. P-core frequency is exactly 4.7GHz all core turbo and E-core is 3.6GHz, max turbo in one thread is 5GHz. I use Super Smash Bros Ultimate as a CPU latency/IPC stress test with 8 ice climbers lv9 CPUs, one P-core with 100% use, frametime is perfectly flat and exactly 16.67ms, not a single stutter (with all shaders compiled).
I'm glad that i have an MSI motherboard because it simply works. I heared bad stuff about ASUS and Gigabyte, like bad quality VRMs or other problems on Intel. The i7-12700K is a beast of a chip, meanwhile i heared that the Ryzen 7 5800X3D has problems because of clockspeed jitter (maybe a thermal issue, power delivery or sillicon problems). If i went for a cheaper motherboard, maybe i would get jitter. I'm 100% that the motherboards used on the video for 12th gen and 13th gen systems have worse quality than 10th gen, it's a downwards trend that i see since Rocket Lake.
I don't know about Zen 3 or 4. I run a Ryzen 5 1600 on my win 10 rig for games. When I browse or manipulate windows around, I don't seem to see the latency, or when selecting multiple files in file manager.
7000 series have a different problem, the off-center placement of the dies under heat spreader, which causes typical cooler installation not all that efficient (and there's now offset brackets to remedy this, which is kind of ridiculous).
Quote from Calypto's Latency Guide:"Earlier Zen CPUs consisted of groups of cores called a Core Complex (CCX). Each CCX has four or less cores, and there are two or more CCXs which are connected together via the Infinity Fabric. The Infinity Fabric is fast, but not fast enough to not have noticeable performance loss in games as well as reduced desktop responsiveness due to inter-CCX communication. On top of this, Ryzen CPUs also have higher RAM latency than desktop Intel CPUs. Starting with Zen 3 (Ryzen 5XXX), each CCD (core complex die) has an eight core CCX which greatly reduces intercore latency, and unifies the split L3 cache previous generations had. This brings massive performance improvements across the board, but unfortunately the memory latency still suffers due to the memory controller being located on the I/O die. For most gamers, a 12th gen. Intel CPU with efficiency cores disabled will perform better than a Zen 3 CPU."
I didn't read through all the replies so I don't know if anybody raised these already. The only other two variables that come to my mind would be to try a DDR4 motherboard so that you would end up using the exact same memory sticks and also to try disabling all the e-course on the 13900K and do a latency comparison again.
I raised that same issue. And I commented that at around 40-50mins since the video was uploaded, there were only about 25 comments back then. It's a bit sus that since then he's "heart-ed" a lot of comments saying the same things he did in the video but not the one comment that was actually suggesting something to update his testing methodology on. And you'd think it'd be obvious to compare a DDR4 system with another DDR4 system but he went with DDR5. That's a bit sus as well.
You can test all you want. If you don`t go the scientific way (real world benchmarks, memory speed tests and testing suits) then you`re just staying in the dark with what you want to believe. It`s self deception and the further you go the crazier it gets. Just read all the insanity here. Then you know what i mean.
It seems to me like it's a windows related issue but only for Intel processors. I am using 1260P laptop with an eGPU to play games and while it plays games really well, it's stuttering when opening like 5 UA-cam tabs at the same time. This never happened with my 2700x, and I am only swapping from it because it had some other issue (it could have been either RAM, CPU or the motherboard but investigating the root cause and trying to solve it is a waste of time, so I am upgrading the whole thing). I am now going to assemble a new PC tomorrow with 7900X, 64GB DDR5 RAM, RTX 3070 and I am going to see if I see similar issues but based on my previous experience with 2700x I am sure I am going to see extremely good performance, and the cost is going to be worth it.
man idk my workflow is super taxing on the cpu and my first try on the new intels was the 12700k, this thing was faster than anything I've used (11900, 10700, etc), now with the 13900k it feels even more responsive and fast, you can have a big 3ds Max scene opened and After Effects for testing different compositions with over 100 gb of ram used and still it feels just fine, other than the noise from the fans you couldn't tell if it's under load or not, it is that fast, my only concern is cpu is it wants to run at 300w at the first glance of heavy work, my 360 aio is almost not able to keep up with it, there's no 1 single application or procedure that I feel slower or less apt than before, even considering I had to move to win 11 🤦♂
You may want to test the 12600 vs the 12600k. One is the fastest 12 Gen no e-cores CPU, the other has 4 e-cores. Comparing these two could be very interesting in light of your latency findings. Thanks for a great video.
Also because the 12600 is derived from a native 6 P-core die (not a full die with disabled cores), which may help given that the bus should be shorter and less complex then too.
Got to love the fact that you are quite possibly one of the only reviewers that has the ability to multitask on one pc enough to overload it and feel the difference. Keep going with your gut feelings on products, it is what sets you apart. !Cheers!
Seriously you are the best, do not get a big head, but yes! If you tube as a whole does not see it so, they need to check what the actual user is into.
I was thinking the same. I also wonder if turning off the E cores would help. I’m not even doing this workflow but I’m just curious because I like nerdy stats 😊 I’d also love to hear about this from an actual engineer.
Same. While I9 from 11 gen was generally considered a trash product, the I5-11400 and i5-11600 were widely considered good products. As an owner of one of these I am curious how much this topic is accurate for 11 vs 12 gen.
Naaaaaa this isn't niche scenarios.. Everyone clicks and search files on their computer. Not everyone just stares at the loading screen while playing games lol This is be part of the standard for test benching new cpus. Latency is important when everyone is doing thousands of clicks a day. If each click takes an extra second to to perform the function, that's 15 mins a day.. That's 91.25 hours a year lost lol.. fffff that
Can't wait till Bryan is back in Australia and do content like Tech Yes Lovin' of filthy parts with his cleaning methods. I love those videos. He is too limited in what he can do for content in Japan. It gets monotonous fast.
Darn.. so now this, combined with 14th gen Intel being just a "refresh" means that I have to cross my fingers and wait until 15th gen to replace my 6800k/GTX1080 desktop. 5090 here I come. lol
@@HeartOfAdel now i have gskill trident neo 64GB 16x4 CL16-19-19-39 3.6GHZ XMP but this ram wont overclock at all tried for days but happy with stock 3.6ghz
SpeedStep ramp. You cannot go from one P-state to the next instantly. There are limits in place which P-states are reachable from which ones. During transition, the CPU is disabled, then clock and voltage adjustment takes place, and the CPU is enabled again. This takes a lot of time. You don't take one ramp step right after the other, you let the voltage regulators settle under load first and also if you did that you'd have a performance dip right up front, because you'd be spending a little too much time with CPU halted. Fine grained P-states and slower more conservative P-state transitions will make for a system which is much more power optimised but less responsive to burst workload. I'm not saying that this is necessarily the cause, but definitely something you want to investigate given the symptoms.
Are you sure it isn’t just ddr4 lower latency feeling snappier ? Would have been more interesting to see if it’s the ddr5 or the board itself. I really enjoy your findings for sure. I have noticed ddr5 6000 in general has higher latency than the best ddr4 has the offer
The DDR4 used in this test was very slow (3000CL16). I guess he did not want to push the DDR4 fast low latency OCs here to not introduce even more variables.
@@innocentiuslacrim2290They should be about the same latency-wise, iirc 3200CL16 and 6000CL30 both work out to 10 nanoseconds, since CL is measured in cycles rather than an absolute time unit?
I upgraded to a 13600KF + Z690 (DDR4) two months ago and the system runs like a dream (Windows 10, E-cores disabled, C-states disabled, core parking disabled). I haven't experienced any of the examples you mentioned. Most of the time I'm running the Balanced power plan with the CPU set to 99%, which locks it to 3.3 GHz (same for the Uncore). I use Bitsum Highest Performance (locked 5.1 GHz with 4.0 GHz Uncore) when I want the full performance for gaming, recording or encoding video. I can fully load the CPU and the system is still completely responsive. I even use a 5940 RPM HDD for most of my data - searching, selecting, moving files is super responsive. Obviously even faster with both of my SSDs - one connected to the CPU, other one to the chipset. It has to be a configuration issue. Are your power saving options disabled? Is Windows running in UEFI or Legacy (CSM) mode? Are you sure you don't have High Precision Event Timer enabled and forced, which is a disaster on modern systems? Or some other internet tweaks (like useplatformtick) that usually have no positive effect, but often they break something. None of it makes any sense. Please investigate this issue further before making these kind of allegations. This is pure speculation at the moment.
The only reliable way to test for "latency" would be to use tools such as xperf and windows performance analyzer and find out which drivers are causing input lag. LatencyMon should only be used to have a quick look at haywire drivers causing stutters, as the program itself causes stutters. Btw, the results are in microseconds, not nanoseconds. Your ISR seems to be spiking pretty high due to presummably the NVidia drivers (cannot possibly tell as you haven't mentioned what you were doing in the background while recording this), the first tab doesn't tell us anything. Use the 'Drivers' tab and sort them by highest execution, depending on what you're doing in the background you can trigger different results. For example: Bad USB drivers (u can easily trigger and see it in real time, especially on a 8KHz mouse) will use the wdf0100.sys driver GPU will trigger interrupts when watching a video and using LatencyMon to record. Bad ethernet drivers (Intel NICs seem to be pretty reliable) Bad storage driver / port (changing port can alleviate stutters, M.2 SSD causes the least amount of stutter cause you're avoiding the SATA controller) I'd recommend you to read through guides such as these ones: github.com/amitxv/PC-Tuning github.com/djdallmann/GamingPCSetup Your 13900k could also not be completely stable, thus leading to these irregularities during file transfer (or perhaps windows not being setup right / storage driver or hardware playing games) Try testing it in bootable porteus linux linpack (updated with latest 2022 libs from SlovenianSlobodan) or this version (testing through windows doens't reveal instabilities easily): github.com/BoringBoredom/Linpack-Extended/
This is ringing a bell. I feel like I recall Hardware Unboxed mentioning differences in percieved desktop "snappiness" between new CPUs in one of their CPU reviews. I'm pretty sure it was AMD vs Intel CPUs rather than intel generations though, and it was either 12th gen vs 5xxx or 13th gen vs 7xxx at the launch of one of those things. I very vaguely recall the observation being that AMD had caught up to intel in terms of the desktop experience feeling snappy. Now I'm thinking it wasn't AMD catching up, so much as Intel losing a step. There is a possibility that it was jayztwocents and not HUB. Or maybe it was both. Either way, there was no discussion beyond the observation.
@techyescity could you please test 12/13 gen with ddr4 in part3 ? I have 12900k and a 13700k (2systems) and I opted for ddr4 on both and I see no issues, tested with ddr5 in the past and I was hiting some issues in windows (slowness etc), maybe this was the issue? The ram controller on these new cpu? Ddr5 latency?
We need to see how AMD does in your tests! The R9 5900HX has a DPC latency of 116 so even lower than 10th gen. Also my AM4 desktop is super responsive in Windows 10 with Atlas.
You want to know what's crazy I just pulled out some of my old boards gigabyte aorus z270x gaming k7 i7 7700k only 16gb ram and m.2 3x4 for storage. And I must say it is seems to be way snappier than my newer PC smh
I have an 11700F system and producing a video on that one, compared to my 18 Core Haswell X99 HP Workstation, I was underwhelmed. I mean in Cinebench and other tests the IPC is much faster but editing videos wasn't multiple times better, I expected a lot more. It's nothing like going from a Pentium to Pentium II THAT was an upgrade!
I have an Erying MATX motherboard with a soldered on Core i9 12900H laptop CPU(14 cores/20 threads w/6P and 8E cores) , and with 32GB of DDR4 3200Mhz RAM it seems to do much better on Manjaro GNOME Linux in traditional layout with way less lag than the brief time I spent with Win 10, although I do have a couple of the P cores that like to do random brief spikes in temps up into the 80's, and 90's but go right back down with a down firing CPU cooler to help cool the VRM, and RAM. So yeah I'm overall happy with it having come from a an AMD A10 5800K with 16GB DDR3 RAM, and an AMD RX560 4GB to a AMD RX 6650XT 8GB GPU.
@@toseltreps1101 Well, from a Commodore 64 to an A10 5800K was a huge jump! But C64s were nevertheless great fun back in the days, no need for CPU and GPU upgrades then... and still many generations of games
This is just the video that I've been looking for. I've been so tired lately of seeing "gaming benchmarks" and "synthetic benchmarks" It's so unrealistic to someone who relies on their computer for work and multitasking day in and day out. thank you tech yes city
Maybe it is just a Windows problem, as on my Linux box it is not like that? Also, I have DDR4, not DDR5 with my 13600KF. EDIT: You should really test on different operating system, not just Windows, as that really might not be a hardware problem. EDIT 2: Jay mentioned a long time ago that AMD feels snappier than Intel with 12th gen.
"Moore's Law Is Dead" recently had an audio-focused custom integrator as a guest, and he seemed to be quite unhappy with recent Intel releases. I wonder if this is related.
In broken silicon's recent video with a DAW enginee/integrator, in the discussion about sapphire rapids he explains he's stuck with providing 10th gen audio workstations specifically because of latency introduced after that gen. He also mentioned turning off core parking had an effect on latency, don't know if you've already looked into that or not.
I just listened to this episode a few days ago and was wondering the same. Maybe Brian can revisit after tweaking core parking. Hopefully Intel is not regressing on responsiveness in pursuit of more power efficiency.
Weird that the latency is higher even on 11th gen, I would've thought the difference was due to the hybrid core architecture introduced in the 12th gen.
@@utubekullanicisi the geared/hybrid memory controller on 11th gen is at least 1 reason why latency is higher on it and why even though it has higher theoretical IPC it basically matched performance in gaming as an example with 10th gen.
13:15 Last Tuesday Windows 11 update caused again a slowing down SSD and BSOD's for some people. The slow down SSD is a months old problem for W11, they are struggling to solve it.
LatencyMon measures MICROseconds, not nanoseconds. The "μ" in μs means micro. Nanoseconds is written as "ns". Also, there are several problems with your use of this tool. Firstly, 1 minute is not enough to compare 2 cpu's because drivers do not get loaded multiple times a minute. For that same reason, looking at the maximum latency doesn't make sense. There might be some dll loaded at the same time as another driver on the one system but not the other. You can even see in your screenshot that dxgkernel was loaded on the 10850K but not on the 13900K. If dxgkrnl was loaded on the 13900K it would have a higher ISR time as well. In short, the results you are looking at are worthless. What you should be looking at is the "current interrupt latency" to see if the two differ when at idle or when performing the same tasks, but even then you'd need multiple runs to decrease the chance of recording background dll calls which happen randomly.
I would look at the DDR5. From memory people complained that the larger timings can slow things down, ie, it needs to be optimised, or even switched for better timing DDR5. Plus you have to look at the numbers of channels, and the number of ranks. Basically the DDR4 setup might have more than the DDR5 setup, ie, it's not apples to apples.
I have been utterly disappointed with the i9 1300k so far. I downgraded to this flagship from an i7 8700k which was blazing fast with display operations and program launches, menu navigation etc. My wonderful top end system with 128GB of ddr5 and the 13900k sucks
Great video Bryan! I had huge issues with DPC latency back a decade or so - and no one on any of the PC tech support forums knew what on Earth I was talking about. Programs like LatencyMon really helped me pinpoint certain driver and hardware issues that was causing massive latency spikes that was creating very much audible "glitching", especially in audio work.
It's interesting, I just did a fresh install of W11 in my 10900k rig and it's significantly more snappy than it was on W10. But I am willing to chalk that up to 3 years of wear and tear in the OS. Overall though, I think W11 is significantly easier to debloat. The whole OS is just more cohesive which I feel contributes to the ease of tuning. Boot times with all programs launched is lower, and initial CPU and RAM usage on idle is significantly better. Subjectively, I also prefer the more cohesive UI of W11. I also have my 10900k highly tuned with a custom overclock, which is actually undervolted and runs at 5ghz all core turbo, slightly higher than stock 4.9 while also retaining the 5.1ghz 4-core boost behavior. Full load TDP is significantly lower than stock, tops out at just under 220W in Cinebench. I was never able to observe the 5.2/5.3 super marketing turbo modes in any real world scenario, so I just ignored those. I could enable them in BIOS with this OC still, but like I said I was never able to observe them even with the stock configuration so I just decided to leave those out of my OC. A 13900k would definitely be better for gaming, but I am yet to run into a situation where I am CPU bottlenecked with my 3080Ti. Will likely upgrade the CPU again before I replace the CPU
@@htoomyatlin123thought this myself but 13 gen performs the same on ddr4 & ddr5 in terms of latency (I have both boards). Its something to do with architecture of the cpu.
From techyes, to techno. You are once again claiming things without proper testing on different hadware setups, and without any hardware, software, OS engeenering knowledge. You are once again fooling your viewers instead on approaching the question, with proper research and testing. You are not able anymore to provide contents, outside the review materials and products companies send you. I have still not forgotten how you blatantly followed the AMD marketing materials sent to mainstream reviewers, when AMD asked to create content, on how good the 7900 was at playing COD.
I'm having this latency desktop issue from Core 2 Duo era. All my builds had in common one thing: NVIDIA GPU's. And it seems, it is true. Some hardware combinations with NVIDIA GPU's has terrible a terrible desktop latency. Recently I bought some Ivy Bridges, and with Intel GPU, they are amazing and responsive. Tested this with a GT 730 and the latency is back. I think in the case of Intel iGPU it makes sense as long they do use a internal bridge different than PCIe.
It’s a needed test so well done for doing it so thoroughly. Two things I’d add as comments. Yes we expect the next technology to be better than before and use performance as that metric. But that isn’t always the case. In recent memory there has been cases of a failed architecture (AMD FX) and outliars from Intel (Any of the CPUs Crystal Well) which showed performance differences in different case. The second point is about test bench work and here I would say that real world testing, use cases and ‘feel’ is actually more important than simple average FPS numbers or synthetic benchmarks. It’s much harder to capture but it is far closer to the real world use a PC will go through for most people. That makes investigations like yours absolutely priceless so don’t do yourself down for being the only one doing it. It’s important work and thank you for doing it.
I havent slept since part 1, I really hope it was worth it.
Off course not there are no problems. It`s just a faulty windows install. Or some incompatibility issues.
Don`t get fooled. The problems mentioned can never cause what he is showing, they are way to small.
@@brugj03 bro who are you?
@@jblpsintel marketing manager probably 😂
@@brugj03 Lol a real intel shill in the flesh!
@@nenadcaric5195 I doubt it. A marketing manager would do a better job of writing his response.
I love these investigation videos! Looking forward to seeing Ryzen 7950x put through the same tests.
And 5000 please ❤
Add 7950x3d to the test list
Ryzen is slower. Ryzen's reaction speed is slower than Intel's 12th generation.
@@KingJeffKiller *5800X3D*, please! :o)
@@overtonecz Well on Latencymon I got 95 nanoseconds on the 5800X3D... for what its worth.
Brian just entered the deepest and darkest path of being a PC enthusiast 😂
😂 the path to insanity
@@ilovehotdogs125790 Hear.. hear..or fantasy island.
@brugj03 Hersey! Hearsay! Hang that WIntchel amd… ok I’m done.
@brugj03 Hersey! Hearsay! Hang that WIntchel amd… ok I’m done.
@@ilovehotdogs125790 You`re quoting me.
I feel honored.
For AMD Users (AM4/AM5): Disabling USB selective suspend setting in the power plan options really made an immediate, noticeable difference in both LatencyMon and real-world performance. Try it out.
True, odd how that works out.
Does that help in intel 11th gen cpus? laptop CPU's ?
@@vmafarah9473 I only have a 9900k/Z390 as my newest Intel platform but it probably does help. Unfortunately the ultimate limiting factor are Nvidia's drivers. There's just no way to control the problem that they have.
This helped me with weird USB issues. (on an Intel system)
This is something i tell everyone to disable, especially if you work with audio. It's extremely helpful
Why not just use an NVIDIA LDAT to measure latency? Also would be really interested to see how the 5800x3D and 7800x3D as well as 11th gen intel compare in these types of testing.
Does 11th gen also have this issue or is it based on the same architecture as the 10th gen?
11th Gen was so similar to 10th Gen people were almost upset by it. There’s fewer cores and a slight ipc and single core uplift. This does improve performance in a few games but has it lose in some games. I would assume the latency difference would be negligible between them on Windows 10.
@@quintrapnell3605I think the upset was more that it was on the same process node and that the 11900k had no reason to exist with the 8 core limit. In case u weren't aware the 11900k had about double the single threaded uplift the 10900k had over the 9900k and that was with the 10900k getting an extra 300mhz while the 11900k got none. 11th gen was much higher ipc than 10th gen. That's also why it only had 8 cores, the cores had much more transistors than comet lake cores bcuz they actually are signicantly different, so when they were forced to backport the cores to 14nm, they could only fit 8 of them on the die.
would love to see the further testing with more architectures :) I'm especially curious how AMD compares (Zen 2/3 mainly, though I don't know how different the various AMD gens are)
this
I'm really curious if Dr. Ian Cutress would chime in, as he is a brilliant person, and he might have some contacts within Intel that could shed a bit of light without breaking any NDAs or such.
You nailed my question ahead of time. Does this happen with AMD CPUs as well? :)
It could be the e-cores. They are all clumped together on one side of the die, so even if the CPU wanted to start using a P-core it could have latency going from an e-core over to a p-core. Maybe see if disabling the e-cores makes a difference
This was addressed in part 1: disabling the e-cores made no difference.
@@lophilip ah, I see. I must've missed that. Next I need to look at the size of the big cores. The 13900k simply has a physically larger die than the 10900k. So maybe extra latency is added there. But maybe it's just from the e-cores. Which is why we need to look at the p-cores specifically. If not that, I'd say it's probably just some architectural issue that we can't really solve
@@jamescampbell6728 Bro you're going in circles. He literally states the technical changes that cause the small latency difference IN THE VIDEO.
@@CyberneticArgumentCreatorI skimmed the video and watched the conclusion segment where he said jack shit about why.
Adding more cache to the frontside bus would help future CPUs with the little core big core model. We’re talking like a 100MB+ of L3 cache and 300MB+ of L4 cache.
As an owner of a 11900K, I'd very much like to see how it performs in these tests.
I guess it would be similar to the 10900k I think those latency problems are somewhat related to the e core architecture.
Should be the same as 10th gen.
11900k is worse than 10900k.
11900k is worse than a 10900k, that’s all you need to know really
My motherboard will take an 11th Gen CPU, but I didn't get one after comparing to my i9-10850k.
Insightful content, well done! Thanks. Loving more my 10900K
In its essence, it's probably a (Microsoft Windows) kernel or some software issue. To be a real hardware issue, it would either break things or it just doesn't work or support something. Code optimization isn't an easy thing at these low levels without knowing and fully understanding the bigger picture.
Increased latency can definitely be explained by increased distance between different parts of a system, i.e. in this case presumably the CPU and the I/O controller hub. Differences in hardware design can reduce performance in specific situations without breaking anything.
@@VVayVVard Yes, true if you look at it purely electrical and from a signal level. It's like considering an expensive or smaller form factor motherboard board. Mostly these issues are "noticeable/measurable" on a system level between chipset and CPU and various IO. Not on the CPU itself, they are either somewhere between specification or not (true you can have little benefits in better silicon quality) but, to say the translation and interpretation of these lightning fast and tinny differences (noticeable in the OS) in a pure hardware issue ... you put a lot of faith in multiple perfect software overhead layers to perform flawless. But hypothetical from a theory perspective, you are right.
Now I don't remember, but I thought the hypothesis here was a new form of branch prediction between gens, no? So that dives into low-level x86 instructions and code, in my opinion, not latency cost by physics.
Sounds like you may be interested in getting an Optane drive. They are on a clearance since earlier this year, due to end of production. I used Optane 905p for photo editing, and Lightroom and Photoshop became a lot snappier. Access latency is 10 us vs 100 us of NAND.
finally part 2 been wating, can't wait to see results!
I was just talking about this "feeling" with a friend. Between an i9-9900KF and r9-5900. The Intel chip just feels snappy when multitasking and doing a bunch of small ops rapid fire.
Kind of like doing a drag race with a diesel versus gas. Intel would be the gas and AMD the diesel. (Just for this comparison on these two chips) doing 2 runs, one with no weight, One with a 5,000 lb trailer behind it.
The 9900kf feels instant right off the line but then bogs down on larger things.
The 5900 feels slower off the line but pulls ahead quickly with the larger ops.
This is interesting, not the first time I hear that. Obviously 9900K is snappier cause it's monolithic with lower latencies, and slower in big apps because it has less cores. (I guess 10900k would be a better comparison here). But microtasks seem more important for a good pc feeling.
Have a look at my channel with tuned i9-9900K, there'll soon be a video comparing 4.7ghz 3200cl16 49ns vs 5ghz tuned 4266cl16 33ns. The difference is Mind-Blowing. It is also as good as tuned 5900X in games.
This kinda reminded me a bit of when I used to have a 3300X and later changed to the 5900X, for the good ol' drag n drop files the 3300X was superbly snappy compared to my 5900X, all due of having the single CCD. Still it's not something as notorious as you have showed here with the 10th vs 13th gen Intel, but I do agree on the moment you changed hardware, there's things that you used to do that are pretty basic, that feel "off"
exactly what i need to know, thank you!
Would be even more interesting to go further back down the rabbit hole with X58 and Win 7. I expect you'll find similar results. For years I thought I was tripping with my intel systems, the best of which was my old X58 920 D0 banger at 4.4, closely followed by the Sandy Bridge 5ghz 2500k both lightning fast in day to day snappiness.
Don't seem to find the same issues on Ryzen 5 series honestly.
You forgot to mention the 486.
I assume you are just being funny.
Nvidia Drivers seem to have big issues with latency on some systems, my 4090+5800X3D shows more latency in Latency Mon if I use both monitors (LG C1 OLED and Asus PG279QZ) than when I use just one of them (doesn't matter which one).
I also tested a 5900X with both a 3080 and a 6800XT, more latency with the 3080.
Maybe something to look into?
Please add i9 9900KS and compare it to 13900K too, but with oced ram and tight timings, interesting how much more responsive it's going to get
9900k is the lowest latency chip I know of. Maybe a 7700k with 4 core ringbus is even more snappy?
@@ilovehotdogs125790 I can't tell really, never had it. But it probably won't take dual rank 4000+ b-die and it makes a big difference in snappiness. So the answer is no I suppose, maybe 8700K :)
@@HeartOfAdel yeah maybe the 8700k, no idea how good the imc is on that chip. The 10900k can do 4400mhz dual rank 32gb. So if the ram speed matters more than the ringbus size then maybe the 10900k is still lower latency.
@ilovehotdogs125790 8700k certainly has a hard time going above dual rank 4000-4200, but 9900ks and 10900k can, not worth it though for the voltage requirement usually. I think all of them will feel about the same.
Lag isn't a problem, it's a feature. Appreciate that you're not using the lower-binned SKUs that have other, uh, features, too.
What if lower binned chips are actually faster? They should have less IO (due to less cores), so less latency too.
what “features” do lower binned skus have.
@@ilovehotdogs125790 Less cores, thus less need for IO transfers.
@@MJ-uk6lu yeah but is it measurably better? Because even on chips like the 10700k, it having 2 less cores didn't help the ringbus since it was just 8 cores with 2 dead cores. So if the lower core chips just have the same die with some disabled cores instead of having a smaller ringbus then will that even help?
@@ilovehotdogs125790 Maybe, low tier chips don't have E cores, which are quite useless, so perhaps.
I wish I had time to try these tests on various Linux distros. Latency has always been a software variable on those OSes. Just last night, a system update took my AMD 5950X from a sluggish UI experience to smooth. The only other variable, was that I disabled the TPM in BIOS but I doubt this had anything to do with it.
what version (old and new) specifically? what distro? did kernel update too? this might be useful for future reference... I'm also an AMD user but I ditched Linux last year...
@@FeelingShred Pop_OS...It wasn't any "version" - just a routine system update which could have included any number of changes that solved the issue. Using Linux is a mater of principle to me. There are no issues that will ever make me switch back to an evil OS like windows or mac. My motto is, if Linux can't do it, it can be done in a VM. If that doesn't work either, then it doesn't need to be done.
thanks for the info it is well deserved to see it is an architect issue then an entire system issue. And the reason why other enthusiasts have not talk about it because they do not know the information that you already have been told by CPU specialist. Always stay busy and keep giving us good hardware content.
Please test at least one 5000 series AMD CPU on part 3 (5600 and 5800X3D are the most popular).
Personally I want to see these CPUs 5600G, 5600, 5800X3D, 7600, 7600X3D.
Monolithic (5600G) vs chiplet (5600) design latency probably not too ineresting based on gaming 0,1% lows but who knows maybe there is an advantage for the monolithic chips.
4:30 - How exactly are you measuring input lag to the milliscond with 120fps? The camera would only have ~10ms resolution. You'd need to go to 1000fps for 1ms resolution..
great content, can't imagine how much work you put in this one. cheers
Always appreciate your hard work Bryan 👍👍👍
My 10900k @5.1 all Cores 1.3 volts still screams and I'm happy AF
THANK YOU I though i was going NUTS i upgrade from a 10850k to a 12900k about a year ago and when i was moving multiple files in davinci resolve it felt like the 12th gen was slower. Thank you Brian for this Video.
I have a 10th gen laptop for work given to me by work. Although it's only an i3 it's the snappiest PC I've ever used. I'm well impressed and now know my gaming PC is going to be a 10th gen if I go Intel.
Glad these are not the types of problems I have to deal with, given my cheap general purpose build...
Huanazhi X99TF ($50-discounted)
Xeon E5-2696v3 (TU and UV: -50/-50mV)-used for $46
48GB DDR4 ($80-used)
RTX2080Ti ($230-used)
PSU 850W Gold (used, $50)
1TB NVME m.2 ($120)
I have my i7-12700K PC that destroys the PCs used on the video for system latency, i built it for music production and emulation.
The i7 cost me 330 dollars, 32GB of DDR5 4800 for 250, MSI Z690 Pro A DDR5 for 150, two 1 TB M2 SSDs 120$, i'm looking for a 8GB RX580 for 100 bucks right now (crazy people still selling garbage GT1030s around the same price and they are slower than the Intel UHD 770 i have right now running Crysis).
Thanks for the video. I have an i9-10850k, this confirms that I have no need to upgrade.
exactly. I have 10700K, same, no need to any upgrade, if only, I would go to 10900K (same MB and RAM).
11700K undervolted @4Ghz all-core on an ASRock Extreme Z590 is my choice for general use.
(While having a virtual LAN party running in the background.)
What are the settings that are using and for what do you use this sistem?
@@puiuus core and ring both x40, 1.15V fixed core, 0.95 fixed sa, 1.10 vio, but 1.40 dram (126gb)
Learning infrastructure as code and throw myself several virtual LAN parties. Above 4Ghz isn't worth the extra heat for me. Attempting everything on just 1 computer though will use remote desktop on cheap notebooks/laptops like an open-box ASUS E410 from Best Buy.
Did you change the power plan in the legacy power panel? Leave it on balanced.
The Northbridge / SA is still located directly on the CPU silicon. I'm not saying your impression of additional latency is incorrect but your technical explanation of why it's happening is dead wrong. It is all still a monolithic die. If you don't believe me, look at an open source die shot of a 13900K.
You're comparing 2 different chipsets as well as 2 different generations of memory (DDR4 & DDR5) between DDR4 and DDR5 and how the memory controller handles data. It has absolutely nothing to do with Raptor Lake moving the northbridge (which it did not).
Whomever you spoke to at Computex either misunderstood you or something got lost in translation.
Isn’t he also comparing across w10 and w11? Like that’s also obviously going to contribute something.
This is fascinating! I went from a 10900K -13900k and although benchmarks prove the new system is much faster the very day experience is different. For music production and photoshop the 10900k was snappier and a better experience!
C states in your BIOS allow your CPU to shut off which takes quite a while to come back on (anti-snappiness). Simple fix, just turn off the C3 C6 C7 wait states so the CPU stays in C1 all the time and you will likely regain the responsiveness you paid for. Hyperthreading (HT) and Efficiency cores are not your friend if you're in a hurry. Buy a "Kill A Watt Electricity Usage Monitor" to measure the wattage used by your computer with different BIOS settings.
@@blormpf1740 Would this solve the problems and differences encountered in this video? He spoke of an IO thing being moved out of the CPU.
@@H53. The effects of the C states are workload-dependent, i.e., how the application runs, how it is used, how it is compiled, etc. That's a lot of syllables to say turning off C states will always help improve "snappiness" and never hurt performance. This holds true for all CPUs with power management features. A $20 kill-a-watt meter will track electricity usage and provide empirical data to show how much (or not) electricity a computer uses, and the effects of changing the C state settings can easily be seen. I turned off the C states and bought a new display and my net electricity use is lower and my computer is always snappy. Life's no bargain buying a fast CPU only to force it to be janky all the time.
Thread Director adds latency and occasionally directs latency sensitive threads on the e-cores instead of the p-cores. It is also super aggressive on moving threads to e-cores from p-cores in the name of “efficiency” so it can park the p-cores faster. I had a 12600 non k. It was quick and handled Cyberpunk 2077 very well. Upgraded to the 13700k and the only thing I noticed is alt tabbing out of games is insanely fast and acts as if the game isn’t running.
@@H53. some IO is off the cpu - depends on what your using and where its connected.
Amazing work, I am always amazed by how much you think and work out of the box to proof that our "gutfeeling" about latency was right
Part 3 should be comparing with Ryzen I guess. Ryzen used to be unresponsive due to slow CCD/CCX and IF but improved over the years with cache, especially with 3D cache.
I upgraded from 4770K to 7800X3D and to Win11, I feel like I'm hitting this same issue.
@@makere Most likely your imagination bro.
@@makere try windows 10 and use it until support ends in 2025. Then rather try windows 12
@@vincentvega3093 I need some win11 features.
@@christopherjames9843 Definetly got some slowdowns I didn't have before, but not sure if it's some hardware/software issue I have, or something related to the latency stuff.
As someone very sensitive to latency I found the most common cause is power management and downclocking, disabling low power states will definitely improve this. Same goes for GPUs, I remember my RX550 used to cause mouse movement stutters on idle desktop when the card was repeatedly downclocking and upclocking. However I do appreciate my 13600k idling at
Yeah, wanted to test out latencymon on my 5800X3D and i got like 700nano seconds instantly, until I randomly had task manager open, got low scores, and it clicked.
Turned on "high perf." power profile and got sub 100 nanosecond scores.
I genuinely miss my i7 7700k vs my 7950x3D. No parking, no asymmetrical cores, no annoying fluctuating clock frequencies, just a pure and simple processor that WORKED. I hate where things are going in the desktop space. We are regressing because these companies can't keep putting out faster hardware due to Moore's law dying. It's bad times ahead, friends.
Thats why i keep my 10900k with ddr4 4600 c16 as long as possible.
No latency issues, no pseudo cores, zero stutter.
@@tobiaspabst5524 I don't blame you. That must be one killer rig for its age. Will surely last you a long time. Enjoy man.
@@tobiaspabst55244600 cl16 dual rank? If it is, you better downclock that thing for slower degradation... I've only seen a couple of dudes reaching that on dual rank on apex but it requires 1.35-1.45V SA/IO usually. If it's single rank, 4200-4300mhz will be faster in games with less voltage requirements.
Myself rocking 9900K with dr 4266cl16 33.8ns, that cpu is very snappy and fast in games. I have a comparison incoming between tuning vs 3200cl16, crazy difference. That's about 5700x3d in gaming.
Holy Hell...I knew it everyone told me it was all in my head. About 6 months ago I gave my brother by 10850k 32mhz cl17,19,19 system components as I went to 13700k ddr4 4266mhz cl15,16,16. And i 1st noticed on startup once i had same startup programs being loaded as the 10850k system.(wish i had a record of the startup time). I had issues when dragging and dropping many multiple large GTA5 mod files as i usually have 4 or 5 different copies at once. Becae a huge pain would cause me to make stupid mistakes like droping files were i did not want to. Got so bad that i took mobo and cpu back to Parkville Microcenter but since it was 34 days since purchase i could not return but exchange. But i did not figure 2 chips or mobo could be faulty so i exchanged for the same(AHHHHH). And of course there was no change. The guys at microcenter said it was all in my head exept for the manager that told me he had noticed it and about a half dozen other customers complained about it as well.
Great video as always Brian!! Thanks again!!
I shoulda kept the damn 10th gen system, screw my little bro LOl LOL LOL. I demand reperations from INTEL....Good luck right???
It would be interesting to see these tests done with a non-heterogeneous chip like the i5 12400
I have the 12400f and I can say that these input latency does happen specially when opening multiple mp4s
Non heterogeneous is simply called homogeneous
@@username8644 🤓
And the i5 12400, if it's an H0 stepping, doesn't just have the P-Cores disabled at a physical level but actually uses a smaller die that only physically has 6 P-Cores (and no E-Cores) to begin with; I thought that might be an advantage too (shorter bus with less stops, etc) vs having a full die but with the E-cores all disabled.
Tech YES hss been on a roll pumpin out banger after banger lately.
DOS had lower latency 😊
When hardware was a constraint they made software actually run well. Imagine if the conservative programming of the past was applied to today's software on current hardware. Everything would be instantaneous.
saw ddr5 6000 and closed the vid. thats all i need to know thanks
Крик отчаяния по поводу единственного способа ускорения твоего 13 поколения. Не расстраивайся, дураки тоже нужны
I knew I wasn’t crazy
The 10900K is literally the last of dying breed 10 REAL CORES vs all these pretend E cores 😆
It`s a nice CPU but really not snappier than a new one.
There are other problems happening here.
He`s playing us, the explanations are ridicilously unscientific. A windows reinstall will do miracles.
@@brugj03 hey we got a scientist over here.
@@brugj03 I mean he literally informs you at the beginning of the video that it's a fresh install of Windows 10 and 11.
E60 M5 V10.
@@brugj03he hasn't isolated the ring bus by turning off the E cores, didn't even mention it. E cores are for 'Cinebench workloads'. If he isn't using those, turn em off.
In RAM overclocking communities, this isn't super obscure information. I don't know the why, all I know is some architectures are just better than others. I know DDR4 is a contributing factor too because even though it doesn't have nearly as much bandwidth potential, it tends to do better latency-wise. The amount that DDR5 factors into it will probably be less and less as DDR5 matures though. AMD also generally does worse than Intel in terms of latency (Not as in their architecture to architecture variance, but they simply tend to have slightly worse latency results than Intel).
The problem is RAM overclocking is fairly niche, especially outside of enthusiast overclocking communities. So not many people are aware of this.
That's because ram overclocking isn't easy and actually requires a lot of knowledge on the actual architecture.
@@username8644 I agree on the first part but not completely on the second. I think it can help to understand the architecture you're dealing with but I don't think it's required.
There are some examples that come to mind, like knowing that 12th/13th gen IMCs are largely random and not binned better the higher you go compared to most other modern CPUs, or that locked 12th/13th gen CPUs have a locked SA voltage.
Other than that I can't think of much. I'm open to being wrong though and i'm no expert by any means.
@@labombaromba Ram architecture, not CPU. You need to understand all the timings and latency, and how they affect one another. There are literally like a hundred things to know and keep track of if you are serious about ram overclocking.
Edit: compared to CPU or GPU overclocking where you really don't need to know much.
@@username8644 Ohhh okay I get you now. You're absolutely right on that.
There are so many different variables compared to CPU and GPU overclocking it's crazy.
DDR5 can have worse latency with its first request command compared to DDR4 with higher TCL but anything else including total system latency on DDR5 is superior. DDR5 is designed to have more banks and bank groups, longer data burst lengths and 2 independent 32 bit addressable sub channels. A lower aida64 latency number doesn't always mean better latency when RAM has many different timing commands.
I'd love to see the comparison with AMD and 11th gen. I believe people noticed Ryzen had better mouse latency when it came out because USB was built into the CPU die.
Does 11th gen also have this issue or is it the same as 10th generation?
@@alikiyani420 as it seems, the 11th Gen is still Monolithic microarchitecture, so it should NOT be affected by this issue. Even more interestingly: according to the Bard AI ... the last monolithic consumer CPU from Intel is: i7-12700KF !!! .......imagine pairing this one with DDR5-6000 CL30 RAM! :O :O :O ...... I guess You would get *nearly* the same responsiveness as Ryzen 7800X3D !
@@overtonecz Is the 12700K (non-F) also monolithic?
I don't know what problems or weird bugs you have for your systems, but i have a 12700K and DPC latency is lower than your 10th gen, 12th gen and 13th gen systems including the i9s.
I use DAWs, lightyears more CPU intensive than Premiere Pro.
Here is what I see.
As an owner of 13900k, I tune my CPU, but don't overclock it.
It has a random frame time spike and it is consistent, but yet random to what is going on in games. It actually shows in other people's benchmarks with 4090 and 4080, I see it too.
LatencyMon completely hard locks my system. I tried different ram 2 motherboards, then my previous one would just say that my PC is not meant for real time recording in LatencyMon.
I have drop outs in my professional audio equipment with 13900k, but amazing fps in games, with an occasional single stutter. Latency in LatencyMon is high, until i turn off High Precision Event Timer in Device Manager, but the stutter does not go away. In fact I saw frame to frame latency get worse with HPET disabled.
LatencyMon still breaks my system on both of my Apex Encore Z790 motherboards.
9900k has none of those issues.
Something in the E-Core/P-Core/Internal Memory Controller of 13900k and the way Windows 11 scheduler talks to it. I saw none of this random stutter with 7950x Ryzen 9. I speculate difference in core frequencies does make a difference, even though I have SpeedStep and SpeedShift turned off. I have 7950x coming to test it, so we shall see if it exhibits same issue. By the way, I tried two 13900k CPUs, same issue.
Head to my channel to find a way how to remove 95% of stutters on 13th, 14th, and 12th gen Intel cpus. I found it this week. Note that new Arrowhead CPU by intel will not have hyperthreading, and I understand why now. I stumbled on partial solution to my stutters. Stutter happened in Ready or Not as you first spawn in, I assume it loads everything to L2/L3 cache, but it's very visible.
Perhaps you shoud have tested the 13th gen with DDR4 RAM like the 10th gen. Not only would that even the playing field a bit, but also potentially remove another unintented difference in results? Because what if the difference or the latency issue is caused by (or at least it plays a part in it) the DDR5 RAM? I mean it shouldn't.. but at this point I have no idea whats possible and what could cause an issue and what couldn't, so why not just even that playing field by having both platforms be on DDR4?
That's most likely the cause. His information is completely wrong about the Northbridge being moved off of the CPU die.
I have a i7-12700 with DDR4 and have never faced those issues ... He should have also tested with a completely different hardware setup for the 13th ... Because probably that is the root cause of his problems ...
I was wondering the same thing
I went with DDR4 3600 CL14 (bdie) dual rank dual channel 4 sticks with my 13700k just because it was an interesting low latency combo
@@Horendus123 are you sure it was low latency? Have you run the same set of tests he did in the video? I'm asking this because I'll have to make a purchase decision soon.
@@Dhruv-qw7jf am I sure the RAM I have is low latency? Yes it is objectively considered low latency and best DDR4 has to offer. Have I run a battery of tests like in the video. No.
This is an EXTREMELY interesting topic, not that niche actually, especially these days where people are BORED with new generations not giving the expected performance uplift.
I truely think you could have gone a bit further because your findings deserve a closer look to not let things unexplained.
Let's not forget that Intel's 12th and 13th gens are heavily tuned for idle efficiency, running on the e-cores, these CPU can draw as little as 6W (as for my 12700k) in idle, so it is not that surprising to observe a somewhat higher latency on these chips in some specific situations.
Some ideas for a part 3 :
* Disable e-cores and run the benchmarks again
* Compare and show the operating frequencies of the CPUs
* Disable any power saving features and C-states in the bios to see if it has an impact on latency
Your content is catching my eye more and more these days, and I think you are making the right choice not trying to show how many FPS a 4090 can pull out of a 13900k in 1080p. People don't really need performance databases, they need to be shown if they are gonna SEE a performance uplift from a hardware upgrade. This information is actually hard to find.
Have a great day.
From France :)
Max DPC latency on my 12700K (5 watts iddle, not 6) is 50 microseconds, average is 20 microseconds.
It's an MSI Z690 Pro A DDR5 motherboard, original 2021 BIOS.
What higher latency?
I do stuff with digital audio workstations and emulators, the 10900K is slow AF, 2015 Skylake cores. Literally is not that different from laptop CPUs at the time for singlethread speeds.
Adobe software has stability problems.
My 12700K is an AVX512 batch from 2021, i disabled the e-cores, enabled AVX512, the OS felt slower, and i lost fps on Switch emulation that supposedly supports those instructions.
P-core frequency is exactly 4.7GHz all core turbo and E-core is 3.6GHz, max turbo in one thread is 5GHz.
I use Super Smash Bros Ultimate as a CPU latency/IPC stress test with 8 ice climbers lv9 CPUs, one P-core with 100% use, frametime is perfectly flat and exactly 16.67ms, not a single stutter (with all shaders compiled).
I'm glad that i have an MSI motherboard because it simply works. I heared bad stuff about ASUS and Gigabyte, like bad quality VRMs or other problems on Intel.
The i7-12700K is a beast of a chip, meanwhile i heared that the Ryzen 7 5800X3D has problems because of clockspeed jitter (maybe a thermal issue, power delivery or sillicon problems).
If i went for a cheaper motherboard, maybe i would get jitter.
I'm 100% that the motherboards used on the video for 12th gen and 13th gen systems have worse quality than 10th gen, it's a downwards trend that i see since Rocket Lake.
I'd be really curious to see how Zen 3 and 4 stack up against these.
I don't know about Zen 3 or 4. I run a Ryzen 5 1600 on my win 10 rig for games. When I browse or manipulate windows around, I don't seem to see the latency, or when selecting multiple files in file manager.
7000 series have a different problem, the off-center placement of the dies under heat spreader, which causes typical cooler installation not all that efficient (and there's now offset brackets to remedy this, which is kind of ridiculous).
Quote from Calypto's Latency Guide:"Earlier Zen CPUs consisted of groups of cores called a Core Complex (CCX). Each CCX has four or less cores, and there are two or more CCXs which are connected together via the Infinity Fabric. The Infinity Fabric is fast, but not fast enough to not have noticeable performance loss in games as well as reduced desktop responsiveness due to inter-CCX communication. On top of this, Ryzen CPUs also have higher RAM latency than desktop Intel CPUs. Starting with Zen 3 (Ryzen 5XXX), each CCD (core complex die) has an eight core CCX which greatly reduces intercore latency, and unifies the split L3 cache previous generations had. This brings massive performance improvements across the board, but unfortunately the memory latency still suffers due to the memory controller being located on the I/O die. For most gamers, a 12th gen. Intel CPU with efficiency cores disabled will perform better than a Zen 3 CPU."
@@zares_fnxexcept for the games that can benefit from the 3d cache
Which is 80% of games, and 100% of simulation games
I agree, I think a newer product should be better then previous all the time. That's lame ha
I didn't read through all the replies so I don't know if anybody raised these already.
The only other two variables that come to my mind would be to try a DDR4 motherboard so that you would end up using the exact same memory sticks and also to try disabling all the e-course on the 13900K and do a latency comparison again.
I raised that same issue. And I commented that at around 40-50mins since the video was uploaded, there were only about 25 comments back then. It's a bit sus that since then he's "heart-ed" a lot of comments saying the same things he did in the video but not the one comment that was actually suggesting something to update his testing methodology on. And you'd think it'd be obvious to compare a DDR4 system with another DDR4 system but he went with DDR5. That's a bit sus as well.
You can test all you want. If you don`t go the scientific way (real world benchmarks, memory speed tests and testing suits) then you`re just staying in the dark with what you want to believe.
It`s self deception and the further you go the crazier it gets.
Just read all the insanity here. Then you know what i mean.
Would really like to see this attempted on Linux. It looks more like a software issue to me. Would be good to get to the bottom of this
It seems to me like it's a windows related issue but only for Intel processors. I am using 1260P laptop with an eGPU to play games and while it plays games really well, it's stuttering when opening like 5 UA-cam tabs at the same time. This never happened with my 2700x, and I am only swapping from it because it had some other issue (it could have been either RAM, CPU or the motherboard but investigating the root cause and trying to solve it is a waste of time, so I am upgrading the whole thing). I am now going to assemble a new PC tomorrow with 7900X, 64GB DDR5 RAM, RTX 3070 and I am going to see if I see similar issues but based on my previous experience with 2700x I am sure I am going to see extremely good performance, and the cost is going to be worth it.
Wondering how my 7800x3d windows 10 stacks up against them.
Thanks for the effort you put into the testing! This is very important information
man idk my workflow is super taxing on the cpu and my first try on the new intels was the 12700k, this thing was faster than anything I've used (11900, 10700, etc), now with the 13900k it feels even more responsive and fast, you can have a big 3ds Max scene opened and After Effects for testing different compositions with over 100 gb of ram used and still it feels just fine, other than the noise from the fans you couldn't tell if it's under load or not, it is that fast, my only concern is cpu is it wants to run at 300w at the first glance of heavy work, my 360 aio is almost not able to keep up with it, there's no 1 single application or procedure that I feel slower or less apt than before, even considering I had to move to win 11 🤦♂
You may want to test the 12600 vs the 12600k. One is the fastest 12 Gen no e-cores CPU, the other has 4 e-cores. Comparing these two could be very interesting in light of your latency findings. Thanks for a great video.
Also because the 12600 is derived from a native 6 P-core die (not a full die with disabled cores), which may help given that the bus should be shorter and less complex then too.
@@rphoenix5908 Yup, the H0 die is much smaller.
Got to love the fact that you are quite possibly one of the only reviewers that has the ability to multitask on one pc enough to overload it and feel the difference. Keep going with your gut feelings on products, it is what sets you apart. !Cheers!
Seriously you are the best, do not get a big head, but yes! If you tube as a whole does not see it so, they need to check what the actual user is into.
I would love to see how the 11th Gen (as well as Ryzen) compare to the 13th Gen in these kinds of tests.
I'd like to see latency results for 11th gen Intel and zen 3 and 4.
I was thinking the same. I also wonder if turning off the E cores would help. I’m not even doing this workflow but I’m just curious because I like nerdy stats 😊 I’d also love to hear about this from an actual engineer.
@@wertywerrtyson5529he already tried disabling E cores before part 1
i remember latency was bit better on zen 3 then 10th gen and better than 11th gen
Latency on 11th gen is outstanding
Same. While I9 from 11 gen was generally considered a trash product, the I5-11400 and i5-11600 were widely considered good products. As an owner of one of these I am curious how much this topic is accurate for 11 vs 12 gen.
By the way, those are microseconds (µs), not nanoseconds (ns).
Naaaaaa this isn't niche scenarios.. Everyone clicks and search files on their computer.
Not everyone just stares at the loading screen while playing games lol
This is be part of the standard for test benching new cpus. Latency is important when everyone is doing thousands of clicks a day.
If each click takes an extra second to to perform the function, that's 15 mins a day.. That's 91.25 hours a year lost lol.. fffff that
Have you turned off CPU power saving features in the BIOS? this affect action latency SO MUCH (example: opening video files).
If both systems were set up identically it shouldn't matter.
Honestly it's a breath of fresh air to see something non gaming focused on this channel.
Is the latency in gaming?
@@adami775he explained that in the first 2 minutes
@WayStedYou he just said that this video isn't about fps so I was curious if it affects game latency
@@adami775 I Highly recommend the channel battle(non)sense for gaming latency of all sorts.
www.youtube.com/@BattleNonSense
Can't wait till Bryan is back in Australia and do content like Tech Yes Lovin' of filthy parts with his cleaning methods. I love those videos. He is too limited in what he can do for content in Japan. It gets monotonous fast.
Went from 11900k to 12900ks both DDR4 tuned and the 12th gen overall felt more responsive and snappy.
Did you fresh install the OS or clone your old drive?
@@sezwo5774 Same NVME ssd (gen 4 z590 w/ 11th gen supported gen 4) fresh installed Windows 11. same RAM on both systems.
I have the 10900k and won't move from it:10 cores, 20 threads?! What's not to like? It's a classic! 👍❤️
I love my 10900k and change it in 2025/26
Me still on a 3800X that needs to communicate between two ccd’s via infinity fabric: “Ah, yes. Interesting.”
Dude Userbenchmarks is going to put a hit out for you.
Darn.. so now this, combined with 14th gen Intel being just a "refresh" means that I have to cross my fingers and wait until 15th gen to replace my 6800k/GTX1080 desktop. 5090 here I come. lol
I am extremely happy with my 10900KF @5.1ghz all cores @1.28v with 4x8GB 3600MHZ
Tune ram, you'll be A LOT happier
@@HeartOfAdel now i have gskill trident neo 64GB 16x4 CL16-19-19-39 3.6GHZ XMP but this ram wont overclock at all tried for days but happy with stock 3.6ghz
SpeedStep ramp.
You cannot go from one P-state to the next instantly. There are limits in place which P-states are reachable from which ones. During transition, the CPU is disabled, then clock and voltage adjustment takes place, and the CPU is enabled again. This takes a lot of time. You don't take one ramp step right after the other, you let the voltage regulators settle under load first and also if you did that you'd have a performance dip right up front, because you'd be spending a little too much time with CPU halted. Fine grained P-states and slower more conservative P-state transitions will make for a system which is much more power optimised but less responsive to burst workload.
I'm not saying that this is necessarily the cause, but definitely something you want to investigate given the symptoms.
Are you sure it isn’t just ddr4 lower latency feeling snappier ? Would have been more interesting to see if it’s the ddr5 or the board itself. I really enjoy your findings for sure. I have noticed ddr5 6000 in general has higher latency than the best ddr4 has the offer
The DDR4 used in this test was very slow (3000CL16). I guess he did not want to push the DDR4 fast low latency OCs here to not introduce even more variables.
@@innocentiuslacrim2290They should be about the same latency-wise, iirc 3200CL16 and 6000CL30 both work out to 10 nanoseconds, since CL is measured in cycles rather than an absolute time unit?
@@innocentiuslacrim2290 Unless there's a platform limitation with DDR5 that adds latency along the way, of course. I wouldn't know.
I upgraded to a 13600KF + Z690 (DDR4) two months ago and the system runs like a dream (Windows 10, E-cores disabled, C-states disabled, core parking disabled). I haven't experienced any of the examples you mentioned. Most of the time I'm running the Balanced power plan with the CPU set to 99%, which locks it to 3.3 GHz (same for the Uncore). I use Bitsum Highest Performance (locked 5.1 GHz with 4.0 GHz Uncore) when I want the full performance for gaming, recording or encoding video. I can fully load the CPU and the system is still completely responsive.
I even use a 5940 RPM HDD for most of my data - searching, selecting, moving files is super responsive. Obviously even faster with both of my SSDs - one connected to the CPU, other one to the chipset.
It has to be a configuration issue. Are your power saving options disabled? Is Windows running in UEFI or Legacy (CSM) mode? Are you sure you don't have High Precision Event Timer enabled and forced, which is a disaster on modern systems? Or some other internet tweaks (like useplatformtick) that usually have no positive effect, but often they break something.
None of it makes any sense. Please investigate this issue further before making these kind of allegations. This is pure speculation at the moment.
The only reliable way to test for "latency" would be to use tools such as xperf and windows performance analyzer and find out which drivers are causing input lag. LatencyMon should only be used to have a quick look at haywire drivers causing stutters, as the program itself causes stutters. Btw, the results are in microseconds, not nanoseconds.
Your ISR seems to be spiking pretty high due to presummably the NVidia drivers (cannot possibly tell as you haven't mentioned what you were doing in the background while recording this), the first tab doesn't tell us anything.
Use the 'Drivers' tab and sort them by highest execution, depending on what you're doing in the background you can trigger different results.
For example:
Bad USB drivers (u can easily trigger and see it in real time, especially on a 8KHz mouse) will use the wdf0100.sys driver
GPU will trigger interrupts when watching a video and using LatencyMon to record.
Bad ethernet drivers (Intel NICs seem to be pretty reliable)
Bad storage driver / port (changing port can alleviate stutters, M.2 SSD causes the least amount of stutter cause you're avoiding the SATA controller)
I'd recommend you to read through guides such as these ones:
github.com/amitxv/PC-Tuning
github.com/djdallmann/GamingPCSetup
Your 13900k could also not be completely stable, thus leading to these irregularities during file transfer (or perhaps windows not being setup right / storage driver or hardware playing games)
Try testing it in bootable porteus linux linpack (updated with latest 2022 libs from SlovenianSlobodan) or this version (testing through windows doens't reveal instabilities easily):
github.com/BoringBoredom/Linpack-Extended/
This is ringing a bell. I feel like I recall Hardware Unboxed mentioning differences in percieved desktop "snappiness" between new CPUs in one of their CPU reviews. I'm pretty sure it was AMD vs Intel CPUs rather than intel generations though, and it was either 12th gen vs 5xxx or 13th gen vs 7xxx at the launch of one of those things. I very vaguely recall the observation being that AMD had caught up to intel in terms of the desktop experience feeling snappy. Now I'm thinking it wasn't AMD catching up, so much as Intel losing a step. There is a possibility that it was jayztwocents and not HUB. Or maybe it was both. Either way, there was no discussion beyond the observation.
@techyescity could you please test 12/13 gen with ddr4 in part3 ? I have 12900k and a 13700k (2systems) and I opted for ddr4 on both and I see no issues, tested with ddr5 in the past and I was hiting some issues in windows (slowness etc), maybe this was the issue? The ram controller on these new cpu? Ddr5 latency?
We need to see how AMD does in your tests! The R9 5900HX has a DPC latency of 116 so even lower than 10th gen. Also my AM4 desktop is super responsive in Windows 10 with Atlas.
Curious, does this issue also happen with hte 12900k? or 12700k?
yes
@@techyescity just saw the part with the 12700k at the end.
You want to know what's crazy I just pulled out some of my old boards gigabyte aorus z270x gaming k7 i7 7700k only 16gb ram and m.2 3x4 for storage. And I must say it is seems to be way snappier than my newer PC smh
I have an 11700F system and producing a video on that one, compared to my 18 Core Haswell X99 HP Workstation, I was underwhelmed. I mean in Cinebench and other tests the IPC is much faster but editing videos wasn't multiple times better, I expected a lot more. It's nothing like going from a Pentium to Pentium II THAT was an upgrade!
I have an Erying MATX motherboard with a soldered on Core i9 12900H laptop CPU(14 cores/20 threads w/6P and 8E cores) , and with 32GB of DDR4 3200Mhz RAM it seems to do much better on Manjaro GNOME Linux in traditional layout with way less lag than the brief time I spent with Win 10, although I do have a couple of the P cores that like to do random brief spikes in temps up into the 80's, and 90's but go right back down with a down firing CPU cooler to help cool the VRM, and RAM. So yeah I'm overall happy with it having come from a an AMD A10 5800K with 16GB DDR3 RAM, and an AMD RX560 4GB to a AMD RX 6650XT 8GB GPU.
@@CommodoreFan64nobody cares
@@toseltreps1101 Well, from a Commodore 64 to an A10 5800K was a huge jump! But C64s were nevertheless great fun back in the days, no need for CPU and GPU upgrades then... and still many generations of games
This is just the video that I've been looking for. I've been so tired lately of seeing "gaming benchmarks" and "synthetic benchmarks" It's so unrealistic to someone who relies on their computer for work and multitasking day in and day out. thank you tech yes city
I wonder if this issue persists that badly in GNU+Linux.
Exactly this. Not testing the biggest part of the software completly invalidates the test results. This here is highly unscientific.
Maybe it is just a Windows problem, as on my Linux box it is not like that? Also, I have DDR4, not DDR5 with my 13600KF.
EDIT: You should really test on different operating system, not just Windows, as that really might not be a hardware problem.
EDIT 2: Jay mentioned a long time ago that AMD feels snappier than Intel with 12th gen.
Dpc latency is very important when it comes to do any kind of real time audio processing/producing(audio buffer underruns) if its to high
"Moore's Law Is Dead" recently had an audio-focused custom integrator as a guest, and he seemed to be quite unhappy with recent Intel releases.
I wonder if this is related.
In broken silicon's recent video with a DAW enginee/integrator, in the discussion about sapphire rapids he explains he's stuck with providing 10th gen audio workstations specifically because of latency introduced after that gen. He also mentioned turning off core parking had an effect on latency, don't know if you've already looked into that or not.
I can confirm turning off core parking does help dpc latency in latency mon for a 12900k.
Seems related to Ryzen audio popping issues.
I just listened to this episode a few days ago and was wondering the same. Maybe Brian can revisit after tweaking core parking. Hopefully Intel is not regressing on responsiveness in pursuit of more power efficiency.
Weird that the latency is higher even on 11th gen, I would've thought the difference was due to the hybrid core architecture introduced in the 12th gen.
@@utubekullanicisi the geared/hybrid memory controller on 11th gen is at least 1 reason why latency is higher on it and why even though it has higher theoretical IPC it basically matched performance in gaming as an example with 10th gen.
Funny this has been my exact experience i thought i was crazy. Our 10850k server felt so much smoother to use directly than any 12th gen
Why did you not use 6400 CL32 for the new Intel system? That's the DDR5 equivalent of 3200 CL32 DDR4 RAM.
13:15 Last Tuesday Windows 11 update caused again a slowing down SSD and BSOD's for some people.
The slow down SSD is a months old problem for W11, they are struggling to solve it.
LatencyMon measures MICROseconds, not nanoseconds. The "μ" in μs means micro. Nanoseconds is written as "ns".
Also, there are several problems with your use of this tool. Firstly, 1 minute is not enough to compare 2 cpu's because drivers do not get loaded multiple times a minute.
For that same reason, looking at the maximum latency doesn't make sense. There might be some dll loaded at the same time as another driver on the one system but not the other.
You can even see in your screenshot that dxgkernel was loaded on the 10850K but not on the 13900K. If dxgkrnl was loaded on the 13900K it would have a higher ISR time as well.
In short, the results you are looking at are worthless. What you should be looking at is the "current interrupt latency" to see if the two differ when at idle or when performing the same tasks, but even then you'd need multiple runs to decrease the chance of recording background dll calls which happen randomly.
I would look at the DDR5. From memory people complained that the larger timings can slow things down, ie, it needs to be optimised, or even switched for better timing DDR5. Plus you have to look at the numbers of channels, and the number of ranks. Basically the DDR4 setup might have more than the DDR5 setup, ie, it's not apples to apples.
I have been utterly disappointed with the i9 1300k so far. I downgraded to this flagship from an i7 8700k which was blazing fast with display operations and program launches, menu navigation etc. My wonderful top end system with 128GB of ddr5 and the 13900k sucks
Great video Bryan! I had huge issues with DPC latency back a decade or so - and no one on any of the PC tech support forums knew what on Earth I was talking about. Programs like LatencyMon really helped me pinpoint certain driver and hardware issues that was causing massive latency spikes that was creating very much audible "glitching", especially in audio work.
It's interesting, I just did a fresh install of W11 in my 10900k rig and it's significantly more snappy than it was on W10. But I am willing to chalk that up to 3 years of wear and tear in the OS. Overall though, I think W11 is significantly easier to debloat. The whole OS is just more cohesive which I feel contributes to the ease of tuning. Boot times with all programs launched is lower, and initial CPU and RAM usage on idle is significantly better. Subjectively, I also prefer the more cohesive UI of W11.
I also have my 10900k highly tuned with a custom overclock, which is actually undervolted and runs at 5ghz all core turbo, slightly higher than stock 4.9 while also retaining the 5.1ghz 4-core boost behavior. Full load TDP is significantly lower than stock, tops out at just under 220W in Cinebench. I was never able to observe the 5.2/5.3 super marketing turbo modes in any real world scenario, so I just ignored those. I could enable them in BIOS with this OC still, but like I said I was never able to observe them even with the stock configuration so I just decided to leave those out of my OC. A 13900k would definitely be better for gaming, but I am yet to run into a situation where I am CPU bottlenecked with my 3080Ti. Will likely upgrade the CPU again before I replace the CPU
I noticed a generally less snappy system switching from 11600k to 13700k, glad I wasnt the only one. Thanks for the videos.
Ddr5 might have something to do with it too. It still has higher latency compared to most ddr4 Kits.
@@htoomyatlin123thought this myself but 13 gen performs the same on ddr4 & ddr5 in terms of latency (I have both boards). Its something to do with architecture of the cpu.
@@pixelztv5254 hmm. Interesting
I guess e-cores are just too slow even for general use. Intel should just go back to making big cores only CPU.
@@htoomyatlin123 hopefully with time these issues are fixed!
@@htoomyatlin123 They do sell 6 p-cores only chips, 12400/500/600 CPU's. I always wondered why not an 8 core equivalent.
From techyes, to techno.
You are once again claiming things without proper testing on different hadware setups, and without any hardware, software, OS engeenering knowledge.
You are once again fooling your viewers instead on approaching the question, with proper research and testing.
You are not able anymore to provide contents, outside the review materials and products companies send you.
I have still not forgotten how you blatantly followed the AMD marketing materials sent to mainstream reviewers, when AMD asked to create content, on how good the 7900 was at playing COD.
I'm having this latency desktop issue from Core 2 Duo era. All my builds had in common one thing: NVIDIA GPU's. And it seems, it is true. Some hardware combinations with NVIDIA GPU's has terrible a terrible desktop latency. Recently I bought some Ivy Bridges, and with Intel GPU, they are amazing and responsive. Tested this with a GT 730 and the latency is back. I think in the case of Intel iGPU it makes sense as long they do use a internal bridge different than PCIe.
Why didn't you test input latency in gaming? In games like Overwatch 2 latency is more important than framerate. Also why no comparison with Zen 4?
It’s a needed test so well done for doing it so thoroughly.
Two things I’d add as comments. Yes we expect the next technology to be better than before and use performance as that metric. But that isn’t always the case. In recent memory there has been cases of a failed architecture (AMD FX) and outliars from Intel (Any of the CPUs Crystal Well) which showed performance differences in different case.
The second point is about test bench work and here I would say that real world testing, use cases and ‘feel’ is actually more important than simple average FPS numbers or synthetic benchmarks. It’s much harder to capture but it is far closer to the real world use a PC will go through for most people. That makes investigations like yours absolutely priceless so don’t do yourself down for being the only one doing it. It’s important work and thank you for doing it.
I'm just here to feel good about my 10900 😂
one of the reasons i keep my 10900k and havent dropped in 11900k is sub 40 ns latency, outside of 2 additional cores