AMD Ryzen Branch Prediction Optimizations Now Available to Windows 11 23H2 (2024)

leezhiranWill this patch do any good to Windows 11 22H2 ?

22H2 support has ended. Upgrade to 23H2.

My small test.
23H2. System admin account. VBS off. 1440p max settings. Not cpu limited.
Newer times are with KB5041587 installed. 2080 Ti. GoT had FSR3 FG enabled. Others were native res/no upscaling.

Tomorrow22H2 support has ended. Upgrade to 23H2.

My small test.
23H2. System admin account. VBS off. 1440p max settings. Not cpu limited.
Newer times are with KB5041587 installed. 2080 Ti. GoT had FSR3 FG enabled. Others were native res/no upscaling.

What GPU are you using?

I guess my theory was right - unless you're playing on an RTX 90909999 Ti Ultra Platinum FX at 1080p, you won't feel any difference. This patch is a nothingburger for 98% of people.

ZubasaBasically there are 2 major issues with older versions of Windows. The Branch Prediction/ scheduling issues, and then the Virtualisation Based Security performance issues.
Running as sys Admin / Root bypass the VBS issue, this is why there are some gains on 23H2 doing that.

Also some early testing shows that 23H2 is still slower than 24H2 after this patch. At lease for these games the patch did F'k all on 23H2, at least for this guy's set up.
Amd/comments/1f2t9s5/_/lk990xr
Imgur: The magic of the Internet
Imgur: The magic of the Internet

The major changes in branch prediction mechanism seem to be present since Zen 3:
hardwaretimes.com/amd-zen-vs-zen-2-vs-zen-3-vs-zen-4-vs-zen-5-core-architecture-road-to-ryzen-9000-cpus/

AMD states that the fix will positively affect Zen 3/4/5 performance. That points to a fact that Windows was not optimized to utilize improved mechanism since Zen 3, while Linux was.
Then there may be another Zen 4/5 branch prediction issue on the top of that. Zen 5 makes the biggest changes to the decoders as well (the biggest since Zen 1).
And after that there might be VBS performance hit.

N/Asome users reported W10 support. AMD stated older operating system will get this. Not just for Zen5, but all later Zen's have been running with this performance untapped until now. What a missed opportunity for AMD to bring this improvement so late. It could have crushed Intel. Well it just did. Better later than never.

Yeah, all the reviews and other performance tests were inaccurate. There has always been some unused performance on the table due to the piece-of-sheet OS.

wolfI suppose when I finally get to Windows 11 this might help me on Zen3, but for now I'm happy with my box being on Win10.

Indeed, same.

AusWolfWhat GPU are you using?

re-read the post, he has the GPU and CPU both mentioned & listed.

Tomorrow22H2 support has ended. Upgrade to 23H2.

My small test.
23H2. System admin account. VBS off. 1440p max settings. Not cpu limited.
Newer times are with KB5041587 installed. 2080 Ti. GoT had FSR3 FG enabled. Others were native res/no upscaling.

Did you place those times correctly?
It looks like in some games you're doing even worse after applying the update. Makes no sense.

b1k3rdudere-read the post, he has the GPU and CPU both mentioned & listed.

Ah, gotcha. I seem to be a bit tired after work. :ohwell: In that case, the rest of my post applies.

AusWolfWhat GPU are you using?

I guess my theory was right - unless you're playing on an RTX 90909999 Ti Ultra Platinum FX at 1080p, you won't feel any difference. This patch is a nothingburger for 98% of people.

Updated after posting: 2080 Ti. About 6750XT/4060 Ti performance.
There are some gains in 0.1% but nothing massive.

LittleBroDid you place those times correctly?
It looks like in some games you're doing even worse after applying the update. Makes no sense.

I could not get the times to line up properly. If its 07:XX then it's pre patch. If it's 09:XX then it's post patch.
Quake Champions and BattleFront II seem to be out of order. Others are Older > Newer in the chart.

TomorrowUpdated after posting: 2080 Ti. About 6750XT/4060 Ti performance.
There are some gains in 0.1% but nothing massive.

So basically, you need a semi-industrial GPU with an integrated nuclear reactor to see any difference, just as I thought.

Another piece of non-news blown out of proportion by the media. What a surprise! :shadedshu:

TomorrowI would be interested in seeing someone test pre-patch 23H2, post-patch 23H2 and 24H2 preview build. Supposedly 24H2 still has the edge.
It would take too long for me to replicate my system environment for 24H2. I ran these tests with my usual background apps running.
If i ran 24H2 clean install the results would be better just due to clean install itself.

Still, your results are infinitely more useful than all those out there on Youtube (including Hardware Unboxed) showing us how much you gain with a 4090 that maybe 1% of gamers have.

Yeah i wanted to mainly test for myself to see how, and if my experience would be changed with the settings i use.
So i figured i might as well share the results. Software i used was CapFrameX. The benchmark runs were relatively short - under 5 minutes.
Rise of the Tomb Raider was the only one where i used canned benchmark. Others were live gameplay. In GoT i ran around in Hyoshi Springs village on my New Game+ save.

Im not saying HUB testing is wrong. Their testing makes sense for them. They're trying to predict the CPU's performance by creating an artificial bottleneck with 1080p low and using 4090. That was no my goal. Im not trying to predict how 5800X3D will handle 5090 or 6090 performance years from now.

I bet performance was fine under Windows 10 since day one, but all reviewers used Windows 11?

Can't wait to see 9000X3D vs Arrow Lake, I will buy the one or the other in Q4/Q1.

TomorrowYeah i wanted to mainly test for myself to see how, and if my experience would be changed with the settings i use.
So i figured i might as well share the results. Software i used was CapFrameX. The benchmark runs were relatively short - under 5 minutes.
Rise of the Tomb Raider was the only one where i used canned benchmark. Others were live gameplay. In GoT i ran around in Hyoshi Springs village on my New Game+ save.

Im not saying HUB testing is wrong. Their testing makes sense for them. They're trying to predict the CPU's performance by creating an artificial bottleneck with 1080p low and using 4090. That was no my goal. Im not trying to predict how 5800X3D will handle 5090 or 6090 performance years from now.

The HUB test makes sense from some perspective, but it's not one of the regular gamer. Yet, here we are, discussing among us, regular gamers, how significant this update is. What a waste of time, isn't it? :laugh:

LittleBroThe major changes in branch prediction mechanism seem to be present since Zen 3:
hardwaretimes.com/amd-zen-vs-zen-2-vs-zen-3-vs-zen-4-vs-zen-5-core-architecture-road-to-ryzen-9000-cpus/

AMD states that the fix will positively affect Zen 3/4/5 performance. That points to a fact that Windows was not optimized to utilize improved mechanism since Zen 3, while Linux was.
Then there may be another Zen 4/5 branch prediction issue on the top of that. Zen 5 makes the biggest changes to the decoders as well (the biggest since Zen 1).
And after that there might be VBS performance hit.

Yeah, all the reviews and other performance tests were inaccurate. There has always been some unused performance on the table due to the piece-of-sheet OS.

What about the HT issues that were suspected initially, and also prompted W1zz to do additional testing? Cores running two hardware threads when they wouldn't have too? Was that suspicion false?

I bet the performance uplift we see with 24H2 update is what Windows 10 performance have been looking all along?

This is probably why Windows 11 gaming performance has been mediocre (compared to 10) for years.

Minus InfinitySo Win 10 has no support for this feature and Zen 5 will run like sh*t on Win 10 or does Zen 5 not even support Win 10?

In my personal opinion since Windows 95, Windows wasted more hardware than Gnu Userspace and the linux kernel. Means hardware has to be bought for an acceptable Windows experience. Nothing changed.

Windows 10 is out of support range in my personal opinion. Nothing new. I doubt there will be any new features for W10.

GoldenXThis is Microsoft's fault, not AMD's. Linux has benefited from this for years.

The crime here is the rewrites that 24H2 implemented taking so long to reach users.

Not really.

sys-devel/gcc hardly has optimizations for Ryzen 5000 / 7000 cpus.
I also check what is new in the kernel version increases because I Build my kernel myself.

There are a few more options recently. A lot of work is being done for scheduler and those security stuff.

Most newbies with their binary distros - do not know what I talk about. They use prebuild software and generic kernel in their binary distros. Which hardly utilize or benefit the hardware at all.

I moved my hardware from Ryzen 5800X -> Ryzen 3 3100 -> Ryzen 7600X from January 2023 till April 2023. Therefore I had to recompile my hole box. And I saw the compile times. A performance indicator is, how long a package compiles. gcc / libreoffice / and so on. I delete from time to time the log files, still I see the difference over the package versions over the time with different cpus.

I also recompile with the corresponding new cpu flags for the Ryzen 7600X. Barely a difference for around 1400 to 1700 installed packages.

The file system also has a big impact on performance. E.G. tmpfs - file system stored in the DRAM.

Summary: gcc is in my point of view far behind in regards of optimizations of current hardware. That means as of now in my point of view Ryzen 7000 or newer. Binary distros with binary generic kernels hardly benefit.

--

Phoronix benchmarks - forget them. Just a pile of numbers for clickbait. Faster compile times -> that is easy to measure and to see over several times of package compiles over time. Less backup time and so on, that is performance I see and can verify.

WirkoWhat about the HT issues that were suspected initially, and also prompted W1zz to do additional testing? Cores running two hardware threads when they wouldn't have too? Was that suspicion false?

Why SMT in Zen 5 scales differently than in Zen 4 remains unknown.
Microsoft or AMD haven't even called it an issue yet, so ...

kondaminwindows 10 and 11 are a horrible intrusive mess people put up with because they have to

QFT.
Linux is also a horrible mess, but that's besides the point. An overwhelming percentage of both commercial and consumer software is made for Windows x86 and sometimes also for Apple. We run Windows because the software we've purchased NEEDS it.

Quick tested in CS2 benchmark from workshop.

My gf computer whit an 5600x and 3060ti
3% gain. In fps before 655 after 668

My own system. 7700x and 4070 ti super
7% gain in fps. Before 660 after 703

Chrispy_QFT.
Linux is also a horrible mess, but that's besides the point. An overwhelming percentage of both commercial and consumer software is made for Windows x86 and sometimes also for Apple. We run Windows because the software we've purchased NEEDS it.

In which way is it a horrible mess? Honest question.

TomorrowI could not get the times to line up properly. If its 07:XX then it's pre patch. If it's 09:XX then it's post patch.
Quake Champions and BattleFront II seem to be out of order. Others are Older > Newer in the chart.

Also these 2 ^^ seem to be either out of order or the update actually decreased performance.
If they are reversed, then those 0.1% improvements are really nice. You've got 33% raise in Hitman, 39% in tit*hima, in SoTR it's above 60%.

mikesgIf prediction optimisations are to decide what to keep in the cache, it's not going to benefit the X3D series much.

They are not to "decide" but indirectly they have some effect, yes, because a (small) amount of mispredicted code and data will go into the cache. The content of the cache is mostly data, not program code.