Watch the Extreme Overclocking of the R9 9950X: ua-cam.com/video/pYWtP4tZe30/v-deo.html Watch our video with Wendell about the Intel problems: ua-cam.com/video/oAE4NWoyMZk/v-deo.html
all that matters is that cinebench score. click, run, collect data. can we stop now. both companies have burned chips with this stupid behavior because of reviewers just benching.
Thanks. Looking to check out that Ryzen 9 in action. I've lucked out on 12th, after a lot of BIOS tuning have the 12700KF & 12900KF gaming stable. I really lucked out last year in not buying 13/14th and having degradation potentially. I hope everyone gets stability though those degraded chips it's likely lower voltages (and likely at lower ram speed) due to damage. Great piece on the Intel issues. I found it useful in the past couple days disabling the ecores has helped me. Mostly being gaming and file server stable are important in my daily workload.
It is the most reasonable option for consumers as well, to be honest. Big like and it makes sense if you are building to specs (like PCIe and DDR gens), which provides you max theorietical numbers. Engineering is better with well defined boundaries.
@@Damicske Socket am2+ cpus could be used in most am2 boards, as long as the mfg did a bios release for it. In addition, AM2+ brought a much improved hypertransport, cpus were moving WAY faster back then and things got real limiting real fast. Socket AM3 brought ddr3, lots of AM2 and AM2+ boards supported AM3 cpus. AM3+ had a new pin, so bulldozer cpus could not be used on AM3 or AM2 boards, they lacked the memory controller compatibility for ddr2, in addition the older boards could not supply enough power, so it really wouldn't have worked anyways. They have always been trying to keep sockets around, but 20 years ago a cpu didn't gain 15% over 2 years. We were seeing 100% gains, going from 1 core to 2, increasing clocks every gen, then 2 cores to 4 with AM3.
@@GewelReal I don't agree. It's bad for the environment . Many AM4 users will be forced to throw away their existing cooler which are perfectly good. Not every company is like Noctua that provides mounting for free.
You forgot about the x670e-e Great boards, but some of the stuff from 870 looks real tempting, like usb 4 support. We need 10gb and multiple USB4 ports on all boards, we've been stuck at 1gb and 2.5gb networking for WAY too long.
@@michaelmoses8745 like the whole spec in a UA-cam comment? It generally requires double sized registers of a specific number, wider data lines and much much more transistor space. This normally results in limited clocks as the silicon is tuned for one workload type at 64 bits. You can see the full specification on Intel's developer site or Wikipedia or chips and cheese.
@@bosstowndynamics5488 It was explained in the video. Zen 4 is still a 256 bit AVX arch and it's double pumped to get 512 ISA working for a mix of performance and efficiency at a loss of raw throughput. Zen5 has a true 512 bit data path when needed with the double pumped pipeline available for presumably less performance centric workloads.
Introducing a brand new PCIe 3.0 motherboard in mid-2024 is wild. Kinda slimy to call it B840 when it's very clear that it should be called A830 or something.
It's an anti-consumer move that only benefits motherboard manufacturers and cheapens AMD's platform reputation. Misguided decision from whichever AMD exec allowed this.
@@fleurdewin7958 Actually Sandy Bridge-E brought PCIe 3.0 13 years to the HEDT platform. I had an i7-3820 on an X79, and "upgraded" to a B450 in 2019. I was shocked that my second NVMe SSD was slower on the B450, and I learned that the chipset provided PCIe 2.0 lanes. But yes, B840 is atrocious and malicious and is deliberately meant to confuse buyers. Just like the Zen 2 Ryzen 7000 chips.
AMD said something to effect of "AI" when they spoke about branch prediction in Zen 2. To be honest it does qualify as AI simply because its job is "decision making" (a thermostat qualifies as that too), but it doesn't qualify as machine learning / deep learning / generative AI.
@@CossackHD to be fair, he was speaking of it in context of vector instructions like AVX. Which are pretty useful to do matrix/tensor math, which is heavily used when doing actual AI task. Meaning AI as in: doing training or inference with a neural network. So as AI talk goes, this was actually on the lower side of the marketing BS scale.
@@Hugh_I Agreed. AI is over-hyped, but is still a real thing. We can see "AI" in things like realtime background removal tools. It's just not the cool "AI" that people think of.
@@stephenallen4635 fair enough, but I'll gladly give up over clocking for PCIE 4.0. over clocking is useless unless that's your bag, but PCIE 4.0 is good for everyone.
For my entire life prior to AM5 every CPU upgrade was an entire new build, and forget the component costs it takes an entire weekend to strip out the motherboard and re-manage all the cables. Upgrading to Zen5 this fall is going to feel sooo good, four screws, plop in the chip, done.
who the fu** is going from Zen4 to Zen5, i get going from Zen2 to Zen3X3D, wait for Zen6 dawg. I went from a 1600AF to a 5800X3D, now i'm waiting for Zen5 and Arrow Lake to launch so i can make my choice.
@@MaxIronsThird could be someone going from an entry level 7000 series cpu like the 8500g to a higher end 9000 series cpu. Could have been waiting it out like by purchasing something to get by a month before the announcement, or could be cause they want to invest more in a PC for various reasons
The new chips will be launching on the 600 series boards initially, so I expect demand for them will go up in the short term, not down. If history is a guide, after the release of a new architecture, AMD will offer great Black Friday/Cyber Monday deals on the last gen stuff
For my audio production usecase i'm most excited about doubling the instruction throughput since most digital audio workstations rely directly on AVX2 instructions so this should, at least in theory, make them incredible for low latency audio processing. These seem like they're going to be absolute monsters for creative work.
I suspect you will also need the cache also the cpu will likely be resting at avx2 avx512 is much much more intensive zen3 and zen4 already had full support pipeline for avx2 dont get me wrong, it will be faster, but not that much faster
@@s-x5373 cache size is really only crucial for audio production when using sample based instruments like Kontakt or Omnisphere, in those scenarios the x3d parts do better than anything else but they actually perform worse than the non-x3d siblings in everything else like overall project size without buffer underruns and with digital signal processing. This is where IPC and clock speed combined are extremely important. CPU's are certainly not resting while executing the AVX2 instructions from the audio engines, idk where you're getting that. And as far as I'm aware to date theres very very few audio production software that makes use of AVX512. Modern digital audio workstations can be incredibly demanding of resources, in large projects I can easily engage over 75% of my Ryzen 5 3600's 12 threads, even more if I push it. Either way, If they actually doubled the instruction throughput then these chips will be a seismic jump for this type of work, of course that remains to be seen.
AVX is SIMD, single instruction, multiple data, not really related to IPC throughput as it's about reducing the instruction pressure via compiler tricks, Early Intel CPUs actually throttled down and caused huge performance issues for other code under AVX 512 workloads. The real issue is Windows (and generally any preemptive multitasking OS) is a shit OS for realtime work, and interrupt based CPU designs like x86, multitasking systems absolutely screw up your latencies as the OS will pause and switch out different apps, hardware will force CPU to pause computations via interrupts, your shitty closed source OEM drivers will hog 1ms of CPU and ruin your latencies.
@@s-x5373cache is really only important for using lots of sample based instruments like Kontakt, in every other scenario (general project size without buffer underruns and digital signal processing) the non-x3d outperform the x3d variants so the bottleneck with audio engines is IPC/clock speed, which at least on paper should be a big increase. We'll have to see when they come out but I've got a feeling these are going to be a massive uplift compared to Intel chips for this.
Where do you feel like you're being held back right now? Recording on PC is not like high end gaming where frames win games. The latency was low enough in the mid 20's when I was doing it.
About 15:35 where you mentioned that AMD and Intel architectures both have challenges when it comes to scheduling, I think that you missed the point Macri was trying to make. There is a massive difference between misscheduling on a 7900x3d and a 14900k. In the first case, you might schedule a clock-friendly task on a high-cache core or vice versa and lose a bit of performance. In the second case, if you schedule an AVX workload on an efficiency core, you get a kernel panic / BSOD. This means you can run older OSes on an x3d CPU. It might not be as good as it can be, but as a user, it will work, and as a dev, you don't have to backport complicated scheduling algorithms to older kernels still in long term support. On the hand, you absolutely have to do this for intel CPUs. They need additional dev time, and that needs to happen before release; and potentially in secret. This is why the initial launch of p/e cores was such a mess. Some apps would randomly crash on Windows, which got support early, and Linux simply didn't work for weeks because it would've leaked the IP. So, AMD doesn't _require_ special drivers, but they do benefit from them, and drivers are the work of the hardware provider putting the onus of producing and maintaining this code on AMD. Intel requires a special scheduler, which is the responsibility of every OS provider and an additional burden on the communit in the case of Linux.
Linux has long had support for heterogeneous compute in the scheduler, because it's important in the HPC world. So both half-X3D and Intel P/E CPUs are directly supported by the kernel. What's weird to me is that neither manufacturer has lent on M$ to get the same support implemented in Windows.
@@GamersNexus Make sense. Though the way you word it makes it feels a bit like misrepresentation because of that. I totally get where you're going though. For a lot of people, if you buy a product and expect it to work as advertised, in the end, there is no difference between a crash and a 20% performance loss. They're both bad situations for consumers, and it's easy to think how less technical people could be convinced by claims like this one.
@@sykescalvin09 Microsoft admited in a marketing way of talking about it, that they already fully knew what they have to do to better manage it. But they are more focusing on building a IP bullshit out of their bad ripoff of existing solutions than pre-emptively testing and developping this ready. And if they had, they would have had pushed that locked on microsoft server licenses first.
@@sykescalvin09 Well you said it yourself: it's important in the HPC world. Windows doesn't really show up over there so it simply wasn't a priority to them. You could even say it's a strategic advantage to wait as it forces people to upgrade to newer version of Windows if they want to use those newer chips. Just to clarify, even if linux did support heterogeneous compute for a while, it still requires every specific core's capability to be declared to the scheduler for it to work, and Intel's P/E required custom code for it to work. Those changes couldn't be published before the release embargo lifted which delayed P/E support for several weeks.
I feel like if anything it would double to 16 but who knows. A 12 core R7 10700X or whatever would be awesome, with a 24 core top end R9. The 16 cores for the #950X is starting to get old. But at the same time, they're getting so fast that in most cases a consumer won't need more from Ryzen 9. My 7800X3D barely runs slower than my 5900X in multi-threaded apps, and in games it's miles faster.
At 4k Ultra/High you will be limited by the GPU It will make no difference if you run 5800x3d 7800x3d or 9800x3d They will all perform exactly the same. But if you're wanting to run super high refresh rates at 1080 on low, then you might have a point. Who knows, that might be your bag. But if you have any of the previous x3d chips, run your game at 4k ultra/high, you would be far better off staying with those and using the money to buy the 5090 instead. You're going to need it.
I built an AM5 PC somewhat recently and am happy AMD seems to be doing well, especially since recent intel xx900 seem to have serious problems. It's good that there's more than one mainstream CPU manufacturer, im case one messes up.
19:20 Although I wouldn't be excited to see AMD lose CPU longevity, that would be an AMAZING episode 10 years from now to see this video dug up and some great content made out of it!
I'm really eager to upgrade as well, but it seems that waiting is just the smart choice for me here, as much as it sucks. Hopefully, they'll release the X3D chips together with the new motherboards so we can get a good comparison and make the most informed choice possible when upgrading to this gen.
@123Suffering456 That'd be nice. But like others have said. X3D hasn't been released for like half a year after thr regular versions. X870E boards are rumored to be out in September
it seems so easy to have multiple x16 slots with how amazing these processors have become but yet most brands are only putting one on the boards. i may have to keep my x570 boards forever.
They don't do it because SLI and Crossfire are dead (as in, no physical connectors to bridge GPUs together anymore). Without that, there's absolutely no point in more than one x16 slot for practically anyone buying a consumer desktop level board. All it would do it is add cost for something no one can even use anymore.
@@wasd____ graphics cards are not the only thing using the x16 slot. Personally I use 1 for a gpu and the other for a 4 slot nvme drive. They will have the slot but instead of two x8 it should be two x16.
@@boosted94sc300 I mean okay but a 4 slot NVMe drive is still a super niche thing for a consumer desktop system that almost no one is going to use. If you need _that_ kind of crazy bandwidth for storage you're probably doing something well beyond the consumer level and really just need to be looking at a higher-class motherboard to begin with.
The CPU and socket is explicitly designed with a specific set of PCIe lanes, an extra 12 lanes is an extra expense, and explicitly why ThreadRipper exists.
@@wasd____ I bought a consumer motherboard, the ASUS B550-XE STRIX because it came with a PCIe 4.0 4 slot NVMe riser, had 3x X16 slots, and supports a total of 6 NVMe SSDs out the box. AM5 is just expensive for being expensive. You get garbage overpriced motherboards. Then people think a "$150 board is $150 quality".
Good video, as always. One quick note, regarding the nice catch on the CCD discrepancy at 10:40 or so: while I totally agree that, ideally, one should try to eliminate as many variables as possible, however minor they may be, when doing these sorts of marketing demonstrations--especially given the fact that I can't imagine AMD themselves could not find more comparable parts on hand--in this particular case, if I were to guess, I would assume this particular instance is actually a less-than-ideal case for AMD. It would be reasonable to guess that, assuming all other attributes are more or less equal (e.g., boost curves, cache, thermals, whatever), that the chip with fewer CCDs (the 7700X, in this example) would have an advantage, albeit miniscule, over the other, since there should be less room for core-to-core latency and other things to affect the benchmarks. Of course, again, sure, it would be better to compare more equal SKUs, but out of all the marketing bullshit, this one seems tames, if not downright not in their favor on paper. --- Longwinded, and tangentially related: I wanted to email you all about latency testing, but never quite got around to it; have you guys considered looking into it for motherboard and CPU reviews? For the overwhelming majority of workloads, I recognize it is typically not something noticeable, but the industry trend seems to have been shifting from complex to manufacture and perhaps more latency-prioritized designs to higher-yield, throughput-oriented designs. We can see this across virtually all SKUs that I am aware of: - AMD's quirky Bulldozer shared FPU/SIMD design. - AMD's CCD/chiplet design in most modern AM4/AM5 (as well as Epyc, Threadripper, etc) processors. - AMD's "heterogeous" 3D-vcache weirdness (e.g., 7950X3D having one CCD with and one without) - Intel's various multi-tile processors. - Intel's heterogenous P vs E-core design. - Intel's switch from the latency-oriented ring bus to throughput-oriented mesh interconnects (IIRC, after the 10980XE; unsure about standard desktop platforms). Similarly, interrupt-based I/O, such as PS/2 ports are seemingly dead [1], with the last board I am aware of being the XOC X670E Gene -- and even then, I'm not actually sure those interrupts are "guaranteed" nowadays in the way they were in the past. Motherboards, as well, tend to pack in all sorts of bells and whistles, making it increasingly difficult to find something barebones with a decent VRM and PCI-E layout, without a variety of crazy onboard -- unfortnately, often not great -- LAN, audio, and various other miscellaneous chips etc, all of which the user has little to no auditability or control over, except perhaps a UEFI toggle that *cross your fingers* truly disables these things -- all of which can add up to be fairly deterimental, and that's before we talk about the actual, often pathologically behaving, vendor firmware/BIOS. The above may seem like a bit of a schizo-placebo rant, but for things like real-time audio, piano synthesizers, and so on, it can make a surprising difference. To the best of my knowledge, Anandtech was one of the few (only?) outlets in recent times that would routinely measure things like DPC latency as part of their motherboard reviews. However, unfortunately, their articles are few and far between. It would be very interesting to me if you guys found the time to integrate some of that into your testing methodology. And I would be very curious whether things like the relatively dogged-on (rightfully so) 8700G may turn out to have some secret advantages for certain workloads, being one of the few monolithic processors remaining -- similarly, I wish more reviewers checked out the w7-2495X during the SPR WS launch rather than jumping straight to the top-end 56-core model, as the
about that long comment, worth noting even programmers have generally switched from optimizing for latency to optimizing for throughput. there are cases where latency is critical, but in most of those, you're not only optimizing code, but also the network around it.
Welp, it seems making the switch from Intel architecture to AMD for the very first time since ye olde days of AMD Athlon was the right call for me. Not only is the 7800X3D an amazingly powerful AND power-efficient CPU for Gaming, I now also already own a motherboard with a seemingly still very high end chipset (X670E) and AM5 architecture that'll allow me to just put in a new CPU when necessary for years to come. Steve is absolutely right in saying that AMD also isn't without its darker aspects, but this is truly one of the best pro-consumer decisions a company of AMD's size has made for quite a long while, especially in the gaming segment. I've been gaming for over 30 years now, both on PC and console since the early 90s, and I've always fondly remembered the times when you bought a decent system for gaming and it was still decent 3 years later... seems like with AM5, these times might, at least for now, be kinda back again.
Had an Athlon as well back then, until AGP graphics were phased out and PCIe + Nehalem took over, 7800x3D on AM5 now as long as it will take me, here we go!
I'm so glad they are giving us more lanes, but I cant wait for when you can get 100 lanes on desktop. I don't want to switch to threadripper, way to expensive.
What I'm hoping for most from Zen 5 is if the 9600X will continue the legacy of the Ryzen 5s being top-tier performers in the gaming realm! If they can, I might have my CPU pick for my next build!
What do you mean by top tier performers? The Ryzen 5’s have become their low end chips. Ryzen 7 is mid range and 9 is high end. I would not recommend a 6 core processor in 2024 going into 2025.
We're at a point where it doesn't make sense to upgrade to a new low-midrange system. If you already have something that works, save until you can afford a high-end system. At one point, upgrading from an i5 to the next i5 was a major improvement. Now it almost doesn't matter.
@@ZackSNetworkFor gaming AMD's Ryzen 5s are generally the best bang for buck. Yes they are the lowest in the stack now, But their gaming performance is usually extremely close to the 8 core chips and the higher end chips are basically irrelevant for gaming.
I love the way y’all stamp clips being used for quotes with an actual profile of the speaker(s), so people are always credited and viewers get a basic idea of credibility on certain topics.
@NielsC68 I'm a renegade enjoying life on the edge. I also like teleporting, skipping, and hitching my way through CPU intensive games with glorious 1% lows that looks like a heart monitor after I blasted coke and ephedrine. It delights me to witness frame rates hop and skip from 28 fps all the way up to 108, only to Plumit again like the doomed aspirations of a minimal wage employee who spent his miniscule savings on a rug pulled crypto alt coin.
FX 6300 and (of all things) a NVIDIA GTX760 here. 16GiB of RAM, doubled from the original 8 I built it with. Data stored on a 1TB piece of 'spinning rust' that luckily seems to still be going perfectly strong -- I've heard that modern consumer-grade hard drives do not enjoy the same long-term reliability, and that NVME Flash memory (at least, models with large endurance guarantees) all around is the way to go when building a new machine. Running Linux Mint 21.3 Cinnamon, because I'm still a boring person, but have been unwilling to deal with Microsoft's BS since Win8 came out, and pretty much everything I'm interested in playing runs through Wine these days anyways, given my hard-line anti-DRM positions. It was my first scratch-built PC, and I'm really happy with all the use I got out of it, both when I initially built it with Win7, when I dual-booted it, and after I got rid of the Windows partition. But suffice it to say, I'm... here watching PC part review channels, in preparation for a new build sometime in the relatively near future (maybe after the X3Ds of this series come out?). Case doesn't have a space LONG enough for modern graphics cards, so this'll be a pretty much completely new build. Probably AM5, planning on going AMD for both CPU and GPU, we'll have to see just what I can afford at the time.
A lot of times the companies will drop the price info just before launch. There have been times when AMD changed pricing of parts literally hours before launch! THAT was wild!
@@rizqi-hiktakaewi678 AMD at the end of the day is still a buisness, and honestly if you need to save money, even the 5000 series really is far from unusable, and still plenty powerful for anything a normal gamer wants to do with their PC
Since 2011 all my CPU's were Intel, until last year when I switched to a 5900x. I think I'm all in on AMD now. Still stuck on Nvidia though, we'll see if AMD can change that in the next few years.
Disappointed that we're still stuck with 8-core CCD's. I'd love to upgrade to a single CCD 12-core x3d chip, but it's looking like we'll at least be waiting another couple years for that. I guess we'll have to wait and see if the eventual 9900x3d turns out better than the 7900x3d. And with all that said I don't think anyone is actually going to be unhappy with performance. Even my 5900x is still an excellent gaming CPU even if it's been well outpaced by the chips released since.
Yeah... and that means more 6-Core Ryzen 5 processors. This really sticks a fork in the whole "Intel being stuck on Quad-Cores" thing, that was funny back in 2018 but isn't funny now. Supposedly AMD did look at 16-core CCD's but they choose to go with 8-Core CCD's for performance reasons. IIRC, it was because the beefed up AVX2, AVX-VNNI, AVX-512, and AI capabilities take up so much chip space. A 10-core or 12-core CCD would have been nice though.
@@m8x425 it would also increase power constraints to a single ccd ccd size is **not** greater mean better it is a **tradeoff** and 8 core (16threads) is already plenty for most tasks it is a great tradeoff (coming from 4 in zen1/2+) it also means you are binning 8core ccds, putting good ones in single products for high number cpus you can't do all that well with a 16core ccd and a 16core ccd aside from caping power and thermal capabilities, it will not change much on computational workloads most games only use 6 THREADS at most (that is only half a ccd currently) I think most people forgot performance on consumer machines does not scale with core count the bigger the number of cores, the less it will scale chrome usually use between 4-8 threads, again that is at most spread accross a whole ccd, without interest in going beyond that and it's only at peak (for 1/10th a second)
PS: note that server cpu (epycs) are running much higher core counts, because these will have workloads that actually use the higher core count (like running thousands vitual machines on a single machine, yes thousands you read me right)
@@s-x5373 - Plenty of non-server workloads that use higher core counts as well. Aside from gaming, I do amateur astrophotography, and my main processing software would happily use 128 cores if I had them. As it stands, I have some tasks that take over an hour on my 5900x (and folks with larger images and higher image counts have jobs that last overnight).
In fairness to AMD's comments on corner cases, I think they're referring specifically to the issues Alder Lake had with AVX 512 support only being available on some cores and processes freaking out if they landed on an e core by mistake. Heterogenous cores inherently have some differences in performance characteristics and will have misses in terms of scheduling but those misses (should) fail gracefully by just resulting in a bit of a performance hit instead of an outright crash
@@infernaldaedra Sure, but there were additional, specific and much more substantial stability issues with the fact that only some cores on Alder Lake supported it at all, and that's the entire point being made there. Minor issues on a CPU with only one type of core are obviously irrelevant to a discussion about what type of heterogenous core strategy you want to use
@@bosstowndynamics5488 sure bit even Intel had issues the entire time they used AVX512 the fact AMD now can use it better than Intel Hardware from like 6 years ago is wild.
@@infernaldaedraavx-512 implementation with Skylake-x was significantly better than the current implementation where you need to disable e cores in bios to actually use it. The only issues Skylake-x had were power draw and additional heat when running avx-512 instructions. Though we can easily give Intel a pass here since they were the first to implement it and this was a pretty long time ago (CPU generation wise).
They definitely look like better CPUs and aligning with what people want, but saying they were going to be the same MSRP as 7000 prices is pretty much a LIE. 7000 was clearly overpriced at launch and pretty much immediately dropped by 20% when Intel released 13th Gen. Now they're capitalising on Intel's fall by actually upping the prices significantly over 7000's actual market MSRP for almost 2 years now. To add to that, the 9600x at projected 295 USD is an absolute joke when for barely any more (sometimes less) you can get a 7800X3D that outperforms the 9600x currently and will continue to do so exponentially because of the lack of 3D cache which is proven over previous generations to fall off more than X3D variants for gaming. As for productivity use, if you're using anything that can utilise 8 cores then that puts it about on par with the 9600X.. THIS CPU ISN'T FOR ANYONE!
considering the 7950x3d and others seem to have a reputation for that.. no reason to, both companies have some interesting failures, as I said a decade ago, intel is now on the losing side of the cpu battle, and it'll be another decade this time where amd will reign at the top for many people while intel struggles, its a back and forth that exists and in some ways i think its on purpose considering they're really the only two cpu manufactures that matter to the majority of people.
I'm still on a 3700X which is on a MSI B450 Tomahawk Max with a Ngreedia 2070 Super and I'm happy as a clam. Going on five years now (as of next January) on this platform and I have zero complaints. PS: running Linux Mint since the day I built this system. FK MS.
All the same, except i have MB Asus Tuf X570,,,man, over 4 years and it runs flawlessly from day one! Now I just got a combo deal with 7800x3d and I'll keep the rest from my old setup and see how it goes! I haven't installed it yet, still sitting in the new boxes...
How so the CCD’s have been the same since zen 2. They aren’t doing anything new just trying to keep manufacturers costs low as possible. Performance gains are not huge either.
@@ZackSNetwork There's more to cpu innovation than ccds. ... and how long has Intel been on 8-cores? I believe since 2018. Three qrtr the yrs they've kept us on 4 and 6-cores. I don't really mind the tick-tock approach. Push technology forward one gen, while they clean and tighten things up the next, with hardware tweaks to improve architecture, power and efficiency. Especially in a climate in which Intel could be facing a class action or large-scale recall. Even while arguably on top, AMD brings a good combination innovation and value compared to the market leaders.
@@ZackSNetworktheir CCD layout is the same yes, but the actual chip designs are fairly different each generation of zen. If they weren't, we would see no IPC increase and only clockspeed increases from smaller nodes.
There won’t be one. Their yields are too good, lol… awkward problem to have. And since they use the same dies in their server chips, intentionally crippling them just so they can sell Ryzen 3 chips makes no sense for them at all - they’d rather sell server chips with those they would otherwise cripple
If R5 becomes 8 core, then the 6 core becomes R3. Buy a R5 and call it a R3, done. And buy a R7 and call it a R5. Happy? Did it change your life in any meningfull way? And then the 12 core R9 would be R7 and the 16 core R9 remains the R9.
Pretty incredible they just confirmed not only Zen 6 support but also Zen 7 support on the same motherboards. If anyone had doubt before they should be diving head first now. I'll be waiting for the X3D model of course but I'm looking forwards to it.
The L3 cache bump on the single CCD parts is healthy, up from the already pretty good 32MB on the 7600X/7700X. Interesting that AMD don’t scale it linearly by core count either this or last generation. Proportionally the 7600X and 9600X have more cache per core than the 8 core parts do. Wonder how much this works to limit the 6c vs 8c performance difference in applications which utilise cache sufficiently
So, what’s that, like twice the cores, about 75% higher clock speed, about a 50% IPC improvement (so you could consider that another +50% effective clock rate, in a sense), twice the cache per core. Yeah, that’s a ridiculously good upgrade
@@mduckernzI know! I had done the math a couple weeks ago and it was like 2.5 times the single thread performance with clocks (@5ghz) and ipc, its going to significantly speed up my handbrake encoding for my 4k blurays. A year ago I had actually thought about getting a 5000 series as a stopgap and dropping it in my motherboard but decided against it as one of the top features im upgrading for is platform features like usb4, usb with power delivery through my case and 10G lan and more nvme storage at the top of the list
I'm loving the reduced temperatures of this new lineup. I'd imagine that's one of the reasons why the 9xxx series performs so will with higher wattages.
That has never been the case since we started exceeding JEDEC spec, and I wouldn't hold my breath on this one. You'll probably have better luck pestering the memory manufacturers to make two-stick kits that max out the platform. 96GB per stick would be nice.... But alas, if you want speed, low latency and capacity your current best option is something like Corsair Vengeance 2x48GB 6000CL30. That said, AMD is upping JEDEC spec from 5200 to 5600 so it's not a stupid loss in speed by populating all four slots.
I went from a 1600X to a 5800X3D in the same motherboard. AM4 has been fantastic. Huge credit for upgrading their AGESA to support new processors on old chipsets. That was huge, and very quiet.
@@redblack9925 where did you find that information ? They literally increased the specced memory speed support, the memory oc speed support, and made it able to be changed on the fly
@@s-x5373 Processor has the same number of PCI-E lanes and can't be increased on AM5 (there are no pins for it). X870E looks like it may have increased speed of some chipset lanes. Not sure.
I am still running on my AM4 platform since 2018 and going strong. I am very satisfied I did not need to change stuff when I upgraded CPUs. I very much like where AMD is going.
Seeing Intel issues and me after having bad experience with rushed AM5 and DDR5, i would recommend everybody to be REALLY cautious before buying a new platform.
Hopefully they can get that IPC up to 18-20%! I also think they should change the IPC comparison to 4.4-4.6GHz fixed instead of 4.0GHz to get a more realistic comparison.
finally all burnt 7000-series cpus replaced with burnt mobos, nicely done ;) 5700 GPU series black screen issues never got solved. Hip hip hooray AMD excellence indeed lmao. Not to mention the adjustments which were done for 3000-series Ryzens... :D Short is fanboy's memory :( Biden-syndrome?
The main thing I would love for socket longevity would be for the same CCX chiplets to be available with different interface chiplets to fit different sockets . So Zen 5 cores in an AM4 socket bottle necked by DDR4 RAM and older PCIe speeds, but with the interface chiplet maximizing bandwidth utilization so all busses run at peek speed to feed the cache from DRAM banks and PCIe memory in stuff like GPUs, SCSI controllers and high speed network cards . Similarly some Zen 4 or Zen 2 cores in the previous AMD socket that was originally used for Opteron and Athlon 64 chips . These old-socket parts would have an in-package bootstrap BIOS flash to make the chip boot in older machines by emulating the old CPUs until enough has run to do a safe transition to full speed of these "overdrive" style CPUs . Similarly Zen 10 cores from 2030 could be fitted with a compatibility chiplet for use in AM 5 boards with old 2023 BIOS code .
I have a 7950X3D myself and I cannot, for the love of me, figure out why people would want a dual V-Cache CCD SKU. As soon as you get cross-talk between the CCDs the gaming performance tanks. And that's regardless of whether both are vanilla or one is vanilla and the other is V-Cache. The games that need manual intervention are all games that spawn more than 8 threads if they see that more than 8 cores are available. Just forcibly locking them to the V-Cache CCD fixes the problem. Getting cross-talk between two V-Cache CCDs will not. I had a 5950X before my 7950X3D, and it was the same there: some games would run worse than on a 5700X, but locking the game to a single CCD instantly fixed it.
@ZackSNetwork it can be time consuming but it's fun. I love my 7950x3d but would definitely not recommend it to those who don't like to tinker. The performance and efficiency I get from mine is amazing
@@ZackSNetwork If you only care about gaming, get a 7800X3D. That said: With an XTX at 1440p I don't notice anything if I leave everything on defaults. I have to set resolutions and settings I would never run at to provoke the issue. But since I'm a software developer by trade I obviously cooked up a powershell script for Windows and a shell script for Linux that launches the game tied to the CCD it respons the best to, and edited my game shortcuts accordingly.
If you set the process affinity for steam to use only the one CCD then any games launched from it will use what steam uses. Edit: Under windows that is, no idea about Linux.
@@ZackSNetwork some x670e do have usb 4, as for wifi 7, pretty much pointless for basically everyone.. I mean the max speed difference is quite large and there's another channel at 320mhz but most places don't even go past the limit of 2.5 gpbs or maybe 10..
I hope they learned from the hurdles of the 7950x3d and 7900x3d. They should just make both ccd's have 3d caches. That would get rid of the scheduler hurdles.
Not completely, unfortunately. The two CCDs each have separate cache, and communication between them has to happen over a slower external bus. Having bigger cache on both CCDs doesn't help if the data that one thread needs is in the cache on the other CCD. So, for gaming workloads, it needs to lock all the threads to one CCD *anyways*.
Hey GN, Is there any chance to get the audio track from your episodes uploaded to Spotify? Just can't keep the screen off and UA-cam running on my phone without having to start another subscription. Thank you! 👍🏼🏆
How long had they been using 8 cores for their X700 and X800, 6 generations? At this point, the 9700X at 65w should've really been a 9600X. Guess both Nvidia and AMD are doing shrinkflation now and we probably won't see upgrade in cores for the mid-low range until the next generation of consoles shows up.😮💨
Can we expect better idle power consumption ? I came back to LGA1700 because the amount of wasted power in the infinity fabric is too high… 20W doing nothing… where Intel can do a lot with 20W.
@@libertyprime9307 not sure to understand. 7600x drinks 20W in idle. Which is ridiculously high and generates heat. I don’t care about cost, but I care about heat.
Some kind of video explaining what PCI lanes imply in daily usage would be interesting... Like... How does the choice of a chipset influence the expandability of a desktop PC? You also mentioned, I think, that motherboard manufacturers could add connectivity with additional chips or something... ... I'm a bit confused.
Better branch prediction is very cool, but it's a double-edges sword, as it might cause security bugs (see spectre, meltdown etc). It would be super interesting to hear from the engineers how they account for those problems.
This is rare for a gamernexus video but you said a lot of things however did not explain a lot of things. All the upgrades to the architecture I would like to understand their impacts, even just briefly. I'm a little disappointed you didn't go in a tad more detail. Thanks for the video though.
I read your comment as: you desired a statement from GN along the lines of, "What this means is that for gaming and for productivity is that it can execute x% more instructions in the same amount of time as previous generation, which _should_ lead to game / task running 10-15% faster" -> Thing is, until GN have the CPUs in their possession, any such statement is speculation -> You will probably get the statement(s) you require when GN gets hold of the new CPUs and does their own benchmarking. From my perspective, GN cut to the AMD engineers talking to these points, then to the marketing slides showing alleged improvements vs selected games and productivity. If it sounded a bit too technical, then fair enough ... GN then reiterated that the stats AMD put forward are cherry-picked stats from the manufacturer and require independent benchmarking when the products are available to validate AMDs claims on performance improvements.
Going full-on Team RED was one of the best decisions I've ever made. I build THREE 4K Ultra 60++ gaming rigs for the same money I would have spent on ONE Intel/nVidia rig.
Nah, this doesn't add up at all. Comparable Geforce cards are like $100 more than Radeon, and for Intel they should be similar price to Ryzen. Post the components.
@@libertyprime9307 except you are completly wrong about intel it's been the intel scam for the last 5 years, you are actually needing new motherboards etc.. and better cooling solution (and maybe even better psu, and that also applies to nvidia gpus) which both increase the cost significantly at start, but also increase extremely the cost throughout the life of the rig to be upgraded sometimes
A few nerds with deep pockets are salty about that, but the ones that make Threadripper sales actually turn profitable use OEM systems. My workplace included.
That's kind of because Threadripper existed the consumer/enthusiast market completely and in place CPU upgrades while maintaining the same motherboard aren't very common in workstation/enterprise settings, so the benefits of adding new features to the socket outweigh the loss of generation to generation compatibility. AM4 on the other hand loses a lot less because consumers are rarely pushing IO to the limit and are much more likely to do incremental upgrades. Plus, with 16 core AM4 parts the enthusiast market for Threadripper is very small (even in my case I would love to run a Threadripper system for my little home lab setup but 16 core AM4 is good enough and a lot cheaper)
two things: 1. threadripper started out for the enthusiast market, who will often upgrade even across one generation. 2. threadripper's second market, the high-end desktop workstation market, most of them are businesses who are wholly willing to throw out their motherboard and even their ram when they go to upgrade ($200 is almost a rounding error compared to the cost of everything else).
@@dead-claudia In the server space the in-place CPU upgrades are even more rare than in the workstations, yet AMD had made 3 generations of EPYC compatible with any SP4 motherboard - if the vendor provided the UEFI updates. Threadripper is a classic case of an engineering pet project getting taken over by marketing, with the usual follow-up of screw you towards both engineers and customers. And screwing over their highest-paying enthusiasts had definitely made a dent in the AMD's sales. Because not only those folks buy the high-end hardware, they convince others to look at the brand they like.
Saying Economy instead of Efficiency really isn't a big deal, IMO. We're all human and sometimes say the wrong words, and the people who are watching that AMD presentation are people who know that he meant Efficiency.
Watch the Extreme Overclocking of the R9 9950X: ua-cam.com/video/pYWtP4tZe30/v-deo.html
Watch our video with Wendell about the Intel problems: ua-cam.com/video/oAE4NWoyMZk/v-deo.html
cool
all that matters is that cinebench score. click, run, collect data. can we stop now. both companies have burned chips with this stupid behavior because of reviewers just benching.
Thanks. Looking to check out that Ryzen 9 in action. I've lucked out on 12th, after a lot of BIOS tuning have the 12700KF & 12900KF gaming stable. I really lucked out last year in not buying 13/14th and having degradation potentially. I hope everyone gets stability though those degraded chips it's likely lower voltages (and likely at lower ram speed) due to damage. Great piece on the Intel issues. I found it useful in the past couple days disabling the ecores has helped me. Mostly being gaming and file server stable are important in my daily workload.
careful Steve. Your inner nerd is showing.
Well that stinks that you are not getting new boards for launch. It would be nice to be able to give the most accurate performance metrics.
Huge opportunity to release a heavily-binned 9950X and call it the 9999X
Would be funny to hear Steve power through a review having to say "9999X" over and over lol
That name is most likely reserved for Threadripper.
Dual vchache 9950x3d would be cooler instead of the “1 die is full clock speed and the other is for 3d cache” scenario the 7950x3d was in
Good point to the poster above about a TR name for that. And no, please do not make me say 9999X 57 times in a review! hahaha
@@GamersNexus opportunity for you to count all of the 9s!
Fun fact: the collective noun for a group of Steves is a "nexus" of Steves
I believe this usage was officially established as part of the Geneva Conventions
@@supdawg7811 *Geneva suggestions* - fixed that for ya :)
I imagine Steve Collective from how I met your mother😂
Lolol
but in Ozzyland it is called a hub of steves
17:02 I'd gladly accept a platform strategy of "the socket lasts as long as the RAM generation".
No point in changing it earlier
I hope they change cooler mounting for AM6. IMO AM5 ahould've not stayed compatible
/me shows benjamin AM AM2 AM2+ AM3 AM3+ and AM4 ;) pitty of the Thread Ripper sockets :(
It is the most reasonable option for consumers as well, to be honest.
Big like and it makes sense if you are building to specs (like PCIe and DDR gens), which provides you max theorietical numbers. Engineering is better with well defined boundaries.
@@Damicske Socket am2+ cpus could be used in most am2 boards, as long as the mfg did a bios release for it. In addition, AM2+ brought a much improved hypertransport, cpus were moving WAY faster back then and things got real limiting real fast. Socket AM3 brought ddr3, lots of AM2 and AM2+ boards supported AM3 cpus. AM3+ had a new pin, so bulldozer cpus could not be used on AM3 or AM2 boards, they lacked the memory controller compatibility for ddr2, in addition the older boards could not supply enough power, so it really wouldn't have worked anyways.
They have always been trying to keep sockets around, but 20 years ago a cpu didn't gain 15% over 2 years.
We were seeing 100% gains, going from 1 core to 2, increasing clocks every gen, then 2 cores to 4 with AM3.
@@GewelReal I don't agree. It's bad for the environment . Many AM4 users will be forced to throw away their existing cooler which are perfectly good. Not every company is like Noctua that provides mounting for free.
“AMD X870E. The E stands for extreme. The X also probably stands for extreme” 😂
Well, at least they didn’t add an Ai in there too…
All new Extreme 870 Extreme chipset!
No, the "X" stands for "eXpensive"
not yet
Now there are multiple AMD staff kicking themselves for not using Ai somewhere in the name.
You forgot about the x670e-e
Great boards, but some of the stuff from 870 looks real tempting, like usb 4 support. We need 10gb and multiple USB4 ports on all boards, we've been stuck at 1gb and 2.5gb networking for WAY too long.
Being able to add full 512bit AVX without the downsides of well, literally everything, is an amazing engineering feat.
Please explain 512 bit AVX implementation pls.
@@michaelmoses8745 like the whole spec in a UA-cam comment? It generally requires double sized registers of a specific number, wider data lines and much much more transistor space. This normally results in limited clocks as the silicon is tuned for one workload type at 64 bits. You can see the full specification on Intel's developer site or Wikipedia or chips and cheese.
I thought Zen 4 already fully supported AVX 512 though, is there some change in Zen 5 to this?
@@JosiahBradley
Thanks.
@@bosstowndynamics5488 It was explained in the video. Zen 4 is still a 256 bit AVX arch and it's double pumped to get 512 ISA working for a mix of performance and efficiency at a loss of raw throughput. Zen5 has a true 512 bit data path when needed with the double pumped pipeline available for presumably less performance centric workloads.
Refreshing to see someone on stage that acts normal about new tech from a company.
all new 3080tai!
@@gooseloose682 It should be a hate crime to call TI as Tie
Introducing a brand new PCIe 3.0 motherboard in mid-2024 is wild. Kinda slimy to call it B840 when it's very clear that it should be called A830 or something.
It's an anti-consumer move that only benefits motherboard manufacturers and cheapens AMD's platform reputation.
Misguided decision from whichever AMD exec allowed this.
You're generous, it should be A810 because even A620 has PCIe 4.0
Yea.. @GamersNexus Please call AMD out for being this cheapskate. PCIe 3.0 was introduced during 3rd gen Intel Ivy Bridge 12 YEARS AGO !!!!
@@fleurdewin7958 Actually Sandy Bridge-E brought PCIe 3.0 13 years to the HEDT platform.
I had an i7-3820 on an X79, and "upgraded" to a B450 in 2019.
I was shocked that my second NVMe SSD was slower on the B450, and I learned that the chipset provided PCIe 2.0 lanes.
But yes, B840 is atrocious and malicious and is deliberately meant to confuse buyers. Just like the Zen 2 Ryzen 7000 chips.
It's especially wild considering they THEMSELVES released multiple PCIe 4.0 x8 GPUs that perform like crap on PCIe 3.0...
Economy core... what a great engineer burn.
I understood "a kind of a p core"
Yeah economy it is..cheap way for realy large margins..
05:54 I love how he completely ignored the "AI acceleration" bs 😆
AMD said something to effect of "AI" when they spoke about branch prediction in Zen 2.
To be honest it does qualify as AI simply because its job is "decision making" (a thermostat qualifies as that too), but it doesn't qualify as machine learning / deep learning / generative AI.
@@CossackHD to be fair, he was speaking of it in context of vector instructions like AVX. Which are pretty useful to do matrix/tensor math, which is heavily used when doing actual AI task. Meaning AI as in: doing training or inference with a neural network. So as AI talk goes, this was actually on the lower side of the marketing BS scale.
So it does have the AI snitch built in or not?
@@Hugh_I Agreed. AI is over-hyped, but is still a real thing. We can see "AI" in things like realtime background removal tools. It's just not the cool "AI" that people think of.
they just simplify every gaming tech term with 'ai', physics ragdoll weather particle simulation etc
The B840 chipset name seems misleading, as the feature set in some spots places below the A620 chipset.
It's an AM5 socketed B450 so far as I'm concerned.
Agreed
@@vedinthorntheres no overclocking, from that alone we can tell its probably very different to b450
@@stephenallen4635 fair enough, but I'll gladly give up over clocking for PCIE 4.0. over clocking is useless unless that's your bag, but PCIE 4.0 is good for everyone.
@@stephenallen4635 well you can't overclock a cpu of a type you were not designed for, B450 is am4
so this kinda makes sense still
Their follow-through for the AM4 platform has truly been astonishing. I do hope such a commitment can be kept similarly with AM5
Agreed. It is a big selling point for DIYers.
Not one reviewer has dropped an “Its over 9000” reference.
The balls are inert.
Shows they know to stay classy.
@@PwntifexMaximus So not using references is classy? Wow, what pretentious train did you get off from?
Ikr the disappointment is real
@@ark_knight
The one being run on your mom.
For my entire life prior to AM5 every CPU upgrade was an entire new build, and forget the component costs it takes an entire weekend to strip out the motherboard and re-manage all the cables.
Upgrading to Zen5 this fall is going to feel sooo good, four screws, plop in the chip, done.
So true! It is not just the chip cost. The time, money (and frustration at times) saved is often understated.
We used to be able to install a different socket on the same motherboard.
who the fu** is going from Zen4 to Zen5, i get going from Zen2 to Zen3X3D, wait for Zen6 dawg.
I went from a 1600AF to a 5800X3D, now i'm waiting for Zen5 and Arrow Lake to launch so i can make my choice.
Honestly I see no reason to go from 7000 series to 9000 series, I'd wait Zen 6 on AM5 and the x3d part if you want the best gaming experience.
@@MaxIronsThird could be someone going from an entry level 7000 series cpu like the 8500g to a higher end 9000 series cpu. Could have been waiting it out like by purchasing something to get by a month before the announcement, or could be cause they want to invest more in a PC for various reasons
I’m only interested in how much the prices for the Ryzen 7000 series will drop for CPUs and 600 series motherboards. Cheers from Australia!
Some places already have some great discounts on x670e boards. Also, from Australia.
The 5800 X3D is the iceberg of value that won't go away. It keeps everything up.
The new chips will be launching on the 600 series boards initially, so I expect demand for them will go up in the short term, not down. If history is a guide, after the release of a new architecture, AMD will offer great Black Friday/Cyber Monday deals on the last gen stuff
@@stevenwest1494I think I’ll be upgrading from the 5800x3d once the AM5 equivalent comes out but we’ll see.
Yep. I am looking to jump from am4 to am5. 7700x then eventually upgrade to 9800x3d or 11800x3d or whatever maxes our the am5 platform.
For my audio production usecase i'm most excited about doubling the instruction throughput since most digital audio workstations rely directly on AVX2 instructions so this should, at least in theory, make them incredible for low latency audio processing. These seem like they're going to be absolute monsters for creative work.
I suspect you will also need the cache
also the cpu will likely be resting at avx2
avx512 is much much more intensive
zen3 and zen4 already had full support pipeline for avx2
dont get me wrong, it will be faster, but not that much faster
@@s-x5373 cache size is really only crucial for audio production when using sample based instruments like Kontakt or Omnisphere, in those scenarios the x3d parts do better than anything else but they actually perform worse than the non-x3d siblings in everything else like overall project size without buffer underruns and with digital signal processing. This is where IPC and clock speed combined are extremely important. CPU's are certainly not resting while executing the AVX2 instructions from the audio engines, idk where you're getting that. And as far as I'm aware to date theres very very few audio production software that makes use of AVX512. Modern digital audio workstations can be incredibly demanding of resources, in large projects I can easily engage over 75% of my Ryzen 5 3600's 12 threads, even more if I push it. Either way, If they actually doubled the instruction throughput then these chips will be a seismic jump for this type of work, of course that remains to be seen.
AVX is SIMD, single instruction, multiple data, not really related to IPC throughput as it's about reducing the instruction pressure via compiler tricks, Early Intel CPUs actually throttled down and caused huge performance issues for other code under AVX 512 workloads.
The real issue is Windows (and generally any preemptive multitasking OS) is a shit OS for realtime work, and interrupt based CPU designs like x86, multitasking systems absolutely screw up your latencies as the OS will pause and switch out different apps, hardware will force CPU to pause computations via interrupts, your shitty closed source OEM drivers will hog 1ms of CPU and ruin your latencies.
@@s-x5373cache is really only important for using lots of sample based instruments like Kontakt, in every other scenario (general project size without buffer underruns and digital signal processing) the non-x3d outperform the x3d variants so the bottleneck with audio engines is IPC/clock speed, which at least on paper should be a big increase. We'll have to see when they come out but I've got a feeling these are going to be a massive uplift compared to Intel chips for this.
Where do you feel like you're being held back right now? Recording on PC is not like high end gaming where frames win games. The latency was low enough in the mid 20's when I was doing it.
Didn't understand a word of that, but great work Steve.
😂 same here
Watching these guys as someone relatively hardware-clueless is just "I like your funny words, magic man"
i came to laugh my A-off
About 15:35 where you mentioned that AMD and Intel architectures both have challenges when it comes to scheduling, I think that you missed the point Macri was trying to make. There is a massive difference between misscheduling on a 7900x3d and a 14900k. In the first case, you might schedule a clock-friendly task on a high-cache core or vice versa and lose a bit of performance. In the second case, if you schedule an AVX workload on an efficiency core, you get a kernel panic / BSOD. This means you can run older OSes on an x3d CPU. It might not be as good as it can be, but as a user, it will work, and as a dev, you don't have to backport complicated scheduling algorithms to older kernels still in long term support. On the hand, you absolutely have to do this for intel CPUs. They need additional dev time, and that needs to happen before release; and potentially in secret. This is why the initial launch of p/e cores was such a mess. Some apps would randomly crash on Windows, which got support early, and Linux simply didn't work for weeks because it would've leaked the IP.
So, AMD doesn't _require_ special drivers, but they do benefit from them, and drivers are the work of the hardware provider putting the onus of producing and maintaining this code on AMD. Intel requires a special scheduler, which is the responsibility of every OS provider and an additional burden on the communit in the case of Linux.
We didn't miss the point. We just thought the language was just as readily turned on them regarding the "confounding" comment.
Linux has long had support for heterogeneous compute in the scheduler, because it's important in the HPC world. So both half-X3D and Intel P/E CPUs are directly supported by the kernel. What's weird to me is that neither manufacturer has lent on M$ to get the same support implemented in Windows.
@@GamersNexus Make sense. Though the way you word it makes it feels a bit like misrepresentation because of that. I totally get where you're going though. For a lot of people, if you buy a product and expect it to work as advertised, in the end, there is no difference between a crash and a 20% performance loss. They're both bad situations for consumers, and it's easy to think how less technical people could be convinced by claims like this one.
@@sykescalvin09 Microsoft admited in a marketing way of talking about it, that they already fully knew what they have to do to better manage it.
But they are more focusing on building a IP bullshit out of their bad ripoff of existing solutions than pre-emptively testing and developping this ready.
And if they had, they would have had pushed that locked on microsoft server licenses first.
@@sykescalvin09 Well you said it yourself: it's important in the HPC world. Windows doesn't really show up over there so it simply wasn't a priority to them. You could even say it's a strategic advantage to wait as it forces people to upgrade to newer version of Windows if they want to use those newer chips.
Just to clarify, even if linux did support heterogeneous compute for a while, it still requires every specific core's capability to be declared to the scheduler for it to work, and Intel's P/E required custom code for it to work. Those changes couldn't be published before the release embargo lifted which delayed P/E support for several weeks.
Strikes back against who? Themselves?!
big oooooooof for Intel!
yep
*whom ;)
@@IvoPavlik American English rarely uses whom, and never ever uses whilst.
@@harleyn3089 Whilst that may be true, Steve is a gentleman of distinguished style to whom linguistic prowess and flair might be of some importance.
I just want a 12 core single CCD X3D CPU :(
Me too..
You'll probably have to wait for Zen6 for that. Doesn't look like AMD is going to increase the core counts this generation at all.
@@garethgobulcoque8668I hope so
I feel like if anything it would double to 16 but who knows. A 12 core R7 10700X or whatever would be awesome, with a 24 core top end R9. The 16 cores for the #950X is starting to get old.
But at the same time, they're getting so fast that in most cases a consumer won't need more from Ryzen 9. My 7800X3D barely runs slower than my 5900X in multi-threaded apps, and in games it's miles faster.
@@mttrashcan-bg1ro There's only one single problem: bandwidth starvation.
More cores are not helpful, if they have to wait for data from RAM.
I think it is nice that the AM5 naming happens to line up with DDR 5 and PCIE 5
Instead of comparing benchmarks vs the 14900k, they could have just said, 'Our chip doesn't melt itself' and just drop the mic.
wake me when x3d comes out
Oh yeah haha. My i7 9700k and co are roasting to death while I wait to replace it with an AMD 9xxx x3d chip :)
So many frames in VR!
Indeed, Should be sometime in September.
At 4k Ultra/High you will be limited by the GPU
It will make no difference if you run
5800x3d
7800x3d
or
9800x3d
They will all perform exactly the same.
But if you're wanting to run super high refresh rates at 1080 on low, then you might have a point.
Who knows, that might be your bag.
But if you have any of the previous x3d chips, run your game at 4k ultra/high, you would be far better off staying with those and using the money to buy the 5090 instead.
You're going to need it.
@@williamtopping Yeah, but it will help at 2k and Ultra-wide.
All I want is the 9800X3D as I found the first gen of AM5 to be a disappointment.
I built an AM5 PC somewhat recently and am happy AMD seems to be doing well, especially since recent intel xx900 seem to have serious problems.
It's good that there's more than one mainstream CPU manufacturer, im case one messes up.
19:20 Although I wouldn't be excited to see AMD lose CPU longevity, that would be an AMAZING episode 10 years from now to see this video dug up and some great content made out of it!
I'll be getting a 9950x most likely. But that fact we will have to wait a month or 2 for x870e motherboards is ridiculous
by that time youll get x3d too likely
@s-x5373 I mean, That'd be cool too I guess. But really the new motherboards should come out at the same time the new CPUs do
@@s-x5373 The past couple X3D chips didn't release for like 6 months after the initial set of chips, so maybe more than "a couple months"
I'm really eager to upgrade as well, but it seems that waiting is just the smart choice for me here, as much as it sucks. Hopefully, they'll release the X3D chips together with the new motherboards so we can get a good comparison and make the most informed choice possible when upgrading to this gen.
@123Suffering456 That'd be nice. But like others have said. X3D hasn't been released for like half a year after thr regular versions. X870E boards are rumored to be out in September
it seems so easy to have multiple x16 slots with how amazing these processors have become but yet most brands are only putting one on the boards. i may have to keep my x570 boards forever.
They don't do it because SLI and Crossfire are dead (as in, no physical connectors to bridge GPUs together anymore). Without that, there's absolutely no point in more than one x16 slot for practically anyone buying a consumer desktop level board. All it would do it is add cost for something no one can even use anymore.
@@wasd____ graphics cards are not the only thing using the x16 slot. Personally I use 1 for a gpu and the other for a 4 slot nvme drive. They will have the slot but instead of two x8 it should be two x16.
@@boosted94sc300 I mean okay but a 4 slot NVMe drive is still a super niche thing for a consumer desktop system that almost no one is going to use. If you need _that_ kind of crazy bandwidth for storage you're probably doing something well beyond the consumer level and really just need to be looking at a higher-class motherboard to begin with.
The CPU and socket is explicitly designed with a specific set of PCIe lanes, an extra 12 lanes is an extra expense, and explicitly why ThreadRipper exists.
@@wasd____ I bought a consumer motherboard, the ASUS B550-XE STRIX because it came with a PCIe 4.0 4 slot NVMe riser, had 3x X16 slots, and supports a total of 6 NVMe SSDs out the box.
AM5 is just expensive for being expensive. You get garbage overpriced motherboards. Then people think a "$150 board is $150 quality".
Good video, as always.
One quick note, regarding the nice catch on the CCD discrepancy at 10:40 or so: while I totally agree that, ideally, one should try to eliminate as many variables as possible, however minor they may be, when doing these sorts of marketing demonstrations--especially given the fact that I can't imagine AMD themselves could not find more comparable parts on hand--in this particular case, if I were to guess, I would assume this particular instance is actually a less-than-ideal case for AMD.
It would be reasonable to guess that, assuming all other attributes are more or less equal (e.g., boost curves, cache, thermals, whatever), that the chip with fewer CCDs (the 7700X, in this example) would have an advantage, albeit miniscule, over the other, since there should be less room for core-to-core latency and other things to affect the benchmarks.
Of course, again, sure, it would be better to compare more equal SKUs, but out of all the marketing bullshit, this one seems tames, if not downright not in their favor on paper.
---
Longwinded, and tangentially related: I wanted to email you all about latency testing, but never quite got around to it; have you guys considered looking into it for motherboard and CPU reviews?
For the overwhelming majority of workloads, I recognize it is typically not something noticeable, but the industry trend seems to have been shifting from complex to manufacture and perhaps more latency-prioritized designs to higher-yield, throughput-oriented designs.
We can see this across virtually all SKUs that I am aware of:
- AMD's quirky Bulldozer shared FPU/SIMD design.
- AMD's CCD/chiplet design in most modern AM4/AM5 (as well as Epyc, Threadripper, etc) processors.
- AMD's "heterogeous" 3D-vcache weirdness (e.g., 7950X3D having one CCD with and one without)
- Intel's various multi-tile processors.
- Intel's heterogenous P vs E-core design.
- Intel's switch from the latency-oriented ring bus to throughput-oriented mesh interconnects (IIRC, after the 10980XE; unsure about standard desktop platforms).
Similarly, interrupt-based I/O, such as PS/2 ports are seemingly dead [1], with the last board I am aware of being the XOC X670E Gene -- and even then, I'm not actually sure those interrupts are "guaranteed" nowadays in the way they were in the past.
Motherboards, as well, tend to pack in all sorts of bells and whistles, making it increasingly difficult to find something barebones with a decent VRM and PCI-E layout, without a variety of crazy onboard -- unfortnately, often not great -- LAN, audio, and various other miscellaneous chips etc, all of which the user has little to no auditability or control over, except perhaps a UEFI toggle that *cross your fingers* truly disables these things -- all of which can add up to be fairly deterimental, and that's before we talk about the actual, often pathologically behaving, vendor firmware/BIOS.
The above may seem like a bit of a schizo-placebo rant, but for things like real-time audio, piano synthesizers, and so on, it can make a surprising difference.
To the best of my knowledge, Anandtech was one of the few (only?) outlets in recent times that would routinely measure things like DPC latency as part of their motherboard reviews. However, unfortunately, their articles are few and far between.
It would be very interesting to me if you guys found the time to integrate some of that into your testing methodology. And I would be very curious whether things like the relatively dogged-on (rightfully so) 8700G may turn out to have some secret advantages for certain workloads, being one of the few monolithic processors remaining -- similarly, I wish more reviewers checked out the w7-2495X during the SPR WS launch rather than jumping straight to the top-end 56-core model, as the
about that long comment, worth noting even programmers have generally switched from optimizing for latency to optimizing for throughput.
there are cases where latency is critical, but in most of those, you're not only optimizing code, but also the network around it.
Interesting choice to compare 9950x vs 7700x in first party benchmarks
think itll be like intel gen 2 vs 4
I love when GN makes CPU content like this. I like fans and all but I nerd out the CPU architecture information.
Welp, it seems making the switch from Intel architecture to AMD for the very first time since ye olde days of AMD Athlon was the right call for me.
Not only is the 7800X3D an amazingly powerful AND power-efficient CPU for Gaming, I now also already own a motherboard with a seemingly still very high end chipset (X670E) and AM5 architecture that'll allow me to just put in a new CPU when necessary for years to come.
Steve is absolutely right in saying that AMD also isn't without its darker aspects, but this is truly one of the best pro-consumer decisions a company of AMD's size has made for quite a long while, especially in the gaming segment.
I've been gaming for over 30 years now, both on PC and console since the early 90s, and I've always fondly remembered the times when you bought a decent system for gaming and it was still decent 3 years later... seems like with AM5, these times might, at least for now, be kinda back again.
AMD has been a valid choice for gaming (compared to Intel) since Zen2.
@@Tigrou7777 I was quite happy with my Core i7 8700K before, easily got it to stable 5.0 GHz for 4 years running. ;)
Had an Athlon as well back then, until AGP graphics were phased out and PCIe + Nehalem took over, 7800x3D on AM5 now as long as it will take me, here we go!
When Steve rattled off the TDP equation… I feel in love in a big nerd way! That’s awesome!
Paper towel getting real close to a replacement now, Steve.
Jajajajaja and I thought was the only one that noticed.
I'm so glad they are giving us more lanes, but I cant wait for when you can get 100 lanes on desktop. I don't want to switch to threadripper, way to expensive.
Wait, Thermalright sponsor? That's a new one. Well deserved imo
What I'm hoping for most from Zen 5 is if the 9600X will continue the legacy of the Ryzen 5s being top-tier performers in the gaming realm! If they can, I might have my CPU pick for my next build!
What do you mean by top tier performers? The Ryzen 5’s have become their low end chips. Ryzen 7 is mid range and 9 is high end. I would not recommend a 6 core processor in 2024 going into 2025.
We're at a point where it doesn't make sense to upgrade to a new low-midrange system. If you already have something that works, save until you can afford a high-end system. At one point, upgrading from an i5 to the next i5 was a major improvement. Now it almost doesn't matter.
@@ZackSNetworkFor gaming AMD's Ryzen 5s are generally the best bang for buck. Yes they are the lowest in the stack now, But their gaming performance is usually extremely close to the 8 core chips and the higher end chips are basically irrelevant for gaming.
you could wait for a 9800x3d then, will likely be the most gaming performance for bucks and efficiency
6 core in 2024 will last you 1 year
I love the way y’all stamp clips being used for quotes with an actual profile of the speaker(s), so people are always credited and viewers get a basic idea of credibility on certain topics.
Watching this video from my FX 8350 pile drive Frankenstein with a 6700XT running on 16 lanes of PCIe 2.0
I switched to a 3600X not long after its release, what's keeping you?
@@NielsC68 He was born to piledrive.
@NielsC68 I'm a renegade enjoying life on the edge. I also like teleporting, skipping, and hitching my way through CPU intensive games with glorious 1% lows that looks like a heart monitor after I blasted coke and ephedrine.
It delights me to witness frame rates hop and skip from 28 fps all the way up to 108, only to Plumit again like the doomed aspirations of a minimal wage employee who spent his miniscule savings on a rug pulled crypto alt coin.
@@Pruflas-Watts You should be a writer, not sure how much it'll help but that was glorious.
Good luck. 👍
FX 6300 and (of all things) a NVIDIA GTX760 here.
16GiB of RAM, doubled from the original 8 I built it with.
Data stored on a 1TB piece of 'spinning rust' that luckily seems to still be going perfectly strong -- I've heard that modern consumer-grade hard drives do not enjoy the same long-term reliability, and that NVME Flash memory (at least, models with large endurance guarantees) all around is the way to go when building a new machine.
Running Linux Mint 21.3 Cinnamon, because I'm still a boring person, but have been unwilling to deal with Microsoft's BS since Win8 came out, and pretty much everything I'm interested in playing runs through Wine these days anyways, given my hard-line anti-DRM positions.
It was my first scratch-built PC, and I'm really happy with all the use I got out of it, both when I initially built it with Win7, when I dual-booted it, and after I got rid of the Windows partition.
But suffice it to say, I'm... here watching PC part review channels, in preparation for a new build sometime in the relatively near future (maybe after the X3Ds of this series come out?). Case doesn't have a space LONG enough for modern graphics cards, so this'll be a pretty much completely new build. Probably AM5, planning on going AMD for both CPU and GPU, we'll have to see just what I can afford at the time.
Release in 2 weeks without price info is wild
A lot of times the companies will drop the price info just before launch. There have been times when AMD changed pricing of parts literally hours before launch! THAT was wild!
@@GamersNexusthis will be priced as high as possible, according how bad Intel is currently.
Not really; it’s pretty common practice.
@@rizqi-hiktakaewi678 AMD at the end of the day is still a buisness, and honestly if you need to save money, even the 5000 series really is far from unusable, and still plenty powerful for anything a normal gamer wants to do with their PC
😆 yoooo the stack of 100s on the B roll! 👀
Since 2011 all my CPU's were Intel, until last year when I switched to a 5900x. I think I'm all in on AMD now. Still stuck on Nvidia though, we'll see if AMD can change that in the next few years.
AMD is the only company that does Tech presentations without treating its audience like idiots
Disappointed that we're still stuck with 8-core CCD's. I'd love to upgrade to a single CCD 12-core x3d chip, but it's looking like we'll at least be waiting another couple years for that. I guess we'll have to wait and see if the eventual 9900x3d turns out better than the 7900x3d.
And with all that said I don't think anyone is actually going to be unhappy with performance. Even my 5900x is still an excellent gaming CPU even if it's been well outpaced by the chips released since.
Yeah... and that means more 6-Core Ryzen 5 processors. This really sticks a fork in the whole "Intel being stuck on Quad-Cores" thing, that was funny back in 2018 but isn't funny now.
Supposedly AMD did look at 16-core CCD's but they choose to go with 8-Core CCD's for performance reasons. IIRC, it was because the beefed up AVX2, AVX-VNNI, AVX-512, and AI capabilities take up so much chip space. A 10-core or 12-core CCD would have been nice though.
@@m8x425 it would also increase power constraints to a single ccd
ccd size is **not** greater mean better
it is a **tradeoff**
and 8 core (16threads) is already plenty for most tasks
it is a great tradeoff (coming from 4 in zen1/2+)
it also means you are binning 8core ccds, putting good ones in single products for high number cpus
you can't do all that well with a 16core ccd
and a 16core ccd aside from caping power and thermal capabilities, it will not change much on computational workloads
most games only use 6 THREADS at most (that is only half a ccd currently)
I think most people forgot performance on consumer machines does not scale with core count
the bigger the number of cores, the less it will scale
chrome usually use between 4-8 threads, again that is at most spread accross a whole ccd, without interest in going beyond that
and it's only at peak (for 1/10th a second)
PS: note that server cpu (epycs) are running much higher core counts, because these will have workloads that actually use the higher core count (like running thousands vitual machines on a single machine, yes thousands you read me right)
@@s-x5373 - Plenty of non-server workloads that use higher core counts as well. Aside from gaming, I do amateur astrophotography, and my main processing software would happily use 128 cores if I had them. As it stands, I have some tasks that take over an hour on my 5900x (and folks with larger images and higher image counts have jobs that last overnight).
If the leaks are true then ZEN 6 should be getting 16c/32t per CCD... And ZEN 6 3D is going to be a no-brainer for everyone! 🥰
Still running the 5800x3D. These design changes sounds great.
Pre-ordered my shirt. Keep doing you guys. Loved the Asus confrontation video!
In fairness to AMD's comments on corner cases, I think they're referring specifically to the issues Alder Lake had with AVX 512 support only being available on some cores and processes freaking out if they landed on an e core by mistake. Heterogenous cores inherently have some differences in performance characteristics and will have misses in terms of scheduling but those misses (should) fail gracefully by just resulting in a bit of a performance hit instead of an outright crash
AVX 512 had issues before that in Haswell E and Skylake it caused memory degradation when used with XMP and using a gaming motherboard with OC preset.
@@infernaldaedra Sure, but there were additional, specific and much more substantial stability issues with the fact that only some cores on Alder Lake supported it at all, and that's the entire point being made there. Minor issues on a CPU with only one type of core are obviously irrelevant to a discussion about what type of heterogenous core strategy you want to use
@@bosstowndynamics5488 sure bit even Intel had issues the entire time they used AVX512 the fact AMD now can use it better than Intel Hardware from like 6 years ago is wild.
@@infernaldaedra Ok, but again, that's completely irrelevant to the discussion about heterogenous core architecture which is what I was commenting on
@@infernaldaedraavx-512 implementation with Skylake-x was significantly better than the current implementation where you need to disable e cores in bios to actually use it. The only issues Skylake-x had were power draw and additional heat when running avx-512 instructions. Though we can easily give Intel a pass here since they were the first to implement it and this was a pretty long time ago (CPU generation wise).
Steve adds LN2 to his coffee.
How else will the ice coffee stay cold
1.8bln more transistors should be interesting
The day this goes on sale, AMD needs to have Dr. Evil on stage saying, "We added 1.8 _billion_ more transistors!"
One million dollar!
@@stefanschuchardt5734 i read this in Gru voice .. (while minions cheers)
Probs mainly the new full avx512 path
They definitely look like better CPUs and aligning with what people want, but saying they were going to be the same MSRP as 7000 prices is pretty much a LIE. 7000 was clearly overpriced at launch and pretty much immediately dropped by 20% when Intel released 13th Gen. Now they're capitalising on Intel's fall by actually upping the prices significantly over 7000's actual market MSRP for almost 2 years now.
To add to that, the 9600x at projected 295 USD is an absolute joke when for barely any more (sometimes less) you can get a 7800X3D that outperforms the 9600x currently and will continue to do so exponentially because of the lack of 3D cache which is proven over previous generations to fall off more than X3D variants for gaming. As for productivity use, if you're using anything that can utilise 8 cores then that puts it about on par with the 9600X.. THIS CPU ISN'T FOR ANYONE!
i recently bought AM4 5950x... its old...but its gold for me... it enabled me to create and render videos....so i am good...
I'm surprised they didn't throw shade on Intels high end cpus for deteriorating in 6 months.
They don't need to. They're letting intel's silence speak
considering the 7950x3d and others seem to have a reputation for that.. no reason to, both companies have some interesting failures, as I said a decade ago, intel is now on the losing side of the cpu battle, and it'll be another decade this time where amd will reign at the top for many people while intel struggles, its a back and forth that exists and in some ways i think its on purpose considering they're really the only two cpu manufactures that matter to the majority of people.
@@yulfine1688 I hope the 7950x doesn't have the same issue. I just ordered one to replace my 13900k because I'm on my 2nd one and it's dead 😭
and a month later AMD could have a fiasco. These mega corporations are much more friendly than most peasants are to their neighbors.
@@yulfine1688another generation and intel will came back at the top in productivity
It's nice to be able to put my feet up with a Zen 4 system and know there's at least one, probably two more generations coming for AM5.
I'm still on a 3700X which is on a MSI B450 Tomahawk Max with a Ngreedia 2070 Super and I'm happy as a clam. Going on five years now (as of next January) on this platform and I have zero complaints. PS: running Linux Mint since the day I built this system. FK MS.
All the same, except i have MB Asus Tuf X570,,,man, over 4 years and it runs flawlessly from day one! Now I just got a combo deal with 7800x3d and I'll keep the rest from my old setup and see how it goes!
I haven't installed it yet, still sitting in the new boxes...
Mint Linux is a great operating system.
I'm 83% certain AMD employ literal wizards these days. Pointy hats n' everything.
At least some priests. A few wololo and now most handhelds are team red.
How so the CCD’s have been the same since zen 2. They aren’t doing anything new just trying to keep manufacturers costs low as possible. Performance gains are not huge either.
thinking lightning in a rock, pure witchcraft
@@ZackSNetwork There's more to cpu innovation than ccds. ... and how long has Intel been on 8-cores? I believe since 2018. Three qrtr the yrs they've kept us on 4 and 6-cores.
I don't really mind the tick-tock approach. Push technology forward one gen, while they clean and tighten things up the next, with hardware tweaks to improve architecture, power and efficiency. Especially in a climate in which Intel could be facing a class action or large-scale recall. Even while arguably on top, AMD brings a good combination innovation and value compared to the market leaders.
@@ZackSNetworktheir CCD layout is the same yes, but the actual chip designs are fairly different each generation of zen. If they weren't, we would see no IPC increase and only clockspeed increases from smaller nodes.
Just nuts to me that the B840 is pcie gen3 when A620 is Gen4.
18:33 - love the old AMD and Intel CPUs as coasters for the current gen! :)
Hoping Zen 5 is as good as claimed and patiently waiting for Shimada Peak next year.
AMD is my choice if I need to update my x86 desktop. I have no confidence in Intel with their current problems.
Intel began to produce buggy CPUs... unbelievable really. their QA sucks badly
man when will ryzen 3's come out. and when will r5 be the 8 core
looks like AMD ditched the 4 core 8 thread CPU's for good.
Never? Lmao
XD when there is a 16 core CCD
There won’t be one. Their yields are too good, lol… awkward problem to have. And since they use the same dies in their server chips, intentionally crippling them just so they can sell Ryzen 3 chips makes no sense for them at all - they’d rather sell server chips with those they would otherwise cripple
If R5 becomes 8 core, then the 6 core becomes R3. Buy a R5 and call it a R3, done. And buy a R7 and call it a R5. Happy? Did it change your life in any meningfull way?
And then the 12 core R9 would be R7 and the 16 core R9 remains the R9.
Pretty incredible they just confirmed not only Zen 6 support but also Zen 7 support on the same motherboards.
If anyone had doubt before they should be diving head first now.
I'll be waiting for the X3D model of course but I'm looking forwards to it.
5:53 I see what you did there
yeah bro wtf you missed out the most exciting part of the new architecture!!!one!!!!
That 9700X is looking like it's right up my alley
I want a pen holder made of RAM sticks
Seems straight forward, if you have non working sticks.
get some old ram and glue
The L3 cache bump on the single CCD parts is healthy, up from the already pretty good 32MB on the 7600X/7700X. Interesting that AMD don’t scale it linearly by core count either this or last generation. Proportionally the 7600X and 9600X have more cache per core than the 8 core parts do. Wonder how much this works to limit the 6c vs 8c performance difference in applications which utilise cache sufficiently
I skipped and heard “first date” and got super excited for Steve
0:58 Me here liking the videos: it ain't much but it's honest work 😆
God I cant wait for these, im gonna be upgrading to a 9950x to from a gen 1 1700 its gonna be glorious T_T
Man, that’s gonna be such a great upgrade! I’m mildly jealous haha. I got a 5900X, so I’m probably good for awhile yet
So, what’s that, like twice the cores, about 75% higher clock speed, about a 50% IPC improvement (so you could consider that another +50% effective clock rate, in a sense), twice the cache per core.
Yeah, that’s a ridiculously good upgrade
@@mduckernzI know! I had done the math a couple weeks ago and it was like 2.5 times the single thread performance with clocks (@5ghz) and ipc, its going to significantly speed up my handbrake encoding for my 4k blurays. A year ago I had actually thought about getting a 5000 series as a stopgap and dropping it in my motherboard but decided against it as one of the top features im upgrading for is platform features like usb4, usb with power delivery through my case and 10G lan and more nvme storage at the top of the list
@@user-hs6ls9gy9mHopefully you upgrade the GPU as well. You might be paired that 9950x with a GTX 1060 lmao.
@@ZackSNetwork Oh no i has a 6700xt which i upgraded from an r9 270x last april ^_^
I'm loving the reduced temperatures of this new lineup. I'd imagine that's one of the reasons why the 9xxx series performs so will with higher wattages.
What I care about most is if I can populate all four memory slots without tanking speeds
That has never been the case since we started exceeding JEDEC spec, and I wouldn't hold my breath on this one. You'll probably have better luck pestering the memory manufacturers to make two-stick kits that max out the platform. 96GB per stick would be nice.... But alas, if you want speed, low latency and capacity your current best option is something like Corsair Vengeance 2x48GB 6000CL30.
That said, AMD is upping JEDEC spec from 5200 to 5600 so it's not a stupid loss in speed by populating all four slots.
I'm hoping we see some CAMM2 boards and high capacity CAMM2 modules to solve this limitation properly.
I went from a 1600X to a 5800X3D in the same motherboard. AM4 has been fantastic. Huge credit for upgrading their AGESA to support new processors on old chipsets. That was huge, and very quiet.
If the memory controller actually improved this is looking like a nice upgrade.
The memory controller itself stays the same, as the IO die is unchanged
@@redblack9925 where did you find that information ?
They literally increased the specced memory speed support, the memory oc speed support, and made it able to be changed on the fly
it also have more pcie5 lanes
@@redblack9925 ngl you sounds like parot from the 3000 5000 series ryzen cpus, which indeed had the same iod for most chip
@@s-x5373 Processor has the same number of PCI-E lanes and can't be increased on AM5 (there are no pins for it).
X870E looks like it may have increased speed of some chipset lanes. Not sure.
I am still running on my AM4 platform since 2018 and going strong. I am very satisfied I did not need to change stuff when I upgraded CPUs. I very much like where AMD is going.
Seeing Intel issues and me after having bad experience with rushed AM5 and DDR5, i would recommend everybody to be REALLY cautious before buying a new platform.
Hopefully they can get that IPC up to 18-20%!
I also think they should change the IPC comparison to 4.4-4.6GHz fixed instead of 4.0GHz to get a more realistic comparison.
AMD is kicking it whilst Intel struggles to deliver a base functional chip
finally all burnt 7000-series cpus replaced with burnt mobos, nicely done ;) 5700 GPU series black screen issues never got solved. Hip hip hooray AMD excellence indeed lmao. Not to mention the adjustments which were done for 3000-series Ryzens... :D Short is fanboy's memory :( Biden-syndrome?
Economy also refers to fuel-saving, efficiency, frugality, and so on.
Looking forward to a new RIP Jay series
The main thing I would love for socket longevity would be for the same CCX chiplets to be available with different interface chiplets to fit different sockets . So Zen 5 cores in an AM4 socket bottle necked by DDR4 RAM and older PCIe speeds, but with the interface chiplet maximizing bandwidth utilization so all busses run at peek speed to feed the cache from DRAM banks and PCIe memory in stuff like GPUs, SCSI controllers and high speed network cards . Similarly some Zen 4 or Zen 2 cores in the previous AMD socket that was originally used for Opteron and Athlon 64 chips . These old-socket parts would have an in-package bootstrap BIOS flash to make the chip boot in older machines by emulating the old CPUs until enough has run to do a safe transition to full speed of these "overdrive" style CPUs . Similarly Zen 10 cores from 2030 could be fitted with a compatibility chiplet for use in AM 5 boards with old 2023 BIOS code .
I wonder if we'll see a 9000 series threadripper or epyc
Of cause we will. The desktop CCDs are the ones that didn't make the cut for Epyc (the absolute best) and Threadripper (the second best).
Thanks for the deep dive! Always appreciated.
As I have a 7950X, I only care about the v-cache versions of Zen 5. I hope they do release the dual v-cache CCD model. Call it the 9990X3D, whatever
I have a 7950X3D myself and I cannot, for the love of me, figure out why people would want a dual V-Cache CCD SKU. As soon as you get cross-talk between the CCDs the gaming performance tanks. And that's regardless of whether both are vanilla or one is vanilla and the other is V-Cache. The games that need manual intervention are all games that spawn more than 8 threads if they see that more than 8 cores are available. Just forcibly locking them to the V-Cache CCD fixes the problem. Getting cross-talk between two V-Cache CCDs will not.
I had a 5950X before my 7950X3D, and it was the same there: some games would run worse than on a 5700X, but locking the game to a single CCD instantly fixed it.
@@andersjjensenThat all sounds annoying as hell and time consuming.
@ZackSNetwork it can be time consuming but it's fun. I love my 7950x3d but would definitely not recommend it to those who don't like to tinker. The performance and efficiency I get from mine is amazing
@@ZackSNetwork If you only care about gaming, get a 7800X3D. That said: With an XTX at 1440p I don't notice anything if I leave everything on defaults. I have to set resolutions and settings I would never run at to provoke the issue. But since I'm a software developer by trade I obviously cooked up a powershell script for Windows and a shell script for Linux that launches the game tied to the CCD it respons the best to, and edited my game shortcuts accordingly.
If you set the process affinity for steam to use only the one CCD then any games launched from it will use what steam uses.
Edit: Under windows that is, no idea about Linux.
8:17 I am shocked that Papermaster managed to go 25 seconds without saying AI here lmao
People already on AM5 should not upgrade to these new boards. No valid reason to.
The main selling points is WiFi 7 and USB4.
@@ZackSNetwork some x670e do have usb 4, as for wifi 7, pretty much pointless for basically everyone.. I mean the max speed difference is quite large and there's another channel at 320mhz but most places don't even go past the limit of 2.5 gpbs or maybe 10..
agreed but if your upgrading to am5 this is a good generation to start on
It really depends on your usecase, if you're pulling in $10k/month in content creation, a Zen5 might be justifiable upgrade.
@@woobilicious. true bro
Thanks for the relatively fast presentation. very time efficient with a lot of useful information. I appreciate that 🙏
Thanks, Steve.
I hope they learned from the hurdles of the 7950x3d and 7900x3d. They should just make both ccd's have 3d caches. That would get rid of the scheduler hurdles.
Not completely, unfortunately. The two CCDs each have separate cache, and communication between them has to happen over a slower external bus. Having bigger cache on both CCDs doesn't help if the data that one thread needs is in the cache on the other CCD. So, for gaming workloads, it needs to lock all the threads to one CCD *anyways*.
Bro at 3nm shining a flashlight on it will install red star os 💀💀💀
Hey GN, Is there any chance to get the audio track from your episodes uploaded to Spotify? Just can't keep the screen off and UA-cam running on my phone without having to start another subscription. Thank you! 👍🏼🏆
How long had they been using 8 cores for their X700 and X800, 6 generations?
At this point, the 9700X at 65w should've really been a 9600X. Guess both Nvidia and AMD are doing shrinkflation now and we probably won't see upgrade in cores for the mid-low range until the next generation of consoles shows up.😮💨
As always, wait for Hardware Unboxed to do a B850 roundup to figure out which ones are worth buying.
HU are not a credible source and haven't been for a long time.
Can we expect better idle power consumption ? I came back to LGA1700 because the amount of wasted power in the infinity fabric is too high… 20W doing nothing… where Intel can do a lot with 20W.
You bought a new mobo + CPU over 20w idle?
@@libertyprime9307 not sure to understand. 7600x drinks 20W in idle. Which is ridiculously high and generates heat. I don’t care about cost, but I care about heat.
@@nicoyt9481 You realize an incandescent light bulb draws 60w? You're talking about 1/3 of a light bulb. It's insignificant.
@@libertyprime9307 here in Europe, it is significant. Nobody uses incandescent light bulb anymore. LEDs are affordable by now
Some kind of video explaining what PCI lanes imply in daily usage would be interesting...
Like... How does the choice of a chipset influence the expandability of a desktop PC?
You also mentioned, I think, that motherboard manufacturers could add connectivity with additional chips or something... ...
I'm a bit confused.
B840 is a disgusting naming scheme, it's just a new A620.
PS: Why are they skipping 700?
Better branch prediction is very cool, but it's a double-edges sword, as it might cause security bugs (see spectre, meltdown etc). It would be super interesting to hear from the engineers how they account for those problems.
Gamer Nexus…what does the Scanner say about it’s Power Level?
It’s Ryzen 9000!!!
What 9000? There’s no way that can be right!!!
Granite ridge is a place in ontario canada and its gorgeous. Probably not related but hearing that made me remember the views up there.
This is rare for a gamernexus video but you said a lot of things however did not explain a lot of things. All the upgrades to the architecture I would like to understand their impacts, even just briefly. I'm a little disappointed you didn't go in a tad more detail. Thanks for the video though.
the short version: more data can be processed faster and more efficiently
You need to ask Ian Cutress about this
I read your comment as: you desired a statement from GN along the lines of, "What this means is that for gaming and for productivity is that it can execute x% more instructions in the same amount of time as previous generation, which _should_ lead to game / task running 10-15% faster"
-> Thing is, until GN have the CPUs in their possession, any such statement is speculation
-> You will probably get the statement(s) you require when GN gets hold of the new CPUs and does their own benchmarking.
From my perspective, GN cut to the AMD engineers talking to these points, then to the marketing slides showing alleged improvements vs selected games and productivity. If it sounded a bit too technical, then fair enough ...
GN then reiterated that the stats AMD put forward are cherry-picked stats from the manufacturer and require independent benchmarking when the products are available to validate AMDs claims on performance improvements.
These videos before the release are just going to be UA-camrs reading AMD's marketing brochures... Don't get your hopes high
It was probably done to keep the length of the video down. We'll probably get more indepth stuff as these chips hit the market
Amazing to see the huge transformation AMD has gone through from Piledriver to Zen 5.
Going full-on Team RED was one of the best decisions I've ever made. I build THREE 4K Ultra 60++ gaming rigs for the same money I would have spent on ONE Intel/nVidia rig.
I disagree, Christmas colors is the best. AMD X3D CPU with Nvidia GPU.
Nah, this doesn't add up at all.
Comparable Geforce cards are like $100 more than Radeon, and for Intel they should be similar price to Ryzen.
Post the components.
@@libertyprime9307 except you are completly wrong about intel
it's been the intel scam for the last 5 years,
you are actually needing new motherboards etc.. and better cooling solution (and maybe even better psu, and that also applies to nvidia gpus)
which both increase the cost significantly at start, but also increase extremely the cost throughout the life of the rig to be upgraded sometimes
so that does add up to having 3 rigs in the end, instead of one with big coolers and psu, and an useless motherboard
@@libertyprime9307 5900x/6950XT my rig, 5900x/6800XT wife's rig, 2600/5700 w/XT BIOS spare rig.
I just noticed the pencil/screwdriver jar made out of memory sticks! Nice :)
Regarding 19:24 AMD already did it with Threadripper. There is no Threadripper Socket with More than 2 generations of CPU Support
This was obviously not about Threadripper and AM4 from 2016(!) is still getting new CPUs which is nuts
A few nerds with deep pockets are salty about that, but the ones that make Threadripper sales actually turn profitable use OEM systems. My workplace included.
That's kind of because Threadripper existed the consumer/enthusiast market completely and in place CPU upgrades while maintaining the same motherboard aren't very common in workstation/enterprise settings, so the benefits of adding new features to the socket outweigh the loss of generation to generation compatibility. AM4 on the other hand loses a lot less because consumers are rarely pushing IO to the limit and are much more likely to do incremental upgrades. Plus, with 16 core AM4 parts the enthusiast market for Threadripper is very small (even in my case I would love to run a Threadripper system for my little home lab setup but 16 core AM4 is good enough and a lot cheaper)
two things:
1. threadripper started out for the enthusiast market, who will often upgrade even across one generation.
2. threadripper's second market, the high-end desktop workstation market, most of them are businesses who are wholly willing to throw out their motherboard and even their ram when they go to upgrade ($200 is almost a rounding error compared to the cost of everything else).
@@dead-claudia In the server space the in-place CPU upgrades are even more rare than in the workstations, yet AMD had made 3 generations of EPYC compatible with any SP4 motherboard - if the vendor provided the UEFI updates. Threadripper is a classic case of an engineering pet project getting taken over by marketing, with the usual follow-up of screw you towards both engineers and customers.
And screwing over their highest-paying enthusiasts had definitely made a dent in the AMD's sales. Because not only those folks buy the high-end hardware, they convince others to look at the brand they like.
Saying Economy instead of Efficiency really isn't a big deal, IMO. We're all human and sometimes say the wrong words, and the people who are watching that AMD presentation are people who know that he meant Efficiency.