Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[Kepler_L2]

It appears that these updates are possibly fixing an issue present in some systems that is not present in other systems, even though they may be running an older version of Windows. So now you have multiple angles of what is causing the uplift and/or the slowdown (depending on your viewpoint). Whats unclear still is, what is the problem and what is the fix? So far:

1.) Windows 11 "Super" Admin - helps fps, but not wise to do
2.) Disable VBS - helps fps, but not wise to do.
3.)Windows 11 2H24 update: Helps fps for some systems, does nothing for others
4.)Windows 11 2H23 branch prediction update: Still waiting on results?

Steve from HUB noted many months ago that he saw lower performance when upgrading from an older Zen generation to a newer one on the same OS install vs a fresh install.

My guess is that until this 24H2 update Windows was incorrectly applying security mitigations based on the original CPU your system was installed with, rather than your current CPU. So newer CPUs get more performance (due to HW-level security fixes) but results will vary depending on how the pre-24H2 OS was setup.

 

[Aeonsim]

Had a quick look at a few tests last night when switching a couple of machines from current release to the windows insider 24H2.

On a 7900 (running at 1080P with a 4070):
Homeworld 3 - +15% (min and average frame rate)
Cinebench 2024 - +8%
Own HPC app - +5%
Cyberpunk 2077 - +5%
Black Myth - 0%

With a 3600:
Homeworld 3 - +10%
Cyberpunk - 0%
Black Myth - 0%

It seems there is something there; for my HPC app, I've run it 20-30x over the last few months and switching to the Windows version beat the best previous result by ~5% in several consecutive runs. The rest were pretty much the best of three runs for benchmarks at lower settings than I'd play the game, so suggestive, but I'd not call it definitive.

This is with VBS and all the other security and virtualization stuff on (for WSL).

 

[igor_kavinski]

24H2 probably has buggy kernel code according to this: https://exploits.forsale/24h2-nt-exploit/

Microsoft, please don't tell me that our games will show different results AGAIN after yet more security fixes!

EDIT: ok maybe it's no issue. Github link mentions these exploits were in the pre-26100 builds. Maybe it's fixed by now?

 

[mostwanted002]

Can't wait for new Rust kernel, now with exploits baked inside! /s

 

[fastandfurious6]

AMD Branch Prediction Optimization For Ryzen 9000 “Zen 5” & Ryzen 7000 “Zen 4” CPUs Now Available In Windows 11 23H2​

AMD Branch Prediction Optimization For Ryzen 9000 "Zen 5" & Ryzen 7000 "Zen 4" CPUs Now Available In Windows 11 23H2

AMD's branch prediction optimizations for Ryzen 9000 & Ryzen 7000 CPUs are now available in Windows 11 23H2 to improve gaming performance.

wccftech.com

finally a windows update KB that actually does something wow

 

[MarkPost]

So I've updated 24H2 to the latest build, and now I see a general improvement compared to 21H2. For example, Geekbench 6 (+5% MT):

Gigabyte Technology Co., Ltd. B650M AORUS ELITE AX ICE vs Gigabyte Technology Co., Ltd. B650M AORUS ELITE AX ICE - Geekbench

 

[SK10H]

finally a windows update KB that actually does something wow

Did some testing on a 7800x3d USB Sata BX100 Win11 23H2 ToGo with 7900xtx 24.3.1, on non built-in user account with local admin. 3rd party anti virus VBS off.
1) August 2024 patch
2) August 2024 preview patch KB5041587
3) August 2024 preview patch KB5041587 + disable Spectre v2
No difference on selected apps.
PYPrime, 7zip, GB6, Shadow of Tomb Raider built-in benchmark 1080p medium custom ~350fps on all run.




Going to stick to the same Win10 install that has gone through Zen1-Zen4 without a format, nv/amd gpu switch, and SSD cloning on different Sata/nvme gen3/4 SSD for 8 yrs. Will try the 24H2 just for testing purpose on the USB SSD.

 

[fastandfurious6]

entirely anecdotal.... but I installed the +11% Fine Wine KB on phoenix (7840hs) 23H2 and some Windows stuff is ultra snappy now

i.e. loading up things instantly like some webpages or even file explorer, very noticeable difference

AMD the gift that keeps on giving~!

 

[lucasworais]

I know this thread is zen 5 related but has anyone seen improvements to Zen 3? I couldn’t find any tests

 

[Mahboi]

Random thought, but since David Huang said that Zen 5c basically has no advantage over Zen 5 past 2Ghz since it's such a gutted small core, I wonder what kind of a 4-6-8 core small solution you could do with it?
What kind of tiny idle power?
What kind of max power?
How much could you get out of say something like Strix Point's 8 Z5c CCD all plugged to 32Go of RAM for basic mini server ops and how good would it be watt/perf and latency wise?

 

[Mahboi]

I know this thread is zen 5 related but has anyone seen improvements to Zen 3? I couldn’t find any tests

We saw massive Zen 4 improvements which is for all the tweaking and extra transistors, still Zen 3 with better frequency and some tweaking.
If Zen 4 explodes, Zen 3 does too.

 

[LightningZ71]

Has anyone seen if any of the reviewers or anyone on the forums has measured processor power usage with the Branch Predictor patch? I suspect that it will provide at least some marginal amount of efficiency improvement as the processor is less likely to mispredict a branch, do a bunch of work that it then needs to discard to go back and do the work that it was originally supposed to do. I don't expect it to be night and day, but perhaps measurable? It would also be highly program dependent given that some are less branchy than others and some are more predictable as well.

 

[Josh128]

Over at Guru3d.com forums, its about 50/50 as to whether or not users get a boost from either the update or newest Windows build. WTH is going on?

Vananovion
5 hours ago
HUB testing showed quite nice gaming performance gains for both Ryzen 4 and 5. Wonder if it's gonna do anything for my 5900. Will benchmark a couple of games.

#6259445

mackintosh
5 hours ago
Already tested my 5900X, didn’t see any performance gains in settings that I normally use in my games. Nothingburger for Zen 3 it seems.

#6259450

Neo Cyrus
4 hours ago

mackintosh:

Already tested my 5900X, didn’t see any performance gains in settings that I normally use in my games. Nothingburger for Zen 3 it seems.

I'm getting the same uplift on a 5800X3D that people are reporting on Zen 4 and 5. So either you don't have the update, or just your settings/what you tested didn't benefit. But that's on 24H2, I already updated to that before these changes were supposedly added to 23H2.

#6259452

vestibule
4 hours ago
Updated to KB5041587.

#6259456

mackintosh
4 hours ago

Neo Cyrus:

I'm getting the same uplift on a 5800X3D that people are reporting on Zen 4 and 5. So either you don't have the update, or just your settings/what you tested didn't benefit. But that's on 24H2, I already updated to that before these changes were supposedly added to 23H2.

Dunno what to tell you. Updated as soon as it went live, tested my usual gameplay in Cyberpunk, RDR2, Forbidden West and the Wukong benchmark, so all the games where significant gains were reported and got exactly zero uplift. Perhaps if I ran at low settings there would have been something there, but that's of no use for me in a real-world scenario 🙂 I guess we'll see when others report in.

#6259464

Ryu5uzaku
3 hours ago

mackintosh:

Dunno what to tell you. Updated as soon as it went live, tested my usual gameplay in Cyberpunk, RDR2, Forbidden West and the Wukong benchmark, so all the games where significant gains were reported and got exactly zero uplift. Perhaps if I ran at low settings there would have been something there, but that's of no use for me in a real-world scenario 🙂 I guess we'll see when others report in.

I need to test the update. But need to disable admin mode first. As that gave me tangible boost. I wonder if 24H2 has some extra apart from the branch prediction. Tho the games you mentioned I doubt I would get any boost even if there was some with my 6800 XT.

#6259467

mackintosh
3 hours ago
Maybe 24H2 does something different indeed. I simply installed the update that was released by MS yesterday. Didn't make any other changes to my system. It's also entirely possible that it's something on my end. This is my main rig, it's full of stuff running in the background all the time. This is why I keep hoping the usual suspects will eventually do a proper, comprehensive retest across the board.

#6259474

undorich
3 hours ago
for me a massive uplift in performance, got from 6600 in 3dmark cpu test to 6900, with stock settings, before i had to rise mem speeds and do curve optm. for the same results, ca. 6900, now it runs without any tweaks the same points. nice ! will test with tweaks later on.

#6259488

bnauk
2 hours ago
To those seeing no performance change, you might be graphics card limited

#6259498

vestibule
2 hours ago
I discovered today that if a game features the NVidia low latency button and that it is enabled that all the setting may not transfer over enabled when switching to the administrator account even though they show that they are enabled. Got that. 😛 So when testing games on the administrator account disable the low latency button in game. Just to be safe reset the game setting you want and re-run the game. Other than that my games run the same form user to admin account no real difference for me worth talking about. 🙂

#6259507

gaz41
One hour ago
Updated to KB5041587 on Windows 11. Didn’t see any performance gains on games that I tested (Cyberpunk 2077, Wukong, Shadow of the Tomb Rider, TimeSpy Extreme, AC Mirage, Warhammer III). Ryzen 7 7800 X3D, 4090 FE, 32 GB ram 6000 mhz.

#6259509

Horus-Anhur
57 minutes ago
In Remnant 2, I went from around 90 fps, to around 100 fps, in the home hub. I'm using 24H2.

#6259511

joshk
39 minutes ago
KB5041587 Probably not working as intended. No perf gains.

 

[PJVol]

Over at Guru3d.com forums, its about 50/50

It's so called "50/50 patch" - a brand new concept in software development.
Together we innovate ! (tm)

 

[Abwx]

Over at Guru3d.com forums, its about 50/50 as to whether or not users get a boost from either the update or newest Windows build. WTH is going on?

Hardewareluxx made a revision of their revised benches, and still, they re currently updating it with yet more revisions of the revised results.

https://twitter.com/x/status/1828767155351167300

Optimierte Sprungvorhersage: Zen 5 mit Windows-Update KB5041587 getestet - Hardwareluxx

Optimierte Sprungvorhersage: Zen 5 mit Windows 11 24H2 getestet.

www.hardwareluxx.de

 

[Josh128]

Hardewareluxx made a revision of their revised benches, and still, they re currently updating it with yet more revisions of the revised results.

https://twitter.com/x/status/1828767155351167300

Optimierte Sprungvorhersage: Zen 5 mit Windows-Update KB5041587 getestet - Hardwareluxx

Optimierte Sprungvorhersage: Zen 5 mit Windows 11 24H2 getestet.

www.hardwareluxx.de

Excellent, I read the update with the KB5041587 instead of the 2H24 (or is it both or with 2H23?) and the saga continues...very little to no regressions this time, which is a bit more comforting...still, how odd is it that none of this shows in any synthetic benchmark and further still, that not everyone who uses the patch is seeing the uplifts in the same games?

 

[StefanR5R]

how odd is it that none of this shows in any synthetic benchmark

Perhaps try a synthetic benchmark which performs many syscalls/ involves device drivers...?

(Common game tests aren't CPU performance tests. They are computer system performance tests/ application framework a.k.a. game engine–OS kernel–device driver performance tests.)

 

[KompuKare]

Over on HardwareLuxx.de, Andreas Schilling has been busy:
https://www.hardwareluxx.de/index.p...rsage-zen-5-mit-windows-11-24h2-getestet.html

Also tested KB5041587 on Win 11 23H2 and Zen 3 3D.
Yes Ryzen 5800X3D sees gains. @lucasworais
In fact, possible the biggest gains if my spreadsheet is correct (looking at their min scores - not the less useful avg FPS):

Actually excluded the Intels has there was no with and without Patch, if I do:

If even a 9600X can outperform the i9-14900K while using a fraction of the power? Well Intel might as well go home? Certainly explains why they cranked up the power so much with the 14th gen.

So does anyone know of any very branchy applications?

 

[igor_kavinski]

Over on HardwareLuxx.de, Andreas Schilling has been busy:
https://www.hardwareluxx.de/index.p...rsage-zen-5-mit-windows-11-24h2-getestet.html

Bless his dear soul. He is certainly doing God's work.

 

[StefanR5R]

So does anyone know of any very branchy applications?

I guess it is not so much about branchy applications, but applications in which Windows used to clear the BTB often.

Well Intel might as well go home?

Below which min FPS figure should they go home?

 

[Josh128]

Here we go....

AMD To Provide Patch That Improves Ryzen 9000 "Zen 5" Inter-Core & Chiplet Latency

AMD is reportedly working on a new BIOS patch that will help improve inter-core latency on Ryzen 9000 "Zen 5" Desktop CPUs.

wccftech.com

 

[KompuKare]

Opps, wrong column in my spreadsheet that was the Avg FPS not the Min FPS.
Here is the min instead:

 

[igor_kavinski]

So does anyone know of any very branchy applications?

Read in some thread here that Stockfish chess engine is branchy as heck.

Download and just pass bench as the parameter and it will quickly give you nodes/second.

See bench under non-standard commands for more options if you want to make it more stressful: https://github.com/official-stockfish/Stockfish/wiki/UCI-&-Commands

On my office i7-4770:

x86_64
Nodes searched : 1303971
Nodes/second : 357840

AVX2
Nodes searched : 1303971
Nodes/second : 593523

65.86% speedup from AVX2 usage.

 

[SK10H]

Did a wipe on the USB SSD and install Win11 24H2 insider preview 26100.1591 ToGo. Wait until Windows is idle long enough.

PYPrime is a bit worst off to the upside, ran a few more times in a bias way in favour of 24H2 actually and still only got it to match the previous result once.
Same score in GB6/Tomb Raider.
7Zip Compress is 1-3% better consistently, decompress is the same.

Not that big of a deal for 7800x3d on these apps.

 

[Ranulf]

Hmm, so maybe a 10-15% uplift for my 5900X? Clearly a MS scam to get me to upgrade to win11 early for that system.