- Mar 3, 2017
- 1,747
- 6,598
- 136
Probably they were testing it in a 100% gpu bound scenario (aka if you don't set grass to extreme, for example)
At least one of them is taking great care to test in cpu bound scenarios. Now I don't have a game and haven't tried to reproduce their results. This is a Polish internet site, I could link to that is also having them equal.Probably they were testing it in a 100% gpu bound scenario (aka if you don't set grass to extreme, for example)
If it does it's error in testing and it's effect of some other change. The codepath should be AMD specific and Tom's Hardware's retesting found 0% change on Intel.
Intel also benefited from the patch, just not as significant as the Ryzen. It's now only 3% behind, whereas previously it was 5% behind.
TPU tests ST power numbers. Maybe they'll do updated testing, but I doubt it. Their graph is why I figured it was a power saving feature to begin with:
Hwcooling tests power consumption by physically measuring draw on the 12V cable feeding the VRM and found idle power to be higher, not lower.
AMD Ryzen 9 9900X in a megatest of 40 CPUs - Page 36 of 41 - HWCooling.net
Average processor power drawWe’re wrapping up tests of the new Ryzen 9000 desktop CPUs. This with a model that once increased the number of cores compared to its predecessor, or a pilot model (Ryzen 9 3900X) from eight to twelve on AMD’s mainstream platform. Even compared to the Ryzen 9 5900X...www.hwcooling.net
If the CCX latency thing improved power, it was not enough to offset the difference.
Instant idle power depends on the whims of the operating system and background processes a lot, so numbers are often going to be inconsistent gen-on-gen though.
Then they should just have released them without the X in the end. The inconsistency and misleading numbers from AMD's marketing is not doing them a favor in "our" community.The crazy thing is had AMD released the 9700X as a 105W chip and it had the additional 10% uplift, no one would have cared what the results were at 65W. But since it was released at 65W and reviewed that way, now all anyone will talk about is the cost of the performance increase, i.e. 40% more power for only 10% more performance.
Oh sorry, seems I should have paid more attention. In that case the audio encoding numbers (commandline FLAC task, strictly 1T) in the link would be applicable. Seems the dual-die SKUs actually also show increase in power in the test. Only 9600X had lower power than 7600X (for the board+memory and samples/condititions used in the test).That's why I used ST power, not idle power.
Edit: BTW, I'm not convinced the latency issue is from a power saving technique, it's just speculation that makes sense to me and seems to be supported by TPU's ST power results, though that could be due to some other factor.
Hwcooling tests power consumption by physically measuring draw on the 12V cable feeding the VRM and found idle power to be higher, not lower.
AMD Ryzen 9 9900X in a megatest of 40 CPUs - Page 36 of 41 - HWCooling.net
Average processor power drawWe’re wrapping up tests of the new Ryzen 9000 desktop CPUs. This with a model that once increased the number of cores compared to its predecessor, or a pilot model (Ryzen 9 3900X) from eight to twelve on AMD’s mainstream platform. Even compared to the Ryzen 9 5900X...www.hwcooling.net
If the CCX latency thing improved power, it was not enough to offset the difference.
Instant idle power depends on the whims of the operating system and background processes a lot, so numbers are often going to be inconsistent gen-on-gen though.
I know others have said the r23 does not use avx-512, but maybe it does ? Regardless the instructions sets are different on Zen 5 vs Zen 4, and the ones it uses in R23 may use more power, just like avx-512 does.Not only that, R23 for their 9950X was around 256W vs 217W for 7950X for a score of 41.6K for the 9950X vs 36827 for the 7950X. The 9950X is not what HWinfo64 shows as PPT for default operation, but the R23 score is. What gives?
This goes counter to every other review Ive seen thus far, which says 9950X PPT is around 199-200W during the R23 nT run, while the 7950X is around 220W.I know others have said the r23 does not use avx-512, but maybe it does ? Regardless the instructions sets are different on Zen 5 vs Zen 4, and the ones it uses in R23 may use more power, just like avx-512 does.
Did you ever run one at 230W but otherwise stock? Curious if that last 30W does anything. Or if that's why AMD cut it.Default PPT power limit for the 9950X is 200w
I would know.. have already binned a few
The only way i could see ~250w being right, is if they are messuring from the 12volt EPS cable (includidng the VRM powerloss), but this is not PPT power limit
Not only that, R23 for their 9950X was around 256W vs 217W for 7950X for a score of 41.6K for the 9950X vs 36827 for the 7950X. The 9950X is not what HWinfo64 shows as PPT for default operation, but the R23 score is. What gives?
Default PPT power limit for the 9950X is 200w
I would know.. have already binned a few
The only way i could see ~250w being right, is if they are messuring from the 12volt EPS cable (includidng the VRM powerloss), but this is not PPT power limit
Their 7950X doesnt perform accordingly, at stock Computerbase got 38649 pts in CB R23 for about 205-215W, 5% better than this one that is eventualy subject to some thermal thorttling wich explain the lower perf and comsumption, btw at stock Computerbase got 189W in Handbrake despite being at the helm for this bench, actualy the 7950X never reach it s 230W limit.They are measuring at the cable, but that still doesn’t explain why the Zen 4 CB power numbers are significantly lower. At worst, they should be equal.
As for the 7950X CB23 MT performance, according to the reviewer that processor gets limited by cooling before it gets limited by its PPT, in that test. The review methodics is locked on using Noctua NH-14S air cooler for all processors, so you could get better results with cooler that manages to get the temperature of 7950X down (good AIO?). The changes in Zen 5 cooling (or temperature reporting...) enable the 9950X to not be temperature-limited with the same cooler though.Their 7950X doesnt perform accordingly, at stock Computerbase got 38649 pts in CB R23 for about 205-215W, 5% better than this one that is eventualy subject to some thermal thorttling wich explain the lower perf and comsumption, btw at stock Computerbase got 189W in Handbrake despite being at the helm for this bench, actualy the 7950X never reach it s 230W limit.
Default PPT power limit for the 9950X is 200w
I would know.. have already binned a few
The only way i could see ~250w being right, is if they are messuring from the 12volt EPS cable (includidng the VRM powerloss), but this is not PPT power limit
WOW!
After hearing rumors about newest beta agesa 1.2.0.2 improving both performance and latency i decided to put it to the test!
I downloaded and flashed to newest ASUS bios 2401 and ran capframeX core-to-core latency
And wow indeed, the wispers were true, with agesa 1.2.0.2 AMD have finally fixed the core-to-core latency plaguing the 9000 from launch
This is with agesa 1.2.0.1A bios 2303 = cross CCD latency is ~180ns
View attachment 107682
Now with agesa 1.2.0.2 bios 2401 = cross CCD latency is ~75ns
View attachment 107683
Otherwise 100% same settings used for both runs, will be interesting to see how this affect the gaming performance / latency bound benchmarks
I'm rebenching 7000 series for hwbot atm while i wait for new delid frame for my delidded 9950X, so it have to wait.. But 7zip is first on my list when i get my Zen5 system put back together.Can you do some test like CPU-z Bench, AIDA64 GPGPU (this test tells a lot about the performance difference because it uses various calculations, I have results from about 3 months ago on 9950X IE to compare)?
There are 2 Polish publications that reported the same in few days after 9900X / 9950X launch so before the AGESA update. But most media outlets at the time were more concerned with special admin accounts, etc so this thing was forgotten. Now if they retest I guess it will be attributed to inter-ccd latency fixI'm rebenching 7000 series for hwbot atm while i wait for new delid frame for my delidded 9950X, so it have to wait.. But 7zip is first on my list when i get my Zen5 system put back together.
But i can share one other little thing interesting thing.
Pre agesa 1.2.0.2 it was recommneded to use the "X3D core parking" meta for Zen5 dual CCD cpus, for best gaming performance.
Now with fixed cross CCD latency fixed ob agesa 1.2.0.2 you actually get higher performance with not using the core parking (high performance powerplan)
Following screenshots belong to other guy from overclock.net forum, so this will only serve a preview:
CCD1 parked -> game is running on CCD0 only (balanced powerplan)
View attachment 107758
Core parking disabled -> both CCD enabled (high performance powerplan)
View attachment 107759
View attachment 107760
He will share more games when he is done with testing
And he have made his findings public 👍I'm rebenching 7000 series for hwbot atm while i wait for new delid frame for my delidded 9950X, so it have to wait.. But 7zip is first on my list when i get my Zen5 system put back together.
But i can share one other little thing interesting thing.
Pre agesa 1.2.0.2 it was recommneded to use the "X3D core parking" meta for Zen5 dual CCD cpus, for best gaming performance.
Now with fixed cross CCD latency fixed ob agesa 1.2.0.2 you actually get higher performance with not using the core parking (high performance powerplan)
Following screenshots belong to other guy from overclock.net forum, so this will only serve a preview:
CCD1 parked -> game is running on CCD0 only (balanced powerplan)
View attachment 107758
Core parking disabled -> both CCD enabled (high performance powerplan)
View attachment 107759
He will share more games when he is done with testing
Apart from WUKONG which seems to be the same or within margin of error, the all core approach now shows better results, then what was previously considered to be the best (parking ccd1). All the other tests were showing quite an uplift in the CPU and overall scores.
but did he even tried to test before core parking on/off before bios update?And he have made his findings public 👍
[Official] Zen 5 Owner's Club (9600X / 9700X /...
www.overclock.net
Yes, he even wrote this in the linked post abovebut did he even tried to test before core parking on/off before bios update?
This is the third time you are trying to downplay the effect of this new agesa i believe, any particual reason for that or ?So far up until now, we have seen that parking (even though mostly for x3d chips) was also beneficial on this new double CCD CPU. Previously I had worse results when using the unparked approach.
Wth is up with those SC2 framerates? It's 2024 and top end CPU can't run that game faster than 100 FPS?Found Starcraft 2 test. Zen5 have advantage over Zen4 non X3D
Also Escape from Tarkov
There seems mass battle. Game also have billion zergs campaign mission, where CPU pretty heavily used. FPS massive drop.Do they have some crazy custom map benchmark or something?
seems DX9Still running DirectX 9 too?