Perfect and clear script, clean and easy to digest graphs (even on the phone), top notch b-rolls, and as unbiased as it gets. Thanks for the hard and honest work Steve and Tim, looking forward to seeing more Zen 4 tests, as well as the incoming ARC and RTX4000 reviews.
So going from a 360mm AIO to the puniest of air coolers only loses you ~5% in multicore and basically nothing in single core. Sounds like you can just get any reasonable air cooler and let the CPU do its thing.
So you want to spend like 900 bucks on the bundle and then cheap out on a 20 dollar cooler because you're leaving "only" 5% performance on the table? Folks usually don't buy high=powered CPUs only to throw away away performance, on the4 contrary, I can see der8auer's upcoming Ryzen 7000 de-lidding and direct-die cooling kit selling pretty darn well. ;) ua-cam.com/video/y_jaS_FZcjI/v-deo.html
@@aoife1122 People would be willing to leave a little bit of performance on an expensive setup if the next best setup was not even close. It's not like losing 5% performance means the new platform is not worth it. This will change when the platform is reaching end of life so you wanna squeeze out as much performance as possible.
Very interested to see testing done with single tower coolers like the noctua nh-u12s/a, currently use the 12s & if I ever built a zen 4 system I'd continue to use it.
Buildzoid just put out a 7950 power vs performance curve yesterday, too, with very encouraging results around 125-150W in Cinebench. Thanks for doing all this testing! HUB is my go to for insightful information that helps me plan my purchases, especially as a more budget conscious consumer.
@@kerbodynamicx472 No. Tech reviewers jumped the gun, and AMD is partially to blame for not clarifying how these new Zen 4 CPUs can be utilized Edit: I mean, of course you're leaving performance on the table if you limit power and use a lesser cooler, but the same can be said if you don't OC a typical CPU. To an extreme degree, you are leaving performance on the table by not liquid nitrogen cooling your parts lol.
@@kerbodynamicx472 It’s been shown through many reviews (and leaks) that aio (even 420mm ones) does jack sh*** to cool the cpu due to how thick the ihs is. Which is really a shame for those who wants to oc the cpu 😭 PS: you can still oc the cpu if you delid the cpu and do direct die cooling but that option is only for xoc enthusiasts and is not practical for the 99% of users
@@kerbodynamicx472 you can limit tpp to 150~175w and you won't lose any gaming performance and maybe ~6% multicore performance depending on the workload. And it will be a lot cooler.
Thanks for looking into stability/ instability with the PBO offset. Having a stable system is hard to guarantee when changing voltage settings. Looks like a simple power limit is pretty effective, though.
Thank you! We were waiting to these. Seems that you can manage with dirt cheap cooler after all! Also Eco mode could suit me well, because I like to keep the computer as quiet as possible without dropping result too much.
Great work Steve ❤ . I like your point about thorough stress testing to assure stability. I overclocked a cpu a couple years ago and all seemed fine until a few days later when it crashed…and then again and again at highly irregular intervals with many hours to days passing before the next crash. After a couple weeks of that, I set the clock back to standard.
I thought I had a stable undervolt that had been using for 4 months but it started to become unstable. Had to increase the voltage slightly to make it stable again
Yeah, I've had stable overclocks stop working after driver (GPU) or BIOS (CPU) updates. It was always good practice to back off a little for long term stability, once finding working settings. I've had some machines running "tuned" for a decade without any issue.
Superb content. This is exactly what I wanted to know, and you've delivered it perfectly. Now I just need to know if I can run RDNA3 on 600W (which is looking, surprisingly, quite likely)
I went through this exercise with my 5800X as it also ran rather hot in all-core workloads, but on the per core level using CoreCycler. I have individual cores running with offsets from -5 to -25 averaging probably around -20. In short, there's quite a bit more to be squeezed out in the per core vs the all core approach. That being said, it takes a while, because you need to be prepared to look out and adjust for instability (using WHEA error reporting, etc.) for a few weeks as the errors on the lower side of the curve are difficult to test for.
Thank you for this comparison. I find it very relevant and was really wandering what performance drop I would have with a lesser cooler. Great review and thanks again for your good work
It's cool to know that you could use the Warith Spire with 65W eco mode and be just fine. That could run you for a while until you can afford a nice cooler.
GDAY JEREMY HIDY…MATE would a NOCTUA NF-A12X25..fam make a difference with the Wraith Spire cooler..if some one could 3 D print SHROUDS…FOR VARIOUS SQUARE FANS..TO FIT WRAITH SPIRE TOPS..SLIDE THE FAN INTO THE SHROUD….SLIDE THE ROUND END OF THE SHROUD DOWN OVER THE WRATH SPIRE TOP….SLIDE IT DOWN TO SNUG FIT..NO NEED FOR SCREWS ..ANYMORE..
With no cheap motherboards the 7600x and 7700x are a bust. Still Valid though for power users on a 7900x/7950x as they wont be buying cheap motherboards anyway (unless they are stupid).
If the wraith spire can keep this Zen4 beast in check Pretty sure most generic 90/120mm single tower cooler will do just fine Im not pushing it past 65w aswell since i only do gaming and occasionally do some recording
A decent "Noctua knock-off" for $35 is perfectly capable of handling the stock 105W power limit and still keep it at or below 95C in any scenario. But with the price of electricity Eco Mode or a manual, say, 80W power limit + 90C thermal limit, is probably where most people will want to gravitate to.
Certainly a good "teaser" for further investigation. I definitely want to see the availability of other "ECO-like" power limits in BIOS and how they affect performance of the larger CPUs. Those of us who want to balance performance and power consumption (for whatever reason), but don't want to get into the experimentation of PBO will definitely appreciate the options different BIOSes can give us.
Many AM4 boards offer 3 ECO modes - 45W TDP (60W PPT), 65W TDP (88W PPT), and 95W TDP (128W PPT). You can also manually set a PPT limit if you choose. Looks like 65W TDP did very well with the Spire - I wonder if its the sweet spot though, as those ~77C temps seem cooler than necessary. I would imagine a ~75W TDP (100W PPT) might be the best balance between MT perf and temp with the Spire. It would be cool to test.
Thank you for addressing the stability. The -30 all core looks magical in results but it's rarely stable. It took me about 3 days of testing to dial in my 5900x for stability with a per core offset. The corecycler script is really useful for getting your per core dialled in and tested. It is great though that we get this level of control. When you see these chips running nearly as well or better well tuned. It really shines a better light on them compared to the release day running.
Yeah, undervolting is what overclocking has become. It is, at least to me, still as interesting to figure out your new CPU, where the limits are, how to stabilize it with LLC tweaks and all. Yes, out of the box is roughly max performance, but there's still fun to be had tuning them.
Thanks for pointing out the issues with Curve Optimizer. It seems to me that reviewers set -30, run a bunch of all-core tests, and go on recommending using it. The issue with CO is that it changes the voltage accross the entire CPU frequency range, reducing power consumption, making dynamic clocks higher. It's extremely difficult to test, and you have to test core per core. It took me weeks of testing to get my CO numbers right, and I consider it the only way to actually overclock these days. Problem is, that you can be stable with every multicore test you throw at it, but the moment a not so heavily threaded application runs your CPU clocks go higher than an all-core test and fail (or maybe not higher but just on a spot where those clocks and volts combination just don't work). Most of the time, instability is seen at idle, browsing, making daily tasks. Also, every CCD will have 2 preferred cores, and those will already have an aggressive curve set by AMD. Logically, those cores don't need much more tweaking, and tend to fail, in some cases even with -1 (Of course it's a lottery, and every CPU acts differenctly, but the point is: good cores already have a good curve).
Take the 7900X as an example, that has one "bad" CCD that won't clock nearly as much as the good one. That CCD has a crappy curve. Forget about the good CCD and only optimize the bad one. It will respond and clock much better with CO and match the good CCD clocks (most likely). But it also has 2 good cores, you have to always pay attention to the good cores on the CCD, they won't go too low on the curve.
@@THEpicND Identify your 2 best cores, and leave them at 0. Then, apply a negative offset on the other cores. Then, identify which of your good cores doesn't tolarate even -1, and lower the other one until it gets unstable.
-30 All-Core works perfectly fine on my 5900HS with Max Processer Frequency in Windows Power Plan set below 3900MHz (CoreCycler ran fine for over 6 hours, idle load & lightly threaded games haven't induced crashing) but seems to crap itself running the max all-core 4.45GHz, as AATU (which is what I used for the undervolt) doesn't have the option for Per-Core CO. For now until it gets updated to support Per-Core CO, I'm probably gonna keep it at 3600MHz -30.
I was only able to do -15mV on 5600X but it was mostly stable at -20mV. I only had 1 crash in a week so I should probably narrow it down and do a per-core setting. Either way it knocked off 5-6C while gaming so I'm happy with that result already.
Nice. I'm planning on waiting until January-February timeframe before starting a new build. At that time cheaper boards will have come out, and hopefully the price of DDR5 will have come down a bit.
Extremely helpful data. You added at least three or four pieces of context, consistently everywhere, that clarify what's going on. Many outlets gave a verbal explanation not directly attached to data, with a couple points of info missing to explain these unusual new temp/power/perf dynamics with AMD's unusual tuning here. It seems like it's a "high power" platform by default*, but with a ton of flexibility that can also work around poorer cooling or lower power caps, _especially well if you configure lower power settings..._ (Which IMO it seems like you generally should, and for sure if you have a basic cooler). (And the performance per temp responds a lot to under-volting, with YMMV effects on stability and maybe not game-changing perf differences.) * Edit: After I watched the gaming power and conclusions at the end... It isn't necessarily even a high-power platform! Rather, it will go high power to meet an extremely core-strenuous workload. But most workloads aren't anywhere near as high-power! So most of the time for a regular desktop user, these aren't really high-power parts! I'm very impressed with the reviewer community (and Hardware Unboxed in particular) for sorting this all out in a matter of a few days. Thanks very much! I wanted to understand, now I feel I do quite a bit better. *Follow-up question:* I wonder if there are any frame-time gotchas, such as 1% and 0.1% lows? Maybe there are difficult frames that would have boosted with higher power limits that become stutters when limited? Or is it really smooth even with the Eco mode? Does the basic air cooler hurt the 1% and 0.1% lows in Eco mode?
what I take away from this is the CPU is yawning at most gaming workloads. there are other bottlenecks in memory/gpu/IO where the cpu doesn't have to work too hard for a cooler to matter that much.
If you feel safe running the CPU on 100°C, then I guess you can use a cheap cooler, but I think the CPU will die after 2-3 years with this kind of stuff going on.
@@damara2268 nope you are wrong. Laptop CPU's are running at 95 degree C and they still run just fine. Yeah sure instead of running 18 years your CPU will only work 15 years now. But that's now a big deal.
@@divanshu5039 after 2 generations I was already looking at my AM4 system, should I move to AM5 or just get a 5800X3D Got the 5800X3D after seeing the price for AM5
@@damara2268 I always thought the general rule was running cooler lives longer, but I maybe older cpus just weren't as robust as newer ones. These things seem to run fine for a long time even at high temps. Not that I would choose to do that though. Haven't heard or read much serious debate on the issue. Seems like it hasn't been one for quite a while now although my dad insists even 75-80C is unacceptable.
Glad you mentioned PBO2 undervolt, because there is no way everyone will get 360mm AIO with B650 + 7600x with frustration that even with AIO temperatures are still near 95c
Buildzoid has a video with data on the performance vs power curves of various programs. Basically as suspected it's almost flat after 200w on the 16c model. AMD marketing should have included a laptop style 1-click app to select different tunings.
Using PBO2 together with Eco mode is the magic combo. With a voltage decrease there is a higher margin to the power limit, so you can regain some of the clock speed you lose by decreasing the power limit. Also, the power limit can be set to other values than 65W like the eco mode does, it would be interesting to see what the most efficient point is.
Wow i would probably just have eco mode on the entire time. The extra temps and wattage usage isnt worth the little out of the box performance when it still rips in eco mode
Wow, thank you for the video. Been waiting for videos like these and encourage more ppl to undervolt their cpus, even their current ones to use less power and produce less heat while keeping close to its stock settings. What's amazing is that a "stock" cooler performs very well vs a 360 rad on a 7600x. Can't wait for the videos with the other tiers of cpus.😁
Are you planning on trying to use ryzen core cycler for per core on curve optimizer? One of my cores can only do a -11 while 3 are stable at -30 and other between 25-20
Setting my PBO for my 5800x to -20 did the trick a few weeks ago. I've kept tweaking the settings and now I can get a single core boost to 4.95Ghz R23 single core test and up to 4.65Ghz on some of the cores during R23 multicore while the temps were under 75 C with my 280mm NZXT. It's a lot of work to tweak these CPUs and I don't plan on upgrading for another couple of years at least. It's taken me the 18 months of owning this CPU to finally get it dialed in and stable. Lol.
You could get a much lower undervolt if you go for per core undervolting. it's usually just one or two cores that can only go to -10, while the rest can go near -30. Thus capping all of the cores at -10, is leaving a lot of temperature/performance on the table.
how would you do this practically? Do you first set all to 10 and increase them to 30 one by one? what if a core crash at 30, do you instantly set it back to 10 or do you gradually decrease it? I am genuinely curious, it feels like youd have to test many combinations, but there might be some smart way around this
@@tandlose When I did it on my 5600X I mapped which prime95 worker was done on which core, and increased the offset on each core until the worker got an error. Something like that anyway... it's been a little while!
The only proper way of undervolting with pbo2 should be per core setting, because cores that are the best and which are most used by system will not take as high undervolting as the others. My 5600 is doing -5, -20, -25, -25, -25, -25. And instability after undervolting comes mostly from lightly threaded usage, so system might crash when idle but will run cinebench just fine.
13:40 You can set any TDP you like just by calculating the correct figures for PPT, TDC, and EDC and plugging them into PBO. For simplicity, you can pretty much just leave TDC and EDC at high values, and adjust PPT to 1.35x the TDP you want.
Surprised they didn't recommend eco mode testing out the gate, as loads of SFF builders may have written Zen 4 off because of the high power & heat ceilings.
Its only high because reviewers attached them to 360mm AIOs. If they did their tests with SFF heatsinks the power draw would be capped around 150w (depending on the cooler), but perform nearly the same. Both AMD and Intel realize that if you give a cpu an unlimited budget you can win the charts game but make a mess for people who dont use the same hardware as the reviewer. Its why I think reviewers need to stop removing every bottleneck and test parts with appropriate system configurations. HUB did a great review last year showing that 12th gen on cheap motherboards could lose up to 30% because the VRM system needed.
@@manicdan481 Yea, hardware testing is for sure going to have to evolve - both CPU & GPU testing in fact, with these driver/game features like DLSS/FSR/XESS becoming rather prominent, and with CPUs becoming more advanced and flexible. I'm not sure how they need to change, and what direction the media is going to have to move towards, but nonetheless testing is going to have to evolve to stay truly consumer relevant with this new hardware for sure.
@@RyTrapp0 yeah, some features i wonder if they go under valued. Sure an amd cpu might perform the same as an intel one for gaming, but what about if you turn on rebar. An extra 5% in gpu performance based on which cpu is used has some value. Its not exactly a fair comparison when looking to review a single component, but as a whole system its exactly the stuff that adds up.
I'm curious to see how the Thermalright peerless assassin would do with these chips. If the wraith spire does fine, then that cooler would probably thrive here. Edit - Especially when you consider it comes remarkably close to the nh-d15 in thermal performance for like, half the cost at 40$ for the rgb version.
(Note Zen4 CO is in points not milivolts now, so -10 is say -30mv) With curve optimizer on Zen3 I had to use per core setting, the 2 best cores are boosted the most so won't undervolt as much, the others can do more but it takes time to nail stable undervolts. There's a tool called Hydra by 1umus to thrash test individual cores and reports errors. I would ideally combine slightly lower max frequency, with CO undervolt and a lower PPT, which might need a reduced thermal limit. AMD marketing should develop a simple 1-click application for user choice. Perhaps later using data obtained by a USB boot validation and tuning data generation utility that can test memory, performance of build, as well as per core offsets to add value. The automated overclock settings they offer doesn't really work as it needs to avoid crashing windows so is too conservative.
Eco Mode has always been incredible since Ryzen 3000. I've been running my 5900x in Eco Mode and a -10UV which gave me stock single core and slightly better multi core performance, while having way better temps and less power draw.
Would be great to see some tests with some common cooling. We've seen extremely good and extremely bad cooling, would be nice to see some real-life examples. ZEN 4 should be a beast with laptops. I wish AMD set them at 85W out of the box because all this heat and power usage are not worth these 2-3% gains.
I'm eager to see how well a 45W Zen4 CPU performs in Gaming Laptops. Would also be nice to see performance if you limit power consumption to 25W. Some games lose very little performance while the whole system stays cooler and quieter.
@@attilad.8142 Definitely. Hell, scale it back to two or four core at 15W and even 10W and see the micro market go insane... it's be really good to see something take on Intel's N-series processors in the ultra-portable market.
You need to do a PBO2 offset for each core. You can actually set a core error warning through windows that will tell you which core is having the instability. I got 10 of my 12 cores on my 5900x at a minus 30 offset the other two are at like 15 and 20. It’s more intensive but man the power saving and thermals were amazing with more performance than stock
interesting, how do you get exact core data? my guess is whea logs but please do share. also why not go for straight all-core oc with c-states enabled?
The one thing that has been talked about, is how with AMD allowing cooler backwards compatibility, made the IHS a bit thicker to allow the coolers to work.
Is that really true though? I read that any cooler with a custom backplate will need an upgrade kit or it won't be compatible. It's pretty much guaranteed that older high-end AM4 coolers like the Dark Rock Pro 3 are not going to get such upgrade kits since those coolers have long since gotten newer versions, so those coolers would be useless for AM5 despite AMD claiming it's "compatible".
@@ApocDevTeam apparently it is compatible as long as your am4 cooler used the original AMD backplate or just the "hooks". You only need an upgrade part if your cooler used a replacement backplate for am4. So the backplate has changed from am4 to am5 but whatever mounted fine on the original am4 backplate will be mounting fine on the am5 backplate.
@@Habixus One other thing was that some brands of coolers come with their own custom backplate. The issue is that the CPU retention mechanism and the backplate of the AM5 boards are screwed together, and as such if you have a cooler with a custom backplate you need to get an adapter for it.
@@Habixus You made it even more confusing.. So are you saying that what I said is true in that any cooler that comes with its own backplate is not compatible? How is that 'fine'? Countless coolers come with their own backplates.
@@ApocDevTeam I didn't say it would be "fine" that all coolers with custom backplates would need a new backplate or adapter! I said coolers that used the original am4 backplate will be fine on am5. I have no statistics about which cooler used the AMD backplate and which cooler used a custom backplate. It's a problem that AMD made a confusing statement and didn't give clear information!
Thank you sir, this was exactly the info I was looking for. My only real reservation about Zen4 - aside from the platform costs - was the worry that it'd be trying to chase TJM even when not needed. I was hoping it wouldn't, but the early reviews made it sound like it really was going to be 95 all day 24/7. I'm not sure why they didn't tout Eco mode more, it works, and seems to work well, and at least in the case of gaming offers no noticeable difference in performance. I've personally seen great results from undervolt/overclock combos on both my i5-12600k/3060TI increasing efficiency quite a bit, reducing temps, and actually boosting overall performance well past stock configurations. From the limited results you showed here, looks promising to do the same with Zen4. Nicely done.
What I am curious about is what the impact of ECO mode on the 5800x3D is, considering it has all that cache. I am thinking that the cache could offset the lower power of ecomode.
Eco mode on Ryzen 7000 was probably how these CPU's were supposed to be launched, I have a suspicion they increased the power because Alder Lake performance was actually super good, going beyond eco mode actually feels like overclocking, you're getting way more heat a just a tiny bit better perfomance.
Since the AMD box cooler is able to operate without serious loss of performance, I would imagine that most tower coolers should be able to do it too. The only thing we need to watch out for is to check that they can be securely mounted on the motherboard as I saw some mountings are not compatible. We will now await the cost of AM5 motherboards and DDR5 to fall before taking the plunge...
As far as tower coolers go, you should check the manufacturer's website for compatibility. A majority of Noctua coolers work, but if a cooler requires a custom backplate them it's a no go.
So suddenly running a CPU at 90-95°C is completely fine just because AMD says so? Until now it wasn't considered safe to run desktop CPU at temperatures like that Why didn't AMD then increase warranty to 4 years to show they're confident in their statement ??
@@damara2268 I don't believe AMD - they are trying to defy the laws of physics. I do not believe that Zen 4 CPUs will last a long time if they are run at 90-95°C. I am OK with running a 7600X at 65W Eco mode as a performance loss of 4% won't be humanly noticeable. Our power bill will also be lower. For those who work in creative apps, the 7950X run at Eco 105W should also be acceptable. AMD is only interested in winning the benchmark wars. They don't care if the CPUs die after 3 years.
@@awright18 Well it required a custom backplate, so the ACTUAL problem is that provider, not the AM4/5 compatibility, without that "custom backplate", that AM4 cooler didn't work on AM4 either: you HAD to replace the AM4 system backplate.
The performance is nice but I live in Florida and it gets hot down here I know you live in Australia but you are doing this is a cool warehouse. That CPU will get uncomfortably hot for me to play with even with an Big AIO and ECO mode. I wonder how well Zen 4 performs under 5Ghz
For the curve optimiser to work properly it needs to be done on per core basis and test stability on the per core basis. Using all cores curve optimiser will introduce instability very quickly or leave a lot of performance on the table. I use OCCT to test stability and assign the stress test only to the cores i am currently working on, stress test and when i find my best value move on to the next one (2 at a time because due to smt each core is seen as 2 cores in windows) in my 5600x the best 2 cores (according to ryzen master & windows always assigning single core loads to them) only accepted offset of -2 but on the worst ones i was able to go over -20.
@@samgoff5289 only if u run it with a stock cooler which is well... if u have money to buy all the shit that come with am4 (Main, ram 5 and this 300$ CPU) but no money to buy an AIO or high quality fan cooler than u better not go for this range of CPU at all.
@@samgoff5289 it's less than 10 degrees C higher/hotter than my 5800x all core load so it ain't that much... Especialy considering the crap ton more frequency you get and the performance.
Mate, loved ya vid. I don't mind the heat in the winter, but in the summer might just pop this badboy into Eco and sacrifice a small percentage of my total power. Cheers.
I'm actually rocking a 7900X with an NH-U12A + Kryonaut Extreme - I've got it in 105W Eco mode with a -10 Curve Optimizer at the moment. Compared to your benchmark I've lost 500 (*edit - fact check, it's ~1000 points after a 10-min warm-up) in Cinebench R23 (even with a worse cooler and slightly worse 6000 CL32 memory). Temps max out at 72c instead of 95c and power draw is down ~50%. I'm going to slowly drop the curve more until it's not stable and see how close I can get it to stock at these settings. It'll happily do ~12 hrs of Prime95 + 12 Hours of AIDA64 with no issue, haven't tried blender yet. (I already had the NH-U12A from ~4 years ago, I had an old i7-3930K overclocked and degraded to near-death and it dried up my H100i)
@@Hardwareunboxed Just re-run it - You're right - it's a ~1000 point drop, not 500 (did a 10min loop instead of a single run). So about 3-4% performance loss. Edited above :)
I would recommend a manual power limit in Ryzen Master closer to 130 - 150w for the 7900X. Temps will probably be in the 80s, but you can manage a bearable fan curve as long as the core temps aren't instantly rocketing to 95.
@@konga382 I'll give that a go :) my old 3930K used to pull 160-170W with the overclock (even gaming as it always hit 100%) so I'm used to ramped fans.
Yes, it's funny in a way, Jayz2 is spas-ing out over temps & expensive cooler incompatibility while some low sub channels already discovered this and recommend air coolers.
Derbaur finding that delidding drops temps by 21 degrees implies that the problem is the IHS as far as peak temps. I would bet that your loop water in the 360rad was cool during those 95 degree spikes with a 7600x. Meaning there is a thermal boost bottleneck more than steady state heat output.
I think you should look at "Actually Hardcore Overclocking" power limit test, it is interesting, and can't wait for your test of multiple coolers. PS in first part of graphs AIO was first, on 2nd part AIO was 2nd, and you not mention about change in order, so it take few graphs to catch it.
Great, informative video Steve. I'm generally getting the impression that this new AMD CPU generation is a genuine step-change that will only improve with subsequent maturity. Aside from the MB/Memory issues,, I wanted to know if you feel this way? I feel stimulated for the first time in years.
Perfect and clear script, clean and easy to digest graphs (even on the phone), top notch b-rolls, and as unbiased as it gets. Thanks for the hard and honest work Steve and Tim, looking forward to seeing more Zen 4 tests, as well as the incoming ARC and RTX4000 reviews.
So going from a 360mm AIO to the puniest of air coolers only loses you ~5% in multicore and basically nothing in single core. Sounds like you can just get any reasonable air cooler and let the CPU do its thing.
True lmao, that 360 doesnt do shit
It's almost as if AMD knows what they're doing.
So you want to spend like 900 bucks on the bundle and then cheap out on a 20 dollar cooler because you're leaving "only" 5% performance on the table? Folks usually don't buy high=powered CPUs only to throw away away performance, on the4 contrary, I can see der8auer's upcoming Ryzen 7000 de-lidding and direct-die cooling kit selling pretty darn well. ;)
ua-cam.com/video/y_jaS_FZcjI/v-deo.html
@@aoife1122 People would be willing to leave a little bit of performance on an expensive setup if the next best setup was not even close. It's not like losing 5% performance means the new platform is not worth it. This will change when the platform is reaching end of life so you wanna squeeze out as much performance as possible.
Very interested to see testing done with single tower coolers like the noctua nh-u12s/a, currently use the 12s & if I ever built a zen 4 system I'd continue to use it.
Can I just say how consistently amazing the B-Roll shots look.
The lighting, backdrops, color and clarity is always superb.
Buildzoid just put out a 7950 power vs performance curve yesterday, too, with very encouraging results around 125-150W in Cinebench.
Thanks for doing all this testing! HUB is my go to for insightful information that helps me plan my purchases, especially as a more budget conscious consumer.
He showcased linpack foremost because cinebench is somewhat trash of a benchmark
So I don't need a 360 AIO for a 7900X?
@@kerbodynamicx472 No. Tech reviewers jumped the gun, and AMD is partially to blame for not clarifying how these new Zen 4 CPUs can be utilized
Edit: I mean, of course you're leaving performance on the table if you limit power and use a lesser cooler, but the same can be said if you don't OC a typical CPU. To an extreme degree, you are leaving performance on the table by not liquid nitrogen cooling your parts lol.
@@kerbodynamicx472 It’s been shown through many reviews (and leaks) that aio (even 420mm ones) does jack sh*** to cool the cpu due to how thick the ihs is. Which is really a shame for those who wants to oc the cpu 😭
PS: you can still oc the cpu if you delid the cpu and do direct die cooling but that option is only for xoc enthusiasts and is not practical for the 99% of users
@@kerbodynamicx472 you can limit tpp to 150~175w and you won't lose any gaming performance and maybe ~6% multicore performance depending on the workload.
And it will be a lot cooler.
Thanks for looking into stability/ instability with the PBO offset. Having a stable system is hard to guarantee when changing voltage settings. Looks like a simple power limit is pretty effective, though.
Thank you!
We were waiting to these. Seems that you can manage with dirt cheap cooler after all!
Also Eco mode could suit me well, because I like to keep the computer as quiet as possible without dropping result too much.
More quality content from hardware Unboxed. You guys never disappoint.😊
Nice to see Steve's enthusiasm this time of year. Great work as always!
Great work Steve ❤ . I like your point about thorough stress testing to assure stability. I overclocked a cpu a couple years ago and all seemed fine until a few days later when it crashed…and then again and again at highly irregular intervals with many hours to days passing before the next crash. After a couple weeks of that, I set the clock back to standard.
I thought I had a stable undervolt that had been using for 4 months but it started to become unstable. Had to increase the voltage slightly to make it stable again
Yeah, I've had stable overclocks stop working after driver (GPU) or BIOS (CPU) updates.
It was always good practice to back off a little for long term stability, once finding working settings. I've had some machines running "tuned" for a decade without any issue.
Phenomenal video... it answered all of my questions about cooling Ryzen 7000. Awesome job!
Amazing video !! great job exploring the PBO utility
Superb content. This is exactly what I wanted to know, and you've delivered it perfectly.
Now I just need to know if I can run RDNA3 on 600W (which is looking, surprisingly, quite likely)
Thank you so much showing Eco mode with gaming benchmarks.
I went through this exercise with my 5800X as it also ran rather hot in all-core workloads, but on the per core level using CoreCycler. I have individual cores running with offsets from -5 to -25 averaging probably around -20. In short, there's quite a bit more to be squeezed out in the per core vs the all core approach. That being said, it takes a while, because you need to be prepared to look out and adjust for instability (using WHEA error reporting, etc.) for a few weeks as the errors on the lower side of the curve are difficult to test for.
i used ryzen master and curve optimiser to do the per core then run OCCT. to check the stability
Excellent video guys. Good info and sound advice. Thank you.
Thank you for this comparison. I find it very relevant and was really wandering what performance drop I would have with a lesser cooler. Great review and thanks again for your good work
It's cool to know that you could use the Warith Spire with 65W eco mode and be just fine. That could run you for a while until you can afford a nice cooler.
GDAY JEREMY HIDY…MATE would a NOCTUA NF-A12X25..fam make a difference with the Wraith Spire cooler..if some one could 3 D print SHROUDS…FOR VARIOUS SQUARE FANS..TO FIT WRAITH SPIRE TOPS..SLIDE THE FAN INTO THE SHROUD….SLIDE THE ROUND END OF THE SHROUD DOWN OVER THE WRATH SPIRE TOP….SLIDE IT DOWN TO SNUG FIT..NO NEED FOR SCREWS ..ANYMORE..
Fine in winter time may be. For summer, a wraith prism is probably the minimum.
Who spends this kind of money on a system but lacks on the cooling? Dont be naïve
With no cheap motherboards the 7600x and 7700x are a bust. Still Valid though for power users on a 7900x/7950x as they wont be buying cheap motherboards anyway (unless they are stupid).
@@sassuki minimum for what?
If the wraith spire can keep this Zen4 beast in check
Pretty sure most generic 90/120mm single tower cooler will do just fine
Im not pushing it past 65w aswell since i only do gaming and occasionally do some recording
A decent "Noctua knock-off" for $35 is perfectly capable of handling the stock 105W power limit and still keep it at or below 95C in any scenario. But with the price of electricity Eco Mode or a manual, say, 80W power limit + 90C thermal limit, is probably where most people will want to gravitate to.
@@andersjjensen $15-20 like idcooling ones or those Snowman should do the job well enough
Thank you so much for this video!
I would love to see some budget tower coolers numbers along side the 360 AIO and the Wraith Spire results!
Certainly a good "teaser" for further investigation. I definitely want to see the availability of other "ECO-like" power limits in BIOS and how they affect performance of the larger CPUs. Those of us who want to balance performance and power consumption (for whatever reason), but don't want to get into the experimentation of PBO will definitely appreciate the options different BIOSes can give us.
Many AM4 boards offer 3 ECO modes - 45W TDP (60W PPT), 65W TDP (88W PPT), and 95W TDP (128W PPT). You can also manually set a PPT limit if you choose. Looks like 65W TDP did very well with the Spire - I wonder if its the sweet spot though, as those ~77C temps seem cooler than necessary. I would imagine a ~75W TDP (100W PPT) might be the best balance between MT perf and temp with the Spire. It would be cool to test.
When you told me you were having a BBQ on the weekend I didn't think you meant this lol
Bravo for doing this! This is the first YT video I’ve seen doing this on Zen 4
Good work looking at alternate cooling solutions and their worst case situations.
Thank you for addressing the stability. The -30 all core looks magical in results but it's rarely stable. It took me about 3 days of testing to dial in my 5900x for stability with a per core offset.
The corecycler script is really useful for getting your per core dialled in and tested.
It is great though that we get this level of control. When you see these chips running nearly as well or better well tuned. It really shines a better light on them compared to the release day running.
Negative Curve Optimizer can actuallly make your system unstable at idle too, so gotta let it idle overnight as well to see if stable
oh wow - i was just thinking about this the other day. great job!
That test was so damn helpful! I think we all were wondering what is happening with thermals Suri gaming. Nice video!
Yeah, undervolting is what overclocking has become. It is, at least to me, still as interesting to figure out your new CPU, where the limits are, how to stabilize it with LLC tweaks and all. Yes, out of the box is roughly max performance, but there's still fun to be had tuning them.
Thank you Hardware UnBOX COOLER
Great content sir! Please do a memory speed investigation as well!
Thanks for pointing out the issues with Curve Optimizer. It seems to me that reviewers set -30, run a bunch of all-core tests, and go on recommending using it. The issue with CO is that it changes the voltage accross the entire CPU frequency range, reducing power consumption, making dynamic clocks higher. It's extremely difficult to test, and you have to test core per core. It took me weeks of testing to get my CO numbers right, and I consider it the only way to actually overclock these days. Problem is, that you can be stable with every multicore test you throw at it, but the moment a not so heavily threaded application runs your CPU clocks go higher than an all-core test and fail (or maybe not higher but just on a spot where those clocks and volts combination just don't work). Most of the time, instability is seen at idle, browsing, making daily tasks. Also, every CCD will have 2 preferred cores, and those will already have an aggressive curve set by AMD. Logically, those cores don't need much more tweaking, and tend to fail, in some cases even with -1 (Of course it's a lottery, and every CPU acts differenctly, but the point is: good cores already have a good curve).
Take the 7900X as an example, that has one "bad" CCD that won't clock nearly as much as the good one. That CCD has a crappy curve. Forget about the good CCD and only optimize the bad one. It will respond and clock much better with CO and match the good CCD clocks (most likely). But it also has 2 good cores, you have to always pay attention to the good cores on the CCD, they won't go too low on the curve.
If I try to use curve optimizer with a negative offset on my 5600x It isn’t stable even at -1. I got a brown sample it seems
@@THEpicND Identify your 2 best cores, and leave them at 0. Then, apply a negative offset on the other cores. Then, identify which of your good cores doesn't tolarate even -1, and lower the other one until it gets unstable.
-30 All-Core works perfectly fine on my 5900HS with Max Processer Frequency in Windows Power Plan set below 3900MHz (CoreCycler ran fine for over 6 hours, idle load & lightly threaded games haven't induced crashing) but seems to crap itself running the max all-core 4.45GHz, as AATU (which is what I used for the undervolt) doesn't have the option for Per-Core CO. For now until it gets updated to support Per-Core CO, I'm probably gonna keep it at 3600MHz -30.
I was only able to do -15mV on 5600X but it was mostly stable at -20mV. I only had 1 crash in a week so I should probably narrow it down and do a per-core setting. Either way it knocked off 5-6C while gaming so I'm happy with that result already.
Nice. I'm planning on waiting until January-February timeframe before starting a new build. At that time cheaper boards will have come out, and hopefully the price of DDR5 will have come down a bit.
Extremely helpful data. You added at least three or four pieces of context, consistently everywhere, that clarify what's going on. Many outlets gave a verbal explanation not directly attached to data, with a couple points of info missing to explain these unusual new temp/power/perf dynamics with AMD's unusual tuning here.
It seems like it's a "high power" platform by default*, but with a ton of flexibility that can also work around poorer cooling or lower power caps, _especially well if you configure lower power settings..._ (Which IMO it seems like you generally should, and for sure if you have a basic cooler). (And the performance per temp responds a lot to under-volting, with YMMV effects on stability and maybe not game-changing perf differences.)
* Edit: After I watched the gaming power and conclusions at the end... It isn't necessarily even a high-power platform! Rather, it will go high power to meet an extremely core-strenuous workload. But most workloads aren't anywhere near as high-power! So most of the time for a regular desktop user, these aren't really high-power parts!
I'm very impressed with the reviewer community (and Hardware Unboxed in particular) for sorting this all out in a matter of a few days. Thanks very much! I wanted to understand, now I feel I do quite a bit better.
*Follow-up question:* I wonder if there are any frame-time gotchas, such as 1% and 0.1% lows? Maybe there are difficult frames that would have boosted with higher power limits that become stutters when limited? Or is it really smooth even with the Eco mode? Does the basic air cooler hurt the 1% and 0.1% lows in Eco mode?
Awesome as usual. Always appreciate the work you put in Steve.
Looks like an acceptable compromise. Very concise and clear explanation. Thank you, HU!
HUB is king when it comes to content like this; intensive testing with multiple sets of hardware variables
All hail the King!
Gotta say, I honestly thought there was going to be a much larger difference between these two. That's just amazing how well these CPUs can be tuned!
what I take away from this is the CPU is yawning at most gaming workloads. there are other bottlenecks in memory/gpu/IO where the cpu doesn't have to work too hard for a cooler to matter that much.
If you feel safe running the CPU on 100°C, then I guess you can use a cheap cooler, but I think the CPU will die after 2-3 years with this kind of stuff going on.
@@damara2268 nope you are wrong. Laptop CPU's are running at 95 degree C and they still run just fine.
Yeah sure instead of running 18 years your CPU will only work 15 years now. But that's now a big deal.
@@divanshu5039 after 2 generations I was already looking at my AM4 system, should I move to AM5 or just get a 5800X3D
Got the 5800X3D after seeing the price for AM5
@@damara2268 I always thought the general rule was running cooler lives longer, but I maybe older cpus just weren't as robust as newer ones. These things seem to run fine for a long time even at high temps. Not that I would choose to do that though. Haven't heard or read much serious debate on the issue. Seems like it hasn't been one for quite a while now although my dad insists even 75-80C is unacceptable.
Thanks for doing the hard work! Great video!
Great info! I'm a believer in water cooling, but it's nice to know the CPU can handle the high temps as described.
Glad you mentioned PBO2 undervolt, because there is no way everyone will get 360mm AIO with B650 + 7600x with frustration that even with AIO temperatures are still near 95c
im just going to get a 250-260W air cooler. and cap the power and the temp desired on the Bios
@@german2122 you can even change those settings from Windows.
so if you needed extra power suddenly then you'd be a single click away
@@german2122 what temp would you set it to?
@@motoxtim34 i think 85 full load is a good temp. but as the video says if the cpu doesn't need it it wont go past 70°
Buildzoid has a video with data on the performance vs power curves of various programs.
Basically as suspected it's almost flat after 200w on the 16c model.
AMD marketing should have included a laptop style 1-click app to select different tunings.
On the topic of UV stability: I've found that Pugetbench for Premiere or Resolve gives a very very good indication for system stability.
Using PBO2 together with Eco mode is the magic combo. With a voltage decrease there is a higher margin to the power limit, so you can regain some of the clock speed you lose by decreasing the power limit. Also, the power limit can be set to other values than 65W like the eco mode does, it would be interesting to see what the most efficient point is.
This was very informative. Thanks.
Wow i would probably just have eco mode on the entire time. The extra temps and wattage usage isnt worth the little out of the box performance when it still rips in eco mode
Too hot for my wallet.
Gotta wait for the B650 and price drop on ddr5 and the cpus
@@50H3i1 the first batch of b650s leaked are starting from 200$
To hot for my wallet and pc case.
Wait till AM5 matures. This is why I like AMD platform with 5+ year socket support
Then get previous gen and be happy. Plenty of budget choices in the market
the little animation flairs in the intros are a nice touch.
Wow, thank you for the video. Been waiting for videos like these and encourage more ppl to undervolt their cpus, even their current ones to use less power and produce less heat while keeping close to its stock settings. What's amazing is that a "stock" cooler performs very well vs a 360 rad on a 7600x. Can't wait for the videos with the other tiers of cpus.😁
This pretty much means my NH-U12A will do just fine on Zen 4 and possibly Zen 5 as well, good to know
Are you planning on trying to use ryzen core cycler for per core on curve optimizer?
One of my cores can only do a -11 while 3 are stable at -30 and other between 25-20
Great vid. Appreciate the effort.
Very interesting! good job!
Setting my PBO for my 5800x to -20 did the trick a few weeks ago. I've kept tweaking the settings and now I can get a single core boost to 4.95Ghz R23 single core test and up to 4.65Ghz on some of the cores during R23 multicore while the temps were under 75 C with my 280mm NZXT. It's a lot of work to tweak these CPUs and I don't plan on upgrading for another couple of years at least. It's taken me the 18 months of owning this CPU to finally get it dialed in and stable. Lol.
You could get a much lower undervolt if you go for per core undervolting. it's usually just one or two cores that can only go to -10, while the rest can go near -30. Thus capping all of the cores at -10, is leaving a lot of temperature/performance on the table.
This! This is what I was preparing to comment
And that is probably tied to the hottest cores as temp spread seems to be +10c
how would you do this practically? Do you first set all to 10 and increase them to 30 one by one? what if a core crash at 30, do you instantly set it back to 10 or do you gradually decrease it? I am genuinely curious, it feels like youd have to test many combinations, but there might be some smart way around this
@@tandlose When I did it on my 5600X I mapped which prime95 worker was done on which core, and increased the offset on each core until the worker got an error.
Something like that anyway... it's been a little while!
@@tandlose I usually change the value by 5. You need to use cpuid and the Event-logger to see what core/thread crashed.
Nice video.
Keep up the Most Excellent Work
The test I was waiting for 👍
The only proper way of undervolting with pbo2 should be per core setting, because cores that are the best and which are most used by system will not take as high undervolting as the others. My 5600 is doing -5, -20, -25, -25, -25, -25. And instability after undervolting comes mostly from lightly threaded usage, so system might crash when idle but will run cinebench just fine.
I'm sure you will revisit all these CPUs again soon with the new GPUs coming.
I love the Overwatch font at the start for the subject title. Very sexy. Didn't think your videos could look any more slick 😎👌
13:40 You can set any TDP you like just by calculating the correct figures for PPT, TDC, and EDC and plugging them into PBO. For simplicity, you can pretty much just leave TDC and EDC at high values, and adjust PPT to 1.35x the TDP you want.
Surprised they didn't recommend eco mode testing out the gate, as loads of SFF builders may have written Zen 4 off because of the high power & heat ceilings.
Its only high because reviewers attached them to 360mm AIOs. If they did their tests with SFF heatsinks the power draw would be capped around 150w (depending on the cooler), but perform nearly the same.
Both AMD and Intel realize that if you give a cpu an unlimited budget you can win the charts game but make a mess for people who dont use the same hardware as the reviewer.
Its why I think reviewers need to stop removing every bottleneck and test parts with appropriate system configurations. HUB did a great review last year showing that 12th gen on cheap motherboards could lose up to 30% because the VRM system needed.
@@manicdan481 Yea, hardware testing is for sure going to have to evolve - both CPU & GPU testing in fact, with these driver/game features like DLSS/FSR/XESS becoming rather prominent, and with CPUs becoming more advanced and flexible. I'm not sure how they need to change, and what direction the media is going to have to move towards, but nonetheless testing is going to have to evolve to stay truly consumer relevant with this new hardware for sure.
@@RyTrapp0 yeah, some features i wonder if they go under valued. Sure an amd cpu might perform the same as an intel one for gaming, but what about if you turn on rebar. An extra 5% in gpu performance based on which cpu is used has some value. Its not exactly a fair comparison when looking to review a single component, but as a whole system its exactly the stuff that adds up.
I'm curious to see how the Thermalright peerless assassin would do with these chips. If the wraith spire does fine, then that cooler would probably thrive here.
Edit - Especially when you consider it comes remarkably close to the nh-d15 in thermal performance for like, half the cost at 40$ for the rgb version.
It's funny how I think... Gee, I'd love to see how this CPU would work with.... and you guys are already making the video! Very nice job! Super R&D.
Keep it up Steve. This answers a big question for me.
(Note Zen4 CO is in points not milivolts now, so -10 is say -30mv)
With curve optimizer on Zen3 I had to use per core setting, the 2 best cores are boosted the most so won't undervolt as much, the others can do more but it takes time to nail stable undervolts. There's a tool called Hydra by 1umus to thrash test individual cores and reports errors.
I would ideally combine slightly lower max frequency, with CO undervolt and a lower PPT, which might need a reduced thermal limit.
AMD marketing should develop a simple 1-click application for user choice.
Perhaps later using data obtained by a USB boot validation and tuning data generation utility that can test memory, performance of build, as well as per core offsets to add value.
The automated overclock settings they offer doesn't really work as it needs to avoid crashing windows so is too conservative.
Eco Mode has always been incredible since Ryzen 3000. I've been running my 5900x in Eco Mode and a -10UV which gave me stock single core and slightly better multi core performance, while having way better temps and less power draw.
wow thats a lot of work. thanks for such high quality content. this is why I recommend Hardware Unboxed to everyone
Very impressive results! WoW! This is the way!
(refrain) This is the way !
Would be great to see some tests with some common cooling.
We've seen extremely good and extremely bad cooling, would be nice to see some real-life examples.
ZEN 4 should be a beast with laptops. I wish AMD set them at 85W out of the box because all this heat and power usage are not worth these 2-3% gains.
I'm eager to see how well a 45W Zen4 CPU performs in Gaming Laptops. Would also be nice to see performance if you limit power consumption to 25W. Some games lose very little performance while the whole system stays cooler and quieter.
@@attilad.8142 Definitely. Hell, scale it back to two or four core at 15W and even 10W and see the micro market go insane... it's be really good to see something take on Intel's N-series processors in the ultra-portable market.
You need to do a PBO2 offset for each core. You can actually set a core error warning through windows that will tell you which core is having the instability. I got 10 of my 12 cores on my 5900x at a minus 30 offset the other two are at like 15 and 20. It’s more intensive but man the power saving and thermals were amazing with more performance than stock
interesting, how do you get exact core data? my guess is whea logs but please do share.
also why not go for straight all-core oc with c-states enabled?
that or just use corecycler
@@emperorSbraz yes its the WHEA logger. Look for the APIC ID, it tells you which thread crashed.
@@rdmz135 gotcha thanks m8
Internet panic over, sanity restored, well done Hardware Unboxed.
Sick video, it is good to know this.
I guess Noctua U12A is going to be just fine.
Be interesting to see what some of the cheapest budget coolers do.
Fuma Scythe 2
The one thing that has been talked about, is how with AMD allowing cooler backwards compatibility, made the IHS a bit thicker to allow the coolers to work.
Is that really true though? I read that any cooler with a custom backplate will need an upgrade kit or it won't be compatible. It's pretty much guaranteed that older high-end AM4 coolers like the Dark Rock Pro 3 are not going to get such upgrade kits since those coolers have long since gotten newer versions, so those coolers would be useless for AM5 despite AMD claiming it's "compatible".
@@ApocDevTeam apparently it is compatible as long as your am4 cooler used the original AMD backplate or just the "hooks". You only need an upgrade part if your cooler used a replacement backplate for am4.
So the backplate has changed from am4 to am5 but whatever mounted fine on the original am4 backplate will be mounting fine on the am5 backplate.
@@Habixus One other thing was that some brands of coolers come with their own custom backplate. The issue is that the CPU retention mechanism and the backplate of the AM5 boards are screwed together, and as such if you have a cooler with a custom backplate you need to get an adapter for it.
@@Habixus You made it even more confusing.. So are you saying that what I said is true in that any cooler that comes with its own backplate is not compatible? How is that 'fine'? Countless coolers come with their own backplates.
@@ApocDevTeam I didn't say it would be "fine" that all coolers with custom backplates would need a new backplate or adapter! I said coolers that used the original am4 backplate will be fine on am5.
I have no statistics about which cooler used the AMD backplate and which cooler used a custom backplate. It's a problem that AMD made a confusing statement and didn't give clear information!
Thank you sir, this was exactly the info I was looking for. My only real reservation about Zen4 - aside from the platform costs - was the worry that it'd be trying to chase TJM even when not needed. I was hoping it wouldn't, but the early reviews made it sound like it really was going to be 95 all day 24/7.
I'm not sure why they didn't tout Eco mode more, it works, and seems to work well, and at least in the case of gaming offers no noticeable difference in performance. I've personally seen great results from undervolt/overclock combos on both my i5-12600k/3060TI increasing efficiency quite a bit, reducing temps, and actually boosting overall performance well past stock configurations. From the limited results you showed here, looks promising to do the same with Zen4. Nicely done.
Even if u run it on 95 for all day it will be normal
I'm planning to buy the B650 together with the 7600x when it goes on sale, so I watched it with great interest. thank you!
Have you had a chance to try out per core undervolting? Does Ryzen Master support automatically configuring the voltages for Zen 4 yet?
I recommend watching Tech Yes City's video on underclocking a 7950X with Ryzen Master.
Ryzen Master is still kinda buggy. Best to do it in BIOS.
What I am curious about is what the impact of ECO mode on the 5800x3D is, considering it has all that cache.
I am thinking that the cache could offset the lower power of ecomode.
The 5000 series has ECO mode??
@@scarkillerful you can set it in ryzen master, but you must enable PBO in the bios.
@@Hjominbonrun PBO is locked on the 5800X3D.
@@JirayD Ah, so 5800X can do it but not the X3d.
@@Hjominbonrun yes
This is exactly the content I was looking for, thanks HWU!
Thank you!
As always Steve, you've saved the day. 65w Eco mode looks like the way to go, tyvm :)
At least for the summer months, during the winter I'm turning my heating off and going full power :)
@@andyk2181 that's the way mate..
Eco mode on Ryzen 7000 was probably how these CPU's were supposed to be launched, I have a suspicion they increased the power because Alder Lake performance was actually super good, going beyond eco mode actually feels like overclocking, you're getting way more heat a just a tiny bit better perfomance.
awesome thanks for sharing
The new outro music is so cool
Since the AMD box cooler is able to operate without serious loss of performance, I would imagine that most tower coolers should be able to do it too. The only thing we need to watch out for is to check that they can be securely mounted on the motherboard as I saw some mountings are not compatible.
We will now await the cost of AM5 motherboards and DDR5 to fall before taking the plunge...
As far as tower coolers go, you should check the manufacturer's website for compatibility. A majority of Noctua coolers work, but if a cooler requires a custom backplate them it's a no go.
So suddenly running a CPU at 90-95°C is completely fine just because AMD says so? Until now it wasn't considered safe to run desktop CPU at temperatures like that
Why didn't AMD then increase warranty to 4 years to show they're confident in their statement ??
@@damara2268 I don't believe AMD - they are trying to defy the laws of physics. I do not believe that Zen 4 CPUs will last a long time if they are run at 90-95°C. I am OK with running a 7600X at 65W Eco mode as a performance loss of 4% won't be humanly noticeable. Our power bill will also be lower. For those who work in creative apps, the 7950X run at Eco 105W should also be acceptable.
AMD is only interested in winning the benchmark wars. They don't care if the CPUs die after 3 years.
Wait for the B650, buying an X670 is like "deciding" to buy a Z790 because it is available first.
@@awright18 Well it required a custom backplate, so the ACTUAL problem is that provider, not the AM4/5 compatibility, without that "custom backplate", that AM4 cooler didn't work on AM4 either: you HAD to replace the AM4 system backplate.
The performance is nice but I live in Florida and it gets hot down here I know you live in Australia but you are doing this is a cool warehouse. That CPU will get uncomfortably hot for me to play with even with an Big AIO and ECO mode. I wonder how well Zen 4 performs under 5Ghz
It actually won't. You need to focus on power consumption, not operating temperature.
@@Hardwareunboxed thanks Steve
FUN to test all these computers!
For the curve optimiser to work properly it needs to be done on per core basis and test stability on the per core basis. Using all cores curve optimiser will introduce instability very quickly or leave a lot of performance on the table. I use OCCT to test stability and assign the stress test only to the cores i am currently working on, stress test and when i find my best value move on to the next one (2 at a time because due to smt each core is seen as 2 cores in windows) in my 5600x the best 2 cores (according to ryzen master & windows always assigning single core loads to them) only accepted offset of -2 but on the worst ones i was able to go over -20.
PBO2 undervolt, here we go again (just like the Ryzen 5000 series)
Except now you are running 90c with an undervolt lmao.....
@@samgoff5289 only if u run it with a stock cooler which is well... if u have money to buy all the shit that come with am4 (Main, ram 5 and this 300$ CPU) but no money to buy an AIO or high quality fan cooler than u better not go for this range of CPU at all.
@@samgoff5289 it's less than 10 degrees C higher/hotter than my 5800x all core load so it ain't that much... Especialy considering the crap ton more frequency you get and the performance.
Just this morning i was thinking if my wraith prism would cool my 5800x3d lol.
Funny enough 90+ was scary and way to high for that cpu, but now suddenly it's fine for the new ones 🤷♂️
We never had an issue with the 5800X3D and cooling.
@@samgoff5289 magic silicon that only amd has access to 🤣
@@samgoff5289 you can thank Intel that made it normal. And Nvidia too for GPU
@@nekogami87 ah yes of course....the amd apologist ready to defend them and point the blame toward another company
Mate, loved ya vid.
I don't mind the heat in the winter, but in the summer might just pop this badboy into Eco and sacrifice a small percentage of my total power.
Cheers.
High casual workload temp usually implies lowered longevity and that it's an overclock.
I'm actually rocking a 7900X with an NH-U12A + Kryonaut Extreme - I've got it in 105W Eco mode with a -10 Curve Optimizer at the moment.
Compared to your benchmark I've lost 500 (*edit - fact check, it's ~1000 points after a 10-min warm-up) in Cinebench R23 (even with a worse cooler and slightly worse 6000 CL32 memory).
Temps max out at 72c instead of 95c and power draw is down ~50%.
I'm going to slowly drop the curve more until it's not stable and see how close I can get it to stock at these settings.
It'll happily do ~12 hrs of Prime95 + 12 Hours of AIDA64 with no issue, haven't tried blender yet.
(I already had the NH-U12A from ~4 years ago, I had an old i7-3930K overclocked and degraded to near-death and it dried up my H100i)
105w Eco Mode on the 7900X will almost certainly result in a larger performance hit.
@@Hardwareunboxed Just re-run it - You're right - it's a ~1000 point drop, not 500 (did a 10min loop instead of a single run). So about 3-4% performance loss. Edited above :)
I would recommend a manual power limit in Ryzen Master closer to 130 - 150w for the 7900X. Temps will probably be in the 80s, but you can manage a bearable fan curve as long as the core temps aren't instantly rocketing to 95.
@@konga382 I'll give that a go :) my old 3930K used to pull 160-170W with the overclock (even gaming as it always hit 100%) so I'm used to ramped fans.
That's pretty damn impressive. I'm really liking Zen 4 more and more
Yes, it's funny in a way, Jayz2 is spas-ing out over temps & expensive cooler incompatibility while some low sub channels already discovered this and recommend air coolers.
Derbaur finding that delidding drops temps by 21 degrees implies that the problem is the IHS as far as peak temps. I would bet that your loop water in the 360rad was cool during those 95 degree spikes with a 7600x. Meaning there is a thermal boost bottleneck more than steady state heat output.
I think you should look at "Actually Hardcore Overclocking" power limit test, it is interesting, and can't wait for your test of multiple coolers.
PS in first part of graphs AIO was first, on 2nd part AIO was 2nd, and you not mention about change in order, so it take few graphs to catch it.
Hi steve ur handsome
holy shit 101°C????
Tech Yes City had some really good results undervolting these.
Great, informative video Steve. I'm generally getting the impression that this new AMD CPU generation is a genuine step-change that will only improve with subsequent maturity. Aside from the MB/Memory issues,, I wanted to know if you feel this way? I feel stimulated for the first time in years.