Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[Hitman928]

What do you define as stability? If my son is able to do all the things he wants to use the PC for without crashing, that, to me, is stable, is it not? He's been using the PC for 9 months now. Heaven was included in testing to make sure the GPU and PSU were also up to the task. So how was this not sufficient? While it may not be sufficient for your specific use case it is what worked for us. How was your 5900x not stable after using benchmarks?

If a CPU is not truly stable then you could get any sort of "random" issues pop up over time. This can lead to OS issues, driver issues, or file corruption. Several years ago someone wrote a great piece on silent data corruption from unstable overclocks but I don't remember where it is anymore. Things like games can keep on rolling through CPU instability and be fine (maybe you get a bad frame that you don't think anything about or something) until you get to a place where the CPU is so unstable it just crashes but all these processes running over time means that its just a matter of time before one of the CPU errors causes some kind of "permanent" issue in the software or data. Now, if on your system you don't care about any of this because maybe all you do is game and potentially having to reinstall stuff once in a while doesn't bother you, then I guess that's ok, everyone has their own risk tolerances.

For me, my 5900x wasn't stable at -30 offset but passed cinebench on a loop and prime95 for hours at -20. I could game, browse, etc. all fine. Then I did more thorough testing and eventually found that some of the cores weren't stable at -20 and I ended up with a distribution of -20, -15, and -10 offsets across the cores. It still allowed me to drop the max power draw from 142W to (from memory without looking into the BIOS) 115W or 120W without any performance loss so I was happy with it overall even though it wasn't nearly as much power savings as some others were showing online.

 

[Panino Manino]

UB now and then:

Pre-release numbers for 7600X:

archive.ph

View attachment 68339

Post-release:

UserBenchmark: AMD Advanced Marketing Devices 7600X Zen 4 Hype Train

View attachment 68341

Just pathetic. 1 core result is now suddenly 20% lower, 2 core is 9% lower. 4 core is 16% lower etc. The guy has edited out all the scores, most likely simply applied some correction factor to the results or updated the benchmark suit....

But the actual results from the review benchmarks match the leaks, right? The 7600X for example IS on average faster than the 12900K in games, right? Meaning that the algorithm changed again?

 

[repoman0]

For me, my 5900x wasn't stable at -30 offset but passed cinebench on a loop and prime95 for hours at -20. I could game, browse, etc. all fine. Then I did more thorough testing and eventually found that some of the cores weren't stable at -20 and I ended up with a distribution of -20, -15, and -10 offsets across the cores. It still allowed me to drop the max power draw from 142W to (from memory without looking into the BIOS) 115W or 120W without any performance loss so I was happy with it overall even though it wasn't nearly as much power savings as some others were showing online.

What did you use to test it? I'm messing with this finally instead of buying new Zen 4 toys ... running CoreCycler for stability testing at -15 all core currently. Like you I care more about absolute stability than performance.

 

[inf64]

But the actual results from the review benchmarks match the leaks, right? The 7600X for example IS on average faster than the 12900K in games, right? Meaning that the algorithm changed again?

As per Tuna-Fish, the UB owner just keeps changing the weights on the UB website. The benchmarks are in line, 7600X is among the best CPUs for gaming, it all depends how much you tune it (Expo or not, IF clocks, PBO etc.). The problem with today's games and benchmarking is that in a lot of titles we are GPU bound even with 3090ti or 6950XT. We need 4090 or RDNA3 to show more tangible difference (if it even exists) between these cpus.

 

[In2Photos]

If a CPU is not truly stable then you could get any sort of "random" issues pop up over time. This can lead to OS issues, driver issues, or file corruption. Several years ago someone wrote a great piece on silent data corruption from unstable overclocks but I don't remember where it is anymore. Things like games can keep on rolling through CPU instability and be fine (maybe you get a bad frame that you don't think anything about or something) until you get to a place where the CPU is so unstable it just crashes but all these processes running over time means that its just a matter of time before one of the CPU errors causes some kind of "permanent" issue in the software or data. Now, if on your system you don't care about any of this because maybe all you do is game and potentially having to reinstall stuff once in a while doesn't bother you, then I guess that's ok, everyone has their own risk tolerances.

For me, my 5900x wasn't stable at -30 offset but passed cinebench on a loop and prime95 for hours at -20. I could game, browse, etc. all fine. Then I did more thorough testing and eventually found that some of the cores weren't stable at -20 and I ended up with a distribution of -20, -15, and -10 offsets across the cores. It still allowed me to drop the max power draw from 142W to (from memory without looking into the BIOS) 115W or 120W without any performance loss so I was happy with it overall even though it wasn't nearly as much power savings as some others were showing online.

Again, if the PC has shown no signs of any issues in 9 months that seems rather stable to me. Obviously if we had stability issues we would look at the voltage offset as that is the only change from stock aside of XMP on the memory. I don't see the need to continue testing for instability at this point.

 

[Hitman928]

What did you use to test it? I'm messing with this finally instead of buying new Zen 4 toys ... running CoreCycler for stability testing at -15 all core currently. Like you I care more about absolute stability than performance.

CoreCycler was one of the main ways. I also used FAH but limited the threads to percentages of total available and let the natural core migration happen. I also did straight Prime95 and FAH with full loads. I also ran blender and a few other things but it's been close to a couple of years now since I built the system so I don't remember exactly everything I tested with.

Again, if the PC has shown no signs of any issues in 9 months that seems rather stable to me. Obviously if we had stability issues we would look at the voltage offset as that is the only change from stock aside of XMP on the memory. I don't see the need to continue testing for instability at this point.

My point is that you don't really know if it's stable or not because nothing you are checking with will actual report miscalculations or wrong bits to you. Maybe it is stable, I don't know, but maybe its not and you are slowing getting background issues you just haven't noticed yet. If you're satisfied with it, I'm not trying to tell you to change anything, just giving my experience and perspective on the subject.

 

[pakotlar]

Core effective clock is not the same as maximum clock some of the cores hit :
View attachment 68363

So yes, some of the cores did hit 6Ghz.

Who cares if they hit 6ghz for a nanosecond? The effective clock is a more accurate representation of the cpu’s behavior.

For reference: https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/

“This new method has been tested on several CPUs and has shown to provide more accurate results especially in scenarios with extremely fluctuating values.”

The 7950x result is great, and I’m sure it can hit 6ghz, but it wasn’t achieved in the post I replied to.

 

[pakotlar]

What did you use to test it? I'm messing with this finally instead of buying new Zen 4 toys ... running CoreCycler for stability testing at -15 all core currently. Like you I care more about absolute stability than performance.

The best test of stability is to leave your PC on for a few weeks. It took me a few months to find stable CO values. Now my 5950x is stable with most cores at -25 to -30, but a few as low as -15. No crashes in 6 months, including a 2 month always on stretch.

 

[RnR_au]

B650 boards will be in channel on 10th Oct according to this (translated) article - https://benchlife-info.translate.go...l=auto&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=wapp

Picture of the Gigabyte B650I AORUS Ultra from the article. Pics of other boards are in there as well.


The interesting thing is that there is no 'E' in the model name of this ITX board. Too many data lines for a little board.

 

[biostud]

I was wondering if TSMC has raised the prices on 5nm vs 7nm, when AMD is launching at same or lower prices. Do you think AMD has the same margins as zen3?

 

[deasd]

AMD subreddit has a deep view on air cooling

https://www.reddit.com/r/Amd/comments/xqsa2j

Looks like 4 pipes 120mm fan is enough to cool everything below 7900X when temperature limit is 95C.

So, overall. while the chips say they run at 95, some of them have a clear happy place. while cpus like the 7600X seems fairly happy on ANYTHING on it.

The 7700x is happy with anything with 4 pipes and a 120mm fan, so hyper 212 fans rejoyce.

What shock me is, he is patient enough that he even tested 2 pipes 80mm('WTF??!!'When I saw this) on a 7600X, and have a nice and unexpected result.

 

[JustViewing]

AMD subreddit has a deep view on air cooling

Nice to see wraith prism doing well with these beasts

 

[coercitiv]

I was wondering if TSMC has raised the prices on 5nm vs 7nm, when AMD is launching at same or lower prices. Do you think AMD has the same margins as zen3?

The Zen 4 CCD die area is smaller than Zen 3, ~70mm2 vs 83mm2. This alone helps optimize the cost structure. Their ASP is probably also going up on AM5 even if prices appear to be same or lower. There is a high chance that consumers will avoid 7600X this round, as it makes less sense considering the overall platform cost. (that may change later as platform costs shift from optimization/competition)

Here's a sales sample data shared by TechEpiphany on Twitter, I'm assuming this is just one retailer (probably Mindfactory?):

• Ryzen 9 7950X: 230 units
• Ryzen 9 7900X: 190
• Ryzen 7 7700X: 200
• Ryzen 5 7600X: 50

Take it as anecdotal evidence.

 

[DrMrLordX]

I would like to know, how many threads of Cinebench R23 you can run on 7950X with 40 USD air coller (as ARCTIC Freezer 34 eSports Edition), without the CPU hitting 95°C and throttling, and what is the score. This cooler should have cca 160W cooling capacity.

Set it to ECO mode. You'll lose maybe 5-10% off your MT results. Maybe.

 

[inf64]

The Zen 4 CCD die area is smaller than Zen 3, ~70mm2 vs 83mm2. This alone helps optimize the cost structure. Their ASP is probably also going up on AM5 even if prices appear to be same or lower. There is a high chance that consumers will avoid 7600X this round, as it makes less sense considering the overall platform cost. (that may change later as platform costs shift from optimization/competition)

Here's a sales sample data shared by TechEpiphany on Twitter, I'm assuming this is just one retailer (probably Mindfactory?):

• Ryzen 9 7950X: 230 units
• Ryzen 9 7900X: 190
• Ryzen 7 7700X: 200
• Ryzen 5 7600X: 50

Take it as anecdotal evidence.

Yeah, the 7600X will either go down in price or the sales will go up once the mid tier boards launch. We'll have to wait and see.

 

[moinmoin]

AMD subreddit has a deep view on air cooling

https://www.reddit.com/r/Amd/comments/xqsa2j

Looks like 4 pipes 120mm fan is enough to cool everything below 7900X when temperature limit is 95C.



What shock me is, he is patient enough that he even tested 2 pipes 80mm('WTF??!!'When I saw this) on a 7600X, and have a nice and unexpected result.

View attachment 68402

Excellent testing. Reinforces my belief that, temperature reading being this useless, there needs to be a secondary reading akin to temperature pressure like I suggested earlier. Really anything that indicates potential throttling due to temperature without having to run benchmarks to detect that would help.

While expected this makes it harder to account for thermal throttling. If temperature is essentially always at 95C anyway there needs to be some additional value to monitor, like thermal pressure or whatever to call it what can eventually lead to thermal throttling.

Yeah, reviews of different cooling systems for Zen 4 will be very interesting.

 

[Ranulf]

Another data point on cutting out the heat by undervolting. 4.8ghz drops temps to 55C.

 

[inf64]

Excellent testing. Reinforces my belief that, temperature reading being this useless, there needs to be a secondary reading akin to temperature pressure like I suggested earlier. Really anything that indicates potential throttling due to temperature without having to run benchmarks to detect that would help.

I think that chips are hitting the PPT limit (while temps are always kept at 95C if that is set in BIOS). So users can either lower the temp limit or lower the PPT limit (or both). Undervolting proved to be the best way to go along with lowering the temp limit.

 

[moinmoin]

I think that chips are hitting the PPT limit (while temps are always kept at 95C if that is set in BIOS).

I don't follow, why would a chip with weak cooling hit the PPT limit? Isn't that a limit that would be reached with very strong cooling?

 

[inf64]

I don't follow, why would a chip with weak cooling hit the PPT limit? Isn't that a limit that would be reached with very strong cooling?

Well, the chip is designed to boost as many cores as possible until it hits 95C and then the power is measured (if below PPT or not) to determine if core clocks are to be decreased. That's my understanding on how it works but I might be totally wrong.

 

[BFG10K]

7600X/7700X provides amazing generational uplift in games.

A stark contrast to the incompetent Intel monopoly that were happy to give us +5% every 18 months on 14nm++++++.

 

[moinmoin]

Well, the chip is designed to boost as many cores as possible until it hits 95C and then the power is measured (if below PPT or not) to determine if core clocks are to be decreased. That's my understanding on how it works but I might be totally wrong.

No, you are right on that, but the issue with temperature reading is that it's essentially always maxed already whenever it would matter, so it's no longer a way to monitor if your chip is throttling due to the temperature limit. And unless you have strong cooling your chip at stock should be hitting the temperature limit way earlier than the PPT limit. To have any indication if and when that throttling due to temperature happens I suggested there should be some additional value to monitor from the Zen cores that shows how much they are adapting to the temperature (eventually leading to throttling).

 

[inf64]

No, you are right on that, but the issue with temperature reading is that it's essentially always maxed already whenever it would matter, so it's no longer a way to monitor if your chip is throttling due to the temperature limit. And unless you have strong cooling your chip at stock should be hitting the temperature limit way earlier than the PPT limit. To have any indication if and when that throttling due to temperature happens I suggested there should be some additional value to monitor from the Zen cores that shows how much they are adapting to the temperature (eventually leading to throttling).

Yeah, it's complicated algorithm for sure (under the hood). I read somewhere that hard limit is 115C (Tjmax) and this is where we see a true thermal throttle begins.
It's a good suggestion to monitor some additional value but I'd don't have a clue what else we might monitor except the temps and power.

 

[maddie]

The pow

No, you are right on that, but the issue with temperature reading is that it's essentially always maxed already whenever it would matter, so it's no longer a way to monitor if your chip is throttling due to the temperature limit. And unless you have strong cooling your chip at stock should be hitting the temperature limit way earlier than the PPT limit. To have any indication if and when that throttling due to temperature happens I suggested there should be some additional value to monitor from the Zen cores that shows how much they are adapting to the temperature (eventually leading to throttling).

The power reading is the "pressure" value you're thinking about. Once the cooling is fixed, meaning an installed cooler with "X" TIM, then cooling temp delta above ambient is purely determined by power.

 

[RnR_au]

Asrock with a B650E mitx... hmm...

Sixteen AMD B650E/B650 motherboards for Ryzen 7000 CPUs have been leaked - VideoCardz.com

AMD B650 motherboards are coming According to early sales data, retailers are seeing smaller interest in Ryzen 7/5 7000 models. The X670E and X670 boards are expensive to say the least, and that’s without including the cost of Ryzen 7000 CPU and DDR5 memory upgrade. However, there are new...

videocardz.com