Windows 11 Performance Issues on Ryzen Fixed by Updates from Microsoft and AMD

[btarunr]

Microsoft and AMD on Thursday released software updates that fix the two performance issues affecting AMD Ryzen processors with Windows 11. The two issues were abnormally high L3 cache latency, and a broken "Preferred Cores" system. The companies had assessed that the issues impact performance of Ryzen processors on Windows 11 by as much as 15%.

The two issues are fixed in separate methods. The L3 cache latency bug is improved through a Windows Update patch, which has been released now as an Update Preview (an Update Preview is not a "beta," but a software update released ahead of its designated "patch Tuesday"). The Update Preview is chronicled under KB5006746, and Windows 11 systems updated with this, get their OS build version set as "build 22000.282." The next update restores the Preferred Cores mechanism that leverages UEFI-CPPC2. This update comes in the form of an AMD Chipset Software update. You'll need to download and install both of the following:

DOWNLOAD: Windows 11 October 21, 2021 Update Preview KB5006746 | AMD Chipset Driver Software 3.10.08.506



View at TechPowerUp Main Site

 

[Flanker]

Well there goes the conspiracy theories

 

[outpt]

No bitching. What’s the world coming to. Oh the agony.

 

[Zubasa]

Well there goes the conspiracy theories

Well the L3 bandwidth is still FUBAR.
From 1TB/s on Win10 to around 400GB/s on Win11 even after the patch.

 

[InhaleOblivion]

Well the L3 bandwidth is still FUBAR.
From 1TB/s on Win10 to around 400GB/s on Win11 even after the patch.
View attachment 221852

Appreciate the heads up. Still sticking with Win10 until this is fully resolved. Not a fan of performance regressions on hardware for the sake of new software.

 

[Raiden85]

Appreciate the heads up. Still sticking with Win10 until this is fully resolved. Not a fan of performance regressions on hardware for the sake of new software.

I've updated to the latest MS L3 patch and installed the latest AMD chipset drivers, speed is looking good here.

 

[prtskg]

I've updated to the latest MS L3 patch and installed the latest AMD chipset drivers, speed is looking good here.

What was your latency in W10? Same, slightly better or slightly behind?

 

[The King]

Well the L3 bandwidth is still FUBAR.
From 1TB/s on Win10 to around 400GB/s on Win11 even after the patch.
View attachment 221852

Did you install the latest AMD chipset driver after installing the Windows 11 patch?

 

[Raiden85]

What was your latency in W10? Same, slightly better or slightly behind?

Pretty much identical on everything from what I remember. RAM latency has always been a bit crap but that was on 10 as well. Running 4x 16GB sticks so that probably doesn't help.

 

[Zubasa]

Did you install the latest AMD chipset driver after installing the Windows 11 patch?

I did before hand, also Zen1 doesn't use CPPC and thus not affected by the other CPPC bug in the first place.
I find it ironic that the CPUs that Microsoft considers E-waste are the ones less affected.

 

[Raiden85]

Did you install the latest AMD chipset driver after installing the Windows 11 patch?

Installed the MS patch first and AMD driver straight after without rebooting after installing the MS patch.

 

[SLObinger]

I also applied both MSFT and AMD updates today for my 5900x on an X570 Asus Dark Hero MB. Latency is improved but still slightly above W10. W11 has definitely had an negatove impact on what I am able to achieve with PBO however. Much better results on W10. There is still some work to be done to make W11 a good idea for AMD.

 

[R-T-B]

Well the L3 bandwidth is still FUBAR.
From 1TB/s on Win10 to around 400GB/s on Win11 even after the patch.
View attachment 221852

Yeah it's far from fully resolved. Conspiracy unsolved.

 

[The King]

I did before hand, also Zen1 doesn't use CPPC and thus not affected by the other CPPC bug in the first place.
I find it ironic that the CPUs that Microsoft considers E-waste are the ones less affected.
View attachment 221857

I have a 1700X and noticed slightly improved L3 cache performance after installing the latest AMD chipset driver.
32ns on your 5950X is way off. My 1700X L3 is 10.3ns at 3.9Ghz.

 

[Zubasa]

I have a 1700X and noticed slightly improved L3 cache performance after installing the latest AMD chipset driver.
32ns on your 5950X is way off. My 1700X L3 is 10.3ns at 3.9Ghz.

I think you miss read my post a bit, that result of the 5950X from the first post is not mine. It was just a pic from another review site.
Notice the CPU I am actually using is the 1950X and 10.2ns is inline with Zen1 latency, but both results are still slightly slower than Win10.

 

[Mussels]

Not reposting my scores yet again, but my results are all back to normal with the W11 update and the AMD driver installed.

Things are not *perfect* but they are massively better.

 

[bug]

Ha, I called bad driver for AMD! I was partially right.

 

[halodies]

My 5900x L3 numbers are back to Windows 10 time, the fix is working perfectly.

 

[HD64G]

Ha, I called bad driver for AMD! I was partially right.

How was the AMD driver bad? The "broken one" works well on Win10.The only thing changed was the MS software between 10 and 11 that broke something and forced AMD to correct something that worked well on another cpu scheduler on the same OS kernel. Win11 is a major update to Win10. Don't forget that.

 

[Metroid]

It fixed for me. I can't see how it was on windows 10 though. This aida64 benchmark is on windows 11 after the fix, before the fix my 5900x could not even boot up to 4.6ghz, now at 4.9ghz no problem, cpu all default.

 

[LocutusH]

I dont see any improvement either.
4900HS laptop with Win11. Updated windows, and driver too.

 

[Metroid]

I dont see any improvement either.
4900HS laptop with Win11. Updated windows, and driver too.

I think this issue is only on zen 3 vermeer 5xxx series.

 

[bug]

How was the AMD driver bad? The "broken one" works well on Win10.The only thing changed was the MS software between 10 and 11 that broke something and forced AMD to correct something that worked well on another cpu scheduler on the same OS kernel. Win11 is a major update to Win10. Don't forget that.

So you're telling me AMD simply repackaged their Win10 driver and called it day?
Something was obviously amiss, otherwise AMD didn't have to issue a new driver.

Raises the question of what kind of testing AMD (and Microsoft) did if they didn't notice that the preferred core detection wasn't working right. Makes you wonder what else they missed.

 

[roberto888]

Before and after.

 

[W4llBangdMyself]

The update seems to have fixed it for the most part, but I still have very slightly higher L3 latency and lower L3 transfer rates on Win 11 than I had on Win10.
The Win10 result is from May and is using older BIOS and AGESA but I don't think that's related. BIOS settings are all the same and I don't have anything overclocked, well, apart from PBO and XMP being enabled

Win10 (May, Agesa 1.2.0.2)

Win11 (yesterday, Agesa 1.2.0.3b)