Z690 Chipset Motherboards and Alder Lake CPU

[energy23]

@energy23

I ran your experiment on my Z690 Aero G DDR4, i9-12900K (stock speed), 8P+8E No HT, 32GB DDR4 @3600MHz, Sapphire RX 580 8GB, WD Black SN750 SE 1TB.

When using 8P+8E No HT, my CPU temp runs around 80°C under heavy load like this test or Cinebench. I can't run with HT enabled as the CPU overheats with these tests. I am currently using an ARCTIC Liquid Freezer II 280. But I had the same heat limitations using an iCue H115i RBG PRO XT, as well as an NOCTUA NH-U12S.

Run time: 6:34

View attachment 539521

Thank you!
Ok this is really interesting. I've seen in Cinebench R23 that the 12900k scores just under 2,000 points PER CORE while my 4790k scores around 1,000 per core, so basically double the performance in the 12900k.

However...

The E-Cores don't seem to be used at all in Handbrake!

Let's keep things rather simple;

4790k: 4 Cores @ 1,000 points each = 4,000 total score (in reality I get a score of about 5,100 but this is with HT enabled)
12900k: 8 cores @ 2,000 points each = 16,000 total score

That's 4X the difference (in VERY simplified terms). Comparing my compression test to yours:

4790: 22 Minutes 49 Seconds (1369 seconds total)
12900k: 6 Minutes 34 Seconds (394 seconds total)

1369 / 394 = 3.47X Difference

I honestly would have expected a bigger difference given that there's that much extra horsepower in the P-Cores alone and STILL another 8 e-cores on top of that.

This is exactly what I was wondering; how do apps like Handbrake see the E-cores and use them?

I have some questions though:

1) I've seen the 12900k tested with e-cores disabled but P-Cores and Hyperthreading enabled. With this setup, I recall temperatures at load skyrocketing to nearly 100 degrees. Is this the same situation for you?

2) What happens if you disable e-cores and run only on E-cores without hyperthreading?

Thanks again!!!

 

[dclive]

@dclive

I just duplicated @energy23's test, as he was looking for volunteers, using the same test file and presets. Test file is shown in the queue - bbb_sunflower_2160p_30fps_normal.mp4

Found it - post 1891. Thanks! Didn't see that, and the details. I'll have to give it a try using both i7-8700 and also via VideoToolBox (for QSV).

 

[dclive]

BBB_sunflower_2160_30_normal test file, i7-8700. 22.15fps with CPU [QSV:41].

i7-8700k: 23.35 fps at 4.9; 24.48 at 5.0 [QSV:49.48].

 

[NCMacGuy]

2) What happens if you disable e-cores and run only on E-cores without hyperthreading?

@energy23

For item (2) did you mean disable E-cores and run P-Cores without HT? I did run the Handbreak test with 8P only and NO HT, Intel Power Gadget reported 25-30% "Core Utilization" and CPU Temp ran an average around 90°C. Run time was 9:06

When running the 8P+8E No HT Handbreak test, Intel Power Gadget was reporting only around 50% "Core Utilization." So apparently Handbreak is not using E-Cores.

As you expected. I disabled E-Cores. With 8P-cores and HT active the Handbreak test caused CPU temps to peg at 100° immediately and the test was stopped.

 

[macoved1974]

Updates in configuration and results.
Capacitor installed correctly.

 

[StefanAM]

Something like running Handbrake would be solid. For instance, compressing Big Buck Bunny from 4k to 1080p using the built-in preset "Super HQ 1080p30 Surround" with no modifications.

I used the 2160p 30fps Normal video fount at the following link:

http://ftp.vim.org/ftp/ftp/pub/graphics/blender/demo/movies/BBB/

Handbrake 1.5.1

HandBrake: Open Source Video Transcoder

HandBrake is an open-source, GPL-licensed, multiplatform, multithreaded video transcoder.

handbrake.fr

My runtime was 22 minutes and 49 seconds:

My system specs:
i7 4790k @ stock speeds
16gb DDR3 RAM @1866mhz
Sapphire RX 6800
256gb SSD WD Blue
MSI Z97 Krait SLI

Background running apps:
Plex Server
iStat Menus
Clipy

I ran the compression after restarting my system with no other apps running (other than background apps).

I'd love to see what others are getting with Alder Lake CPUs @Middleman @ori69 would you be willing to do a compression test?

This test is quite challenging..... I had to put PL2 to 190 W and make other undervolt changing, if not, HandBrake crash. Like you see the E cores are working (I use CpuTopologyRebuild.kext). If you fix your computer to run this, you are ready for any stress test . Now I'll do others benchmark to see how much I lose in performance keeping this settings.

 

[dclive]

Has anyone installed 1MM washers under the ILM in order to lower temps by 5 degrees C?

Alder Lake’s cooling problem straightened out by 5 degrees! - Simple ILM-Mod for Intel’s LGA-1700 socket | Practice | igor´sLAB

Intel’s Alder Lake CPUs run hot, very hot. A few weeks ago, Igor had already identified the bending of the CPU and motherboard due to the LGA1700 socket as a potential cause for this. Today I’ll show…

www.igorslab.de

 

[hboy89]

A simple performance comparison of i9-10900K multicore h264 encode, multcore vs UHD 630.

View attachment 539493

I am drastically simplifying a lot of concerns in the above comparison, but I think it's fair to make a point.

On an Apple supported design, you get your cores, plus the benefit of task-specific silicon.

For a hackintosh, there was a crossroads that got passed in 2021. Whatever happens with hackintosh on RL and beyond, is going to miss out on any specicial silicon value Intel brings to party, beyond brute force.

Things are messy even for Windows! Currently, Intel is retracting support for AVX512, and dropping the security support that allows UHD vid decode a la disc payers. Yikes.

AS to Intel trying to match Apple on core function: Intel efficiency core is an odd-duck, trading off die area, performance and power in manner to copy Apple, but because their market is horizontal, even on Windows, you don't know what you will get nor how long it will work because the total system performance regarding power is a PC vendor concern, not an Intel / Microsoft concern. Whatever progress is made here, Hackintosh can just forget about it. If you get all the cores engaged, screw power and be happy (desktop only because a 12th gen hackinotsh laptop is gonna be a dumb mess).

Maybe some hackintosh desktop users don't care about silicon refinements because they can afford to throw kit at their work and see what sticks. But the sane 12th gen user will move back to Windows — God help you.

Meanwhile Apple Si designs are moving towards more task-specific silicon optimizations fully supported by Apple stack and all power / perf benefits realized up and down the line.

I'm not trying to hate on people having fun. But it's past the end of an era. This scene can only tread water from here on out.

Damn. That sounds pretty bad...

I got to this party very very late, but it was still fun! Thanks for the good community.

That's all fine and good, but it misses something.
Everyone running Z690 will almost assuredly be running a modern AMD GPU. Either an rx5000 series or rx6000 series. So they would not be encoding with just the CPU, but would be able to offload to a pretty powerful GPU
(one that we can keep upgrading as long as Apple keeps supporting new AMD GPUs in their 2019 Mac Pro). It probably won't outperform QuickSync, and it certainly won't be as power efficient, but that's a much fairer comparison.

Does Handbrake not allow that on a Mac? It works on Windows and Linux. If it does, then it would be interesting to test. Because UHD630 may be more power efficient, but I cannot imagine AMD VCE adds nothing.

 

[dclive]

That's all fine and good, but it misses something.
Everyone running Z690 will almost assuredly be running a modern AMD GPU. Either an rx5000 series or rx6000 series. So they would not be encoding with just the CPU, but would be able to offload to a pretty powerful GPU
(one that we can keep upgrading as long as Apple keeps supporting new AMD GPUs in their 2019 Mac Pro). It probably won't outperform QuickSync, and it certainly won't be as power efficient, but that's a much fairer comparison.

Does Handbrake not allow that on a Mac? It works on Windows and Linux. If it does, then it would be interesting to test. Because UHD630 may be more power efficient, but I cannot imagine AMD VCE adds nothing.

This should answer most of your questions: https://handbrake.fr/docs/en/latest/technical/video-videotoolbox.html

Key part: you can't pick which hw it will use; it will use the "highest quality" option.

 

[energy23]

This test is quite challenging..... I had to put PL2 to 190 W and make other undervolt changing, if not, HandBrake crash. Like you see the E cores are working (I use CpuTopologyRebuild.kext). If you fix your computer to run this, you are ready for any stress test . Now I'll do others benchmark to see how much I lose in performance keeping this settings.

Thank you!

So your runtime was 341 seconds versus my 4790k's 1369 second runtime, so....

1369 / 341 = 4.01

Basically exactly 4 times the difference and a little bit better than @NCMacGuy achieved. Looks like you got Hyperthreading enabled as well? If so, it also looks like your temperatures are pretty good hovering at under 70C.

Still, that 4X the performance is somewhat around what the 10900k gets over the 4790k, so paying that much more for a new CPU that doesnt get utilized all the way doesn't really make a whole lot of sense to me. This is of course, all the kinks have been worked out with both hardware and software;

1) Does Handbrake actually take full of advantage of all of a CPU's cores? There's been some debate about that.

2) Do things get any better in Handbrake under Windows where all cores are easily identified? Well, some have issues:

12700k will only engauge eCores, while pCores stand by laughing (Windows 10) - HandBrake