Tuesday, October 12th 2021

First Windows 11 Patch Tuesday Makes Ryzen L3 Cache Latency Worse, AMD Puts Out Fix Dates

Microsoft on October 12 put out the first Cumulative Updates for the new Windows 11 operating system, since its October 5 release. The company's monthly update packages for Windows are unofficially dubbed "patch Tuesday" updates, as they're scheduled to come out on the second Tuesday of each month. Shortly after Windows 11 launch, AMD and Microsoft jointly discovered that Windows 11 is poorly optimized for AMD Ryzen processors, which see significantly increased L3 cache latency, and the UEFI-CPPC2 (preferred cores mechanism) rendered not working. In our own testing, a Ryzen 7 2700X "Pinnacle Ridge" processor, which typically posts an L3 cache latency of 10 ns, was tested to show a latency of 17 ns. This was made much worse with the October 12 "patch Tuesday" update, driving up the latency to 31.9 ns.

AMD put out a statement on social media, which surfaced on Reddit. The company stated that patches for the two issues have been developed, and specified dates on which they'll be released. The patch for the Preferred Cores (UEFI-CPPC2) bug will be released on October 21. Customers can request the patch even earlier. By "customers," AMD is probably referring to big enterprise customers running mission-critical applications on Threadripper or EPYC-powered workstations. The L3 cache latency bug will be fixed through the Windows Update channel, its release is scheduled for October 19.
If rumors surrounding the late-October/early-November launch dates of 12th Gen Intel Core "Alder Lake" processors are true, then the situation with these patches will have a direct impact on AMD. Processor reviewers will be compelled to use Windows 11 for their Core "Alder Lake" testing, as the new operating system supposedly has greater awareness of the heterogeneous core design. The switch to Windows 11 will force a re-bench of all processors, including the AMD Ryzen chips. With AMD cautioning of an up to 15% performance hit from the added cache latency and Preferred Cores bugs, results of AMD processors in 12th Gen Core launch reviews could be affected. It is advisable for AMD to reach out to the press with these patches immediately, if they are ready.
Source: destiny2sk (Reddit)
Add your own comment

157 Comments on First Windows 11 Patch Tuesday Makes Ryzen L3 Cache Latency Worse, AMD Puts Out Fix Dates

#126
zlobby
Selayathats on you then
i would only ever buy big cores but idt there's an 8/0 SKU?
so i get unwanted useless little cores shoved down my throat, that don't even want to play nice w/ all OSes so yeah, -> disable
You are both right. Little.big makes little to no sense on desktop and even on laptops. It's a good waste of silicon real estate. Only ultra portables, e.g. phones have non-marginal benefits.

I'm guessing now the next big thing would be who has the smallest cores.
Posted on Reply
#127
blunden
MusselsNo, really. Alder lake requires windows 11.

It's two different CPU architectures in one, Win 10 would throw crysis on the low performance core and minesweeper on the big ones

The fix is live on the beta channel

While much better, the Read, Write and Copy numbers for the L3 cache still seem pretty low to me. Even my currently stock 3900X easily beats those numbers on Windows 10 by quite a bit.

I think it needs further investigation. Maybe the difference is due to different AGESA versions, maybe it's due to the OS.

Posted on Reply
#128
Mussels
Freshwater Moderator
blundenWhile much better, the Read, Write and Copy numbers for the L3 cache still seem pretty low to me. Even my currently stock 3900X easily beats those numbers on Windows 10 by quite a bit.

I think it needs further investigation. Maybe the difference is due to different AGESA versions, maybe it's due to the OS.

my test was absolutely not optimised, i really should throw 10 on my other SSD
Posted on Reply
#129
blunden
Musselsmy test was absolutely not optimised, i really should throw 10 on my other SSD
Yeah, that is probably a good idea. Some methodical comparisons on the same hardware and settings would be much appreciated. Otherwise, it's hard to accurately compare besides seeing whether the latency has gotten back in check. I've seen others with similar numbers on Windows 11 though, so they don't stick out as low compared to others on that OS version.

I also saw someone mention that they saw the high latency return after a while after playing some games and some normal browsing. Might also be worth investing. :)
Posted on Reply
#130
Raven Rampkin
FireFoxDid anyone force you to buy AMD?
This goes surprisingly well with your status.
bugI will laugh so hard if at the end of the day, this all turns out to be yet another botched driver from AMD.
CPU driver, noice...
Posted on Reply
#131
Mussels
Freshwater Moderator
blundenYeah, that is probably a good idea. Some methodical comparisons on the same hardware and settings would be much appreciated. Otherwise, it's hard to accurately compare besides seeing whether the latency has gotten back in check. I've seen others with similar numbers on Windows 11 though, so they don't stick out as low compared to others on that OS version.

I also saw someone mention that they saw the high latency return after a while after playing some games and some normal browsing. Might also be worth investing. :)
I downloaded and compiled an update ISO last night to do a 10/11 dual boot

performance slows down because explorer has a memory leak - its slow, but some of us dont reboot.
Yes, 11 feels a little rushed... doesn't mean its terrible or unusable, at least the fixes are rolling out weekly.
Posted on Reply
#132
blunden
MusselsI downloaded and compiled an update ISO last night to do a 10/11 dual boot

performance slows down because explorer has a memory leak - its slow, but some of us dont reboot.
Yes, 11 feels a little rushed... doesn't mean its terrible or unusable, at least the fixes are rolling out weekly.
Great! I'm eagerly awaiting your results. :)

I'm also squarely in the "I don't reboot except when needed for OS updates" camp, so hearing that the memory leak is noticable after a while is valuable. I thought simply restarting explorer.exe temporarily "solved it" without a full reboot?

I wanted to install Windows 11 as soon as it was released, but was looking for indications that they would address this in the release build or with a day 1 patch. Sadly, that didn't end up being the case. I like to move to new OS versions early (except on work machines), and probably installed every single leaked build of Windows 7. In general, I've moved to the new Windows versions as soon as the RTM release dropped, long before official public release. :)

In other words, my interest in Ryzen + Windows 11 performance issues is simply because I want to confirm that they are fixed so I can make the jump. :D
Posted on Reply
#133
Mussels
Freshwater Moderator
Absolutely clean W10 install





And now the clean, everything quit W11 result



I see only margin of error stuff here.
Posted on Reply
#134
blunden
MusselsAbsolutely clean W10 install





And now the clean, everything quite W11 result



I see only margin of error stuff here.
Thank you! It does indeed appear as if the difference compared to my results is likely to be the result of other factors. Makes me curious what those factors are, but at least it seems to be unrelated to Windows 11. :)
Posted on Reply
#135
Mussels
Freshwater Moderator
Now i gotta work out why some people get L3 cache so far higher than mine... is it just my mere mortal 5800x not keeping up with the 12 and 16 core kings?
blundenThank you! It does indeed appear as if the difference compared to my results is likely to be the result of other factors. Makes me curious what those factors are, but at least it seems to be unrelated to Windows 11. :)
Heres something from before W11 was a thing, L3 cache problems can be related to BIOS, PBO, and core sleep settings (windows power plan??)
Low L3 cache bandwidth on AMD Ryzen 7 5800X - Benchmarking, system performance - AIDA64 Discussion Forum

Below be the rabbit hole

Uhhh i'm seeing weird things like "Set EDC to 500 and L3 cache results double"
This is one hell of a rabbit hole, if you find stuff before W11 was blamed

"i set an all core OC and my L3 cache result doubled"





These might be from a few AGESA updates ago, but they're still showing its a weird ongoing problem


And heres what happens if i max out PBO, and just farm the results for the fastest run:



What i'm seeing is L3 cache performance is all about single core clock speed
The moment i enable +200MHz in PBO, it jumps. The low scores are when its testing on SMT cores, it would seem.
Posted on Reply
#136
blunden
MusselsNow i gotta work out why some people get L3 cache so far higher than mine... is it just my mere mortal 5800x not keeping up with the 12 and 16 core kings?


Heres something from before W11 was a thing, L3 cache problems can be related to BIOS, PBO, and core sleep settings (windows power plan??)
Low L3 cache bandwidth on AMD Ryzen 7 5800X - Benchmarking, system performance - AIDA64 Discussion Forum

Below be the rabbit hole

Uhhh i'm seeing weird things like "Set EDC to 500 and L3 cache results double"
This is one hell of a rabbit hole, if you find stuff before W11 was blamed

"i set an all core OC and my L3 cache result doubled"





These might be from a few AGESA updates ago, but they're still showing its a weird ongoing problem


And heres what happens if i max out PBO, and just farm the results for the fastest run:



What i'm seeing is L3 cache performance is all about single core clock speed
The moment i enable +200MHz in PBO, it jumps. The low scores are when its testing on SMT cores, it would seem.
Yeah, it seems to vary a lot between different people's rigs. I've seen 3900X results close to yours, so it's not a simple 12 cores == better cache performance. Also, like I said, I'm running mine stock so my boost frequency is definitely lower than yours, even before BIOS tweaks. That's why I'm guessing it could be due to AGESA versions or possibly some other settings that happen to differ between different motherboards. The power plan could also make a difference, like you say. Does setting it to max performance make a difference for you? I'm still on the old Ryzen Balanced power plan I think (not at the computer at the moment), back from when I had an 1800X.

Can you confirm what AGESA version you're on?
Posted on Reply
#137
Mussels
Freshwater Moderator
AM4 AGESA V2 PI 1.2.0.3 Patch C
It seems based on single core clock speed , and ram latency. High core boost + C14 = best result.
(and this is why W10 gets higher results on average, W11 doesnt choose the preferred cores yet)
In W11 you can click and reclick and see what i assume is slowest core, fastest core, and SMT cores as you refresh it.

Here is my standard/average results from a 4GHz (4MT/s?) OC, still at C18 cause my ram is poo.
Posted on Reply
#138
blunden
MusselsAM4 AGESA V2 PI 1.2.0.3 Patch C
It seems based on single core clock speed , and ram latency. High core boost + C14 = best result.
(and this is why W10 gets higher results on average, W11 doesnt choose the preferred cores yet)
In W11 you can click and reclick and see what i assume is slowest core, fastest core, and SMT cores as you refresh it.

Here is my standard/average results from a 4GHz (4MT/s?) OC, still at C18 cause my ram is poo.
Cache performance being tied to core clock is reasonable since the cache has at least historically been running at the same frequency as the core clock, or at least traditionally been tied to it.

It will be interesting to see how the measurements change when the CPPC2 fix is released, hopefully today. Last I checked, Microsoft didn't hit the October 19 date rumored though. Let's hope AMD doesn't delay the release too. :)
Posted on Reply
#139
TomasK
OK so the fix for the preferred core issue is out via AMD Chipset Driver Software, now I wonder why is it taking Microsoft so long to push the update fixing the L3 cache latency to the non-beta channel... Maybe the longer they delay the more money Intel pays them?
AMD updated their KB page about those 2 Win11 problems today, including adding the link to the chipset driver
Windows® 11 Performance Variation in Certain Applications on Compatible AMD Processors | AMD

edit a few hours later:
OK so KB5006746 is out now, nice
Posted on Reply
#140
lexluthermiester
TomasKedit a few hours later:
OK so KB5006746 is out now, nice
That's just a preview page. There's nothing to download yet.
Posted on Reply
#142
lexluthermiester
windwhirlYou're likely expected to get it from Windows Update once it lands in whatever channel you're, as an optional update. Though, it's already available from Microsoft Catalog too
www.catalog.update.microsoft.com/Search.aspx?q=KB5006746
Nice! It's still only a preview. Might be wise to wait for the final that has the .CAB version of the update.
Posted on Reply
#143
Mussels
Freshwater Moderator
AMD's fix is out, and i've been told the MS update is out on general releases too.

AMD Ryzen™ Chipset Driver Release Notes (3.10.08.506) | AMD

Install both, enjoy happy times!

My results post fixes:


~630 to ~650GB/s

Again, you can reroll those tests and see big variances, so i did it a bunch and went a common, high result (spam clicking it trashes the numbers, what a shock)
Posted on Reply
#144
blunden
lexluthermiesterNice! It's still only a preview. Might be wise to wait for the final that has the .CAB version of the update.
Microsoft stated that it would be a preview update they would release for "seekers". For everyone else, it will land in the November update.
Posted on Reply
#145
stimpy88
blundenMicrosoft stated that it would be a preview update they would release for "seekers". For everyone else, it will land in the November update.
Well after the Alder Lake reviews are published... Now those of us in the know will see how honest the tech reviewers are when they publish (or don't) the Windows build number they used for their benchmarks.
Posted on Reply
#146
bug
stimpy88Well after the Alder Lake reviews are published... Now those of us in the know will see how honest the tech reviewers are when they publish (or don't) the Windows build number they used for their benchmarks.
I expect almost everybody will apply this patch. I trust TPU and Anand the most, I'm positive they will.
Posted on Reply
#148
Liviu Cojocaru
Things look good on Win 11 now, my 5600X seems to be working as it should, got back to Windows 11 today as I downgraded last week because of this issue.
Posted on Reply
#149
bug
Liviu CojocaruThings look good on Win 11 now, my 5600X seems to be working as it should, got back to Windows 11 today as I downgraded last week because of this issue.
Going back to Win10 and now returning to Win11 lost you more time than you would have lost staying on Win11 with reduced performance. Just sayin'.
Posted on Reply
#150
Mussels
Freshwater Moderator
bugGoing back to Win10 and now returning to Win11 lost you more time than you would have lost staying on Win11 with reduced performance. Just sayin'.
I mean... i set up an entirely seperate OS on a secondary SSD and it only took 45 minutes.
Maybe he clicked downgrade and went to bed
Posted on Reply
Add your own comment
May 2nd, 2024 15:09 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts