Windows 10 & Intel Core i9-12900K Alder Lake Performance

[Pickoff]

Hello,

Thanks for the comparison.

VBS/ Memory integrity in W11 was on or off for the tests ?



For info: Process Lasso allows you to select cores affinity for tasks (permanently).
Every process spawned by one process for which you selected a specific core affinity will have the same core affinity. (i.e, if you set only E-cores for Origin, if you launch the game, it will be allocated only to E-cores.)

 

[AlwaysHope]

Yeah not upgrading to W11 until at least 2023.

And now is not the best time to jump into Intel's new hybrid arch experiment, I'll see where they're at after they've ironed out the kinks comes the 3rd generation of E and P cores.

+1 on that with the hardware, but may try win11 next year on one of my frequently used PCs.

 

[mrnagant]

b) there is LTSC, big fan of that, using it right now .. and eagerly waiting for update, cause while its great in many aspects, its currently seriously behind and Im starting to chew my fingers a bit since there is few new features that I kinda need hopefully, anyday soon

Anyway, LTSC should be good for 2025+5 so up to 2030. At that time Win 10 probably will be really obsolete anyway. I expect to move to Win 11 LTSC when its on like 2nd or 3rd release. If first will be like Win 10, then that will be a bit too much of beta.

Inability to select what core should do what is fairly amazing, given its something Windows struggle with since birth of Hyper-Threading. But then its MS.

Sry, but no more 10yrs of LTSC support. Plus if you are already getting itchy for the next version of LTSC, what is the point in going LTSC?
The next Windows 10 Long Term Servicing Channel (LTSC) release - Microsoft Tech Community

With Win 10 Edu/Ent you already get up to 30 months of support on fall feature builds. LTSC 2019 is only 3 years old and you already want to upgrade to what I presume will be LTSC 2022 for the new features. Windows 11 feature builds are moving to 3 year support on Edu/Ent.

The next LTSC will still be "Windows 10". Will be interesting if it is Build 20348 like Server 2022 (which is still Win10 like) or be something like build 19043/44.

 

[Why_Me]

just wait until we get to the court case that e-cores aren't real cores

 

[W1zzard]

VBS/ Memory integrity in W11 was on or off for the tests ?

Off, like for all my testing

 

[yeeeeman]

@W1zzard : can you please do a test with 35/65/95/105/150W TDPs? on 12900K/12700k/12600k.

 

[Crackong]

just wait until we get to the court case that e-cores aren't real cores

And buildzoid isn't happy since e-cores limits ring frequency by 20%

 

[W1zzard]

@W1zzard : can you please do a test with 35/65/95/105/150W TDPs? on 12900K/12700k/12600k.

You forgot 125 W. Unlikely, 6 TDPs x 3 CPUs x 5 hours per run = 90 hours, so almost 2 weeks of work

 

[Vayra86]

just wait until we get to the court case that e-cores aren't real cores

Lol. Knowing Intel they've already launched a full on investigation on the possibilities of that. But its an interesting thought nonetheless. What's different now, one might ask.

Look at their marketing, the presentation slides and the number of disclaimers everywhere. They're not overselling those E cores anywhere, except in net performance graphs.

oh, yesss, for 2022 => win 11 + i7 and i9 alder lake + 64gb ddr5 + rtx 4080-4090 = it is the future of pc dekstop for gaming......

For two years

I don't think Alder Lake is the optimal choice going forward for the big little technology. Its an early adopter gen. Buying anywhere in 2021, 2022 or 2023 is buying a box of issues waiting to be fixed. Especially for gaming - note the fact consoles carry Zen cores and not a Big Little setup. It shows in the results even today. Matter of fact, for gaming, those E cores are mostly just in the way.

 

[R0H1T]

I hope Intel gets sued for not letting us use all of the E cores all the time!

Spoiler

No not really, but knowing the US of A it's a possibility with their trigger happy lawyers

. Especially for gaming - note the fact consoles carry Zen cores and not a Big Little setup.

Speaking of there's a good chance the next consoles from AMD(?) could have this Big Little setup with possibly the big cores dedicated towards games & the rest of the OS & background tasks limited to small cores.

You can already do this using software on Windows, so I guess having that on consoles could yield much better efficiency results! Not to mention the M1xxx ~ I have a feeling the next gen could feature a massive MCM setup.

 

[Vayra86]

Speaking of there's a good chance the next consoles from AMD(?) could have this Big Little setup with possibly the big cores dedicated towards games & the rest of the OS & background tasks limited to small cores.

Possibly, could lend somewhat more TDP budget to GPU. But otherwise I don't see a purpose, its not like consoles are productivity machines.

 

[R0H1T]

But otherwise I don't see a purpose, its not like consoles are productivity machines.

Yes but having even a couple of small cores dedicated to the OS & other background tasks is just better than doing it on the big cores, especially during gaming. For instance you can try this using Process Lasso, or Process hacker, & see the difference yourself.

 

[Vayra86]

Yes but having even a couple of small cores dedicated to the OS & other background tasks is just better than doing it on the big cores, especially during gaming. For instance you can try this using Process Lasso, or Process hacker, & see the difference yourself.

Theoretically, but in practice, will you even notice? I have never seen OS interrupt my gaming tbh, even on 6 cores.

I get a strong impression we are seeing problems because there is a new solution , not because there are actual problems. And we are still running those tasks over the same RAM anyway, too.

 

[bug]

Theoretically, but in practice, will you even notice? I have never seen OS interrupt my gaming tbh, even on 6 cores.

I get a strong impression we are seeing problems because there is a new solution , not because there are actual problems. And we are still running those tasks over the same RAM anyway, too.

It seems people have forgotten we've had thread priorities since forever.

 

[chrcoluk]

Maybe add process lasso notes to the end? this allows you to set default affinity for a process so doesnt need to be done again and again, in addition can do with process hacker, also can even be done without 3rd party software using command line, here is an example.

C:\Windows\System32\cmd.exe /c start "Program name" /affinity 14 "C:\program folder\program,exe"

This basically forces only 2 cores to be used, the affinity flag uses a hex code to define which cores.

More info here.

Set affinity with start /AFFINITY command on Windows 7

I am using the start /AFFINITY [n] [.exe] command to start an executable with the specified affinity. I have a system with 8 processors (1,2,3,4,5,6,7,8). I'd like to set the process to use all of...

stackoverflow.com

 

[persondb]

Speaking of there's a good chance the next consoles from AMD(?) could have this Big Little setup with possibly the big cores dedicated towards games & the rest of the OS & background tasks limited to small cores.

Consoles already had that for quite a long while.

On top of my head, the Wii had the Starlet processor integrated into it which ran another OS and abstracted security and I/O. Another example is the PS4 that included an ARM co-processor for OS and background stuff. There are other ones afaik.
If the whole purpose would be to do background and OS, then in my view it's likely better to implement them as co-processors than to complicate your CPU clusters, busses and etc.

If you are going to do little cores then the purpose of those will probably be for enhancing multi-threading. For area-wise, they are a lot better than the equivalent number of big cores, so for example you could run a lot of physics calculations or whatever MT that your future game might need.

One thing is that AMD little cores aren't 'little' by the rumors that we have the Zen 4D will be (mostly, I think, maybe some customization too) Zen 4 cores with half the cache. So they would be very fast and the shebang. Probably too big and fast to use for miscellaneous.

 

[rhaoul]

And performance under linux ? I'm curious...

 

[Why_Me]

And performance under linux ? I'm curious...

Linux Patch Fixes Alder Lake XMP, Overclocking Performance

The culprit is Intel Turbo Boost Max Technology

www.tomshardware.com

 

[bug]

And performance under linux ? I'm curious...

Better than Zen3, even without Thread Director support: https://www.phoronix.com/scan.php?page=article&item=intel-12600k-12900k&num=1

 

[bobmllr]

Great analysis. Can you tell me when compiling in Visual Studio were all 24 cores hammered, on Win 10 and Win 11? Cheers.

 

[bug]

Great analysis. Can you tell me when compiling in Visual Studio were all 24 cores hammered, on Win 10 and Win 11? Cheers.

That usually depends on the compiler flags and the project itself. But if you have work for 24 threads, they would end up on all cores, whether it's Win10 or Win11.

The real nasty scenario is when you don't have work for all the cores. Win10 will randomly select a core and Win11 will relegate your compile work to E-cores as soon as you click away from the compiler window.

 

[Wirko]

The real nasty scenario is when you don't have work for all the cores. Win10 will randomly select a core and Win11 will relegate your compile work to E-cores as soon as you click away from the compiler window.

Depends on how the threads are distributed. What does Windows scheduler usually do with a low-threaded load, assuming a high performance power plan is selected? Does it avoid MT as much as possible?

 

[bug]

Depends on how the threads are distributed. What does Windows scheduler usually do with a low-threaded load, assuming a high performance power plan is selected? Does it avoid MT as much as possible?

I'm not sure what exactly depends on the thread distribution.
As soon as a window loses focus, Windows decreases its priority. When running (certain SKUs of) AL on Win11, that gives that windows a high chance of being kicked to an E core.
Imho, this is not strictly a problem of Win11, it's hard for to describe in words what the proper behavior would be, that's usually a sign that is not straightforward to put into code.

 

[Wirko]

I'm not sure what exactly depends on the thread distribution.

For example, an app with 8 threads running on an 8-core CPU with HT but no E-cores. Does Windows attempt to put one thread or two threads on each core? Does that depend on power plan? Does it depend on whether the app is in foreground? I mean, scheduler can make bad decisions even if E-cores aren't even there.

As soon as a window loses focus, Windows decreases its priority. When running (certain SKUs of) AL on Win11, that gives that windows a high chance of being kicked to an E core.
Imho, this is not strictly a problem of Win11, it's hard for to describe in words what the proper behavior would be, that's usually a sign that is not straightforward to put into code.

Yeah, it's not like the same behaviour is best for everyone. However, letting MySQL running on E-cores while P-cores sit idle is probably always bad on a desktop CPU.

 

[bug]

For example, an app with 8 threads running on an 8-core CPU with HT but no E-cores. Does Windows attempt to put one thread or two threads on each core? Does that depend on power plan? Does it depend on whether the app is in foreground? I mean, scheduler can make bad decisions even if E-cores aren't even there.

Well, that wouldn't be an AL specific problem (were we talking AL? I lost track).
Fwiw, I've always been wondering whether batching threads that do light work (as opposed to spreading them across as many execution units the CPU presents) is a pro or a con. I still haven't been able to figure that out.

Yeah, it's not like the same behaviour is best for everyone. However, letting MySQL running on E-cores while P-cores sit idle is probably always bad on a desktop CPU.

You're probably not hammering MySQL all day long a desktop, so an E core is the perfect place for it to loiter. Of course, we'd want that load to move to a P-thread as soon as it starts to, you know, work. A damn good example of a workload that's not that easy to figure out, thank you very much.