Info AMD confirms Windows 11 slow down its CPUs up to 15%

[deasd]

AMD processors hit by performance issues on Windows 11, patch on the way

AMD has said that its CPUs are taking a performance hit on Windows 11 systems and that it has teams working on the issues. Patches should be released later this month through Windows.

www.neowin.net

AMD: Windows 11 Slows Our CPUs Up To 15%, Patch Coming

High latency bug hits fast

www.tomshardware.com

AMD: Windows 11 May Cause Performance Dips Of Up To 15% On Ryzen CPUs

AMD reported issues that their Windows 11 compatible AMD Ryze CPUs were having with performance while running certain applications.

wccftech.com

The issue mentioned in the opening paragraph is caused by functional L3 cache latency which has increased by around three times on affected hardware. The programs that will suffer include those sensitive to memory subsystem access times. AMD noted another problem too, explaining that UEFI CPPC2 may not schedule threads on the processor’s fastest core preferentially.


Regarding the latter issue, applications sensitive to the performance of one or a few CPU threads will see a performance hit. The issue will be more noticeable on greater than 8-core processors that operate at over 65W. This issue should also be fixed this month.

hmmmm, whose fault is it this time? It screw up the future hardware review if true? what about Intel side?

 

[jpiniero]

anyway, if its nonfixable isnt it just a matter of time before its backported to 10?

AMD says they are working on a fix, but what that means we will have to see. 10 could be affected eventually.

 

[deasd]

Isn’t this just a benchmark reporting issue? How could an OS have such an effect on CPU internals?

I don't know, maybe we should wait until some other real-world test come out for comparison... since some gaming tests already showed a noticeable fps down. Theoretically IF this is just a report issue it should not lead to performance regression in other tests.

Another example of Vermeer's L3 being poorly reported in Win11:

 

[Zucker2k]

For example, if you take into account up to 15% worse(L3 Cache bug) performance + this comparison.

https://twitter.com/x/status/1446742599193141253

Maybe AMD could license thread director from Intel? Should help. Just thinking out loud.

What's interesting is AMD is silent on the issue while forum warriors are having fits. AMD has even promised a "fix" yet that has gone largely ignored. Why blame AMD when you can blame Intel? Lol

 

[Fanatical Meat]

Typically I am in the early adopter club regarding windows. Going to wait longer before I change to windows 11.
Seems like too many performance issues and not much gain other than better protection vs ransomeware, which is a good thing for a corporate environment not so much value for my home machine. Even in the event I got infected I could easily get my stuff back from a backup.

 

[Joe NYC]

L3 performance collapsed when switch from win10 to win11:
Read from 898GB/s to 136GB/s (-85%)
Write from 565GB/s to 51GB/s (-91%)
Copy from 701GB/s to 66GB/s (-91%)
Latency from 11.2ns to 32.1ns (-187%)

if future patches doesn't work effectively the AMD processors would be condemned to death at win11

View attachment 51143

What an interesting coincidence, that Windows 11, at the same time is optimized for Alder Lake and sabotages the performance of AMD's strongest feature - L3 - for current and upcoming processors..

 

[Hans de Vries]

L3 performance collapsed when switch from win10 to win11:
Read from 898GB/s to 136GB/s (-85%)
Write from 565GB/s to 51GB/s (-91%)
Copy from 701GB/s to 66GB/s (-91%)
Latency from 11.2ns to 32.1ns (-187%)

if future patches doesn't work effectively the AMD processors would be condemned to death at win11

View attachment 51143

It looks like the L3 cache is hardly used here as L3 bandwidth slows down to main memory speeds and L3 latency is more than 50% of main memory latency. The same happens on the 5800X with a single CCD.

Also keep in mind that he Linux scheduler is way better than the Windows 10 one. What happens here with Windows 11 is extreme by any measure.

Maybe the result of some extreme Core hopping? The L3 fills up after the L2 overflows via victim caching in L3. This and some cache policy settings in combination with cache/TLB flushing/invalidating?

More like a "thread derailor" instead of a "thread director" Blaming AMD here looks like victim blaming to me. Learn from the open source Linux scheduler, which does much better than Windows 10 with less information.

The fix promised for this month is probably just getting rid of this 'Intel thread director' in the AMD case, so that we have the Windows 10 performance back again.

 

[jpiniero]

More like a "thread derailor" instead of a "thread director" Blaming AMD here looks like victim blaming to me. Learn from the open source Linux scheduler, which does much better than Windows 10 with less information.

Until AMD gives any kind of guidance on what's actually going on it's just speculation about what is causing this. You can certainly blame AMD for not having the real fix in time for the Windows 11 launch.

 

[Zepp]

2k: 👍
XP: 👍
Vista: 👎
7: 👍
8: 👎
10: 👍
11: 👎

I'd rate them like this:

2k: 👍
Millennium: 👎👎
XP: 👍👍
Vista: 👎
7: 👍
8: 👎
8.1: 👍👍
10: 👎
11: 👎

I think 10 was worse than vista or 8.0 when it first launched.
7 was like Vista.1 and 11 seems like 10.1 only not an improvement

 

[Fanatical Meat]

for me it would look like

2k: 👍
Millennium: 👎👎
XP: 👍👍
Vista: 👎
7: 👍
8: 👎
8.1: 👍👍
10: 👎
11: 👎

I even think 10 was worse than vista or 8.0 when it first launched

While I had zero problems with windows 8/8.1 I found windows 10 better all around.

 

[BTRY B 529th FA BN]

AMD "hey our new 3D Vcache chips will increase game performance up to 15%"

MS "hold my beer"

 

[Insert_Nickname]

I'd rate them like this:

2kSP4: 👍👍👍
Millennium: 👎👎👎
XP (RTM): 👎
XP (SP2): 👍
XP (SP3): 👍👍
Vista (RTM):👎
Vista (SP2): 👍
7: 👍
8: 👎
8.1: 👍
10: 👍
11: 👎👎

 

[DigDog]

Enterprise LTSC (Long-Term Servicing Channel) (formerly LTSB (Long-Term Servicing Branch))
..is a long-term support variant of Windows 10 Enterprise released every 2 to 3 years. Each release is supported with security updates for either 5 or 10 years after its release, and intentionally receive no feature updates.[11] Some features, including the Microsoft Store and bundled apps, are not included in this edition.[12][1][3] This edition was first released as Windows 10 Enterprise LTSB (Long-Term Servicing Branch).[13] There are currently 3 releases of LTSC: one in 2015 (version 1507), one in 2016 (version 1607), and one in 2018 (labeled as 2019, version 1809).[14] "

small thumbs up icon goes here

 

[UsandThem]

AMD "hey our new 3D Vcache chips will increase game performance up to 15%"

MS "hold my scheduler"

FIFY

 

[Hans de Vries]

Until AMD gives any kind of guidance on what's actually going on it's just speculation about what is causing this. You can certainly blame AMD for not having the real fix in time for the Windows 11 launch.

AMD gave this guidance: https://www.amd.com/en/support/kb/faq/pa-400

One of the two problems is with 100% certainty scheduler related. The other (L3) most likely as well IMHO.

Does AMD even has access to Windows code containing proprietary Intel Thread Director source code?
Is AMD allowed to modify that source code in order to experiment with it to fix it?

AMD depends on Microsoft's top management to act after a certain deadline passes. Apparently that deadline has now passed.

 

[jpiniero]

I'd assume that any scheduler changes related to Thread Detector only runs on Alder Lake. It would be silly otherwise.

 

[BorisTheBlade82]

The question remains why MS not just uses the Win10 scheduler logic depending on the identified CPUID. That is no rocket science and it is not as if people were changing their CPUs once a second

 

[moinmoin]

The Windows' scheduler is unadulterated rubbish, part 32789.

 

[|JosephB17331|]

I'd rate them like this:

2k: 👍
Millennium: 👎👎
XP: 👍👍
Vista: 👎
7: 👍
8: 👎
8.1: 👍👍
10: 👎
11: 👎

I think 10 was worse than vista or 8.0 when it first launched.
7 was like Vista.1 and 11 seems like 10.1 only not an improvement

Put Windows 10 LTSC in the slot between 10 and 11 with a thumbs up 👍

 

[Jaskalas]

I dunno dude. This entire OS seems like a rushed release. Heck, the updated paint doesn't even have a dark mode. Let me repeat that. Their own app doesn't support the theme in their own OS. Just crazy.

That's just Microsoft.
Did you see how they handled Skype over the years? Important options REMOVED after an update. You appear to be describing more of the same.

 

[Jaskalas]

Typically I am in the early adopter club regarding windows.

I believe as of Windows 10 their QA department was, uh.... downsized. Let's just say the last update I volunteered for was followed by a system restore. Test at your own risk. Early adoption is not risk free, not anymore.

 

[Geegeeoh]

Early adoption is not risk free, not anymore.

Never was.

 

[LikeLinus]

AMD gave this guidance: https://www.amd.com/en/support/kb/faq/pa-400

One of the two problems is with 100% certainty scheduler related. The other (L3) most likely as well IMHO.

Does AMD even has access to Windows code containing proprietary Intel Thread Director source code?
Is AMD allowed to modify that source code in order to experiment with it to fix it?

AMD depends on Microsoft's top management to act after a certain deadline passes. Apparently that deadline has now passed.

Did AMD not do throughout testing on Windows 11 before the release and never told the public until after it was launched and others found out?

Both companies are to blame here. AMD can't plead innocence if they weren't doing their due diligence and testing a new OS along the way.

No one really knows how things transpired. But it's not like MS hid Win11 from them and they didn't have the ability to test and raise the red flag to MS and consumers earlier.

 

[BorisTheBlade82]

Did AMD not do throughout testing on Windows 11 before the release and never told the public until after it was launched and others found out?

Again: AMD very probably was under the impression that the fix was included in the RTM! That it was not was Microsofts decision. They could have simply included it or otherwise delayed the release. They did not. And for good reason: Win11 is vital for ADL and new hardware sells a new OS.

 

[leoneazzurro]

And what if AMD fond the issue, communicated it to MS and MS decided to launch Win 11 neverthless? It seems that the fix was present on developer releases before AMD spoke about it, that is, before Win 11 was launched.

 

[jpiniero]

And what if AMD fond the issue, communicated it to MS and MS decided to launch Win 11 neverthless? It seems that the fix was present on developer releases before AMD spoke about it, that is, before Win 11 was launched.

Is the patch actually available in some form? I haven't seen anything suggesting that.