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

[R-T-B]

thats 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

This is why options are nice.

 

[zlobby]

thats 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.

 

[blunden]

No, 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

View attachment 221062

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.

 

[Mussels]

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.

my test was absolutely not optimised, i really should throw 10 on my other SSD

 

[blunden]

my 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.

 

[Raven Rampkin]

Did anyone force you to buy AMD?

This goes surprisingly well with your status.

I 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...

 

[Mussels]

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.

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.

 

[blunden]

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.

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.

 

[Mussels]

Absolutely clean W10 install

And now the clean, everything quit W11 result

I see only margin of error stuff here.

 

[blunden]

Absolutely clean W10 install

View attachment 221546



And now the clean, everything quite W11 result
View attachment 221549


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.

 

[Mussels]

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?

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.

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

Spoiler

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.

 

[blunden]

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?


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

Spoiler

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"
View attachment 221587

View attachment 221588


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:
View attachment 221593

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?

 

[Mussels]

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.

 

[blunden]

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.
View attachment 221710

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.

 

[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

 

[lexluthermiester]

edit a few hours later:
OK so KB5006746 is out now, nice

That's just a preview page. There's nothing to download yet.

 

[windwhirl]

That's just a preview page. There's nothing to download yet.

You'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

Microsoft Update Catalog

 

[lexluthermiester]

You'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

Microsoft Update Catalog

Nice! It's still only a preview. Might be wise to wait for the final that has the .CAB version of the update.

 

[Mussels]

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)

 

[blunden]

Nice! 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.

 

[stimpy88]

Microsoft 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.

 

[bug]

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.

I expect almost everybody will apply this patch. I trust TPU and Anand the most, I'm positive they will.

 

[acu]

 

[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.

 

[bug]

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.

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'.