Do EPYC Milan CPUs even exist for sale?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

lihp

Active Member
Jan 2, 2021
186
53
28
Shouldn't the 7443p be boosting closer to 4.0GHz, not 3.0GHz?

Actually slightly more. In my last few tests I had all cores beyond 3.90 GHz and some hitting 4.01 GHz and 4.02 GHz (according my RHEL server). I still blame my cooler and the liquid metal thermal paste ;).
 

lihp

Active Member
Jan 2, 2021
186
53
28
can you elaborate a bit? which settings?
SMT Control, Core Performance Boost, C-State Control (various settings), APIC mode,... Basically everything under CPU and Northbridge settings.

Examples:
  1. Due to recent security vulnerabilities in Cloud environments, companies tend to disable SMT. RedHat additionally recommends SMT disabling during installation.
  2. Many settings are on auto, where you are better off to set them yourself (performance-wise). Local APIC mode to x2APIC, CCD and Core Control - its 4 CCDs and 6 Cores in case of the 7443, ...
  3. Review the rest of the settings for plausiblity.
 

ca3y6

New Member
Apr 3, 2021
24
3
3
I did look at the -threads parameter of ffmpeg in the past, but it's a weird one. I think by default it uses two times the number of cores on the machine though in practice the CPU is rarely fully utilised if it does, plus the codec itself will have its own limitations in term of how many cores it can use, which is a function of the resolution and the codec. So the parameter is useful to reduce the number of threads used but not to increase it.

Hyperthreading is on (as the number of blue boxes in task manager shows)

But what I find interesting is that if the task is neither single threaded, nor does it uses all cores, it seems the CPU will run below its max clock speed, which surprised me.
 

lihp

Active Member
Jan 2, 2021
186
53
28
Hyperthreading is on (as the number of blue boxes in task manager shows)
Problem is: it works on the Rome one. And both designs are too similar for such a difference. So its either some bug or config in Windows or Hardware.

Try your benchmark with Hyperthreading off. Maybe its too many cores for ffmpeg.
 

ca3y6

New Member
Apr 3, 2021
24
3
3
Well, the number of cores is quite different so it may have to do with a function of % of physical cores used. If ffmpeg only uses 12 cores for instance, then the 16-cores 7302P is almost at capacity whereas the 24-cores 7443P has ample leg room, and it is possible that it drives a different cpu frequency. Just speculating. The 7443P is a new Windows install, so it's pretty much all windows defaults.
 

DRW

New Member
May 1, 2021
29
11
3
But what I find interesting is that if the task is neither single threaded, nor does it uses all cores, it seems the CPU will run below its max clock speed, which surprised me.
Even when you were running 100% on all 24 cores, that was 100% at 3.0GHz. Quite a bit lower than the 3.9 all-core seen by @lihp
 

lihp

Active Member
Jan 2, 2021
186
53
28
... If ffmpeg only uses 12 cores for instance, then the 16-cores 7302P is almost at capacity whereas the 24-cores 7443P has ample leg room,...
So far I only implied but please @ca3y6 could you monitor CPU temperatures under ffmpeg load? And come back an tell us? Maybe a bad contact on the cooler to CPU lid, maybe a setting in BIOS/Windows or you missed some settings when running ffmpeg on the 7443P.
 

ca3y6

New Member
Apr 3, 2021
24
3
3
Sure. The screenshot below is with one ffmpeg compression (h264@1080p), only some of the cores are used, the CPU clock is well below its capacity (1.6GHz), the CPU is "cold" (40C)

7443P-1 ffmpeg.PNG

The second screenshot is with two ffmpeg compressions in parallel (h264@1080p), all cores fully used, CPU going to base clock, CPU is warm but well within limits (60C)
7443P-2 ffmpeg.PNG
So the temperatures do not look abnormal to me.

Regarding the BIOS, all CPU options are set to default (and it is a clean install of Windows Server 2019, I didn't tweak anything related to the CPU or power savings).
bios1.PNG

bios2.PNG
 

uldise

Active Member
Jul 2, 2020
209
71
28
Windows Power plan, is it at High performance or something like that?
you can boot some Linux live, stress the Cpu and watch for Cpu speed.. - it will prove Windows or Bios..
 

Mace

New Member
Feb 3, 2021
26
6
3
You won't reach the maximum core-speeds without adjusting some things in the Bios. First of all, set the cTDP and PPLC to the max wattage of your CPU. Next, set the Determinism Control to Performance.

Reboot and see what that does. I'm still fiddling with other settings (DF Cstates and CCLK) to get al the available power unlocked.
 

ca3y6

New Member
Apr 3, 2021
24
3
3
Windows Power plan, is it at High performance or something like that?
you can boot some Linux live, stress the Cpu and watch for Cpu speed.. - it will prove Windows or Bios..
Actually you were right. If I set the power plan to High Performance (from the default Balanced), then I get the expected performance profile. Mystery solved! Thanks
 
  • Wow
  • Like
Reactions: DRW and uldise

Dolpa

Member
Apr 27, 2017
54
30
18
Hungary
I finally managed to order a 7443P in an EU shop. They can ship to many countries in Europe in 3-4 business day and at the moment 14 pcs on stock.
My order will be arrive in the next week. :)
 
  • Like
Reactions: lihp

lihp

Active Member
Jan 2, 2021
186
53
28
Shouldn't the 7443p be boosting closer to 4.0GHz, not 3.0GHz?
Just to add what I meant with 3.90+ GHz. Right now doing some tests (not full load but close) and thought to make you a screenshot. Once I make a full load test again I will see if I can hit a screenshot at 4.00 GHz +

Unbenannt-1.jpg
 

DRW

New Member
May 1, 2021
29
11
3
Just to add what I meant with 3.90+ GHz. Right now doing some tests (not full load but close) and thought to make you a screenshot. Once I make a full load test again I will see if I can hit a screenshot at 4.00 GHz +
I'd be interested in seeing what temps are like at full load. I was planning on using a dynatron A26.
 

lihp

Active Member
Jan 2, 2021
186
53
28
250W TDP on the consumer be quiet! cooler vs 280W on the A26. The A26 is a lot noisier at load though.
In that league it probably comes down to personal taste. I was actually quite "afraid" of the noise my server will make in my office. I dont hear the server at all - only rarely like every two days I hear the Ultrastar drives working for a brief moment. Thats it. My Workstation is (still for now) louder, while I dont hear anything of the server (besides above mentioned drives).