Info Upscayl CPU benchmark (much improved with recompiled DLL)

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Jul 27, 2020
28,071
19,157
146
Great! It still has some blocking limit in the thread count. It does not go above 41% utilization for my Epyc 64c/128T CPU.

It's using this for multithreading: https://github.com/google/marl

May have to take a closer look at it to see what the limiting factor may be.
 
Jul 27, 2020
28,071
19,157
146
Thanks, Det0x!

Seems this is one of those "hard" benchmarks where squeezing more performance won't be easy without tinkering with the source code itself.
 
Jul 27, 2020
28,071
19,157
146
@511 I think your 185H can slot in between the 245KF and the Epyc.

If you run the benchmark, make sure to do it on AC power and max performance mode.
 
Jul 27, 2020
28,071
19,157
146
1758706965753.png

Did not expect this CPU to finish so quickly at only 3.56 GHz max boost and 41% overall CPU utilization.

Possibly more RAM bandwidth or AVX-512 optimization by the VS compiler could be the reason.
 
Jul 27, 2020
28,071
19,157
146
I should use the ICX/ISPC Compiler
I would be very interested in your results. You just need to compile the vk_swiftshader.dll from the Google source code and then rename it to vulkan-1.dll and replace in root folder and bin folder of Upscayle.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,250
16,108
136
How many tasks or threads can it run at once ? I may try my 9755 on this
 
Jul 27, 2020
28,071
19,157
146
My 9755 uses only 4800 but 12 channels
Yeah, it's worth a try. Make sure that the NPS setting is at zero in the BIOS. Epyc performs twice as slow with NPS4.

The benchmark behaves weirdly on different systems.

On my Epyc Rome, it showed around 15 threads pegged at 100%. On Ryzen 9950X3D, it was spread over all the cores with lower CPU utilization. On the Xeon 6248R, it was again spread out over the cores, not maxing any cores at 100% as far as I could see but still managed a decent score.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,250
16,108
136
Yeah, it's worth a try. Make sure that the NPS setting is at zero in the BIOS. Epyc performs twice as slow with NPS4.

The benchmark behaves weirdly on different systems.

On my Epyc Rome, it showed around 15 threads pegged at 100%. On Ryzen 9950X3D, it was spread over all the cores with lower CPU utilization. On the Xeon 6248R, it was again spread out over the cores, not maxing any cores at 100% as far as I could see but still managed a decent score.
What about your 128 core (2 x 64) Milan ?
 
Jul 27, 2020
28,071
19,157
146
What about your 128 core (2 x 64) Milan ?
I will get to building it, eventually. I want it to be fancy with AIO cooling and I found something cheap (Coolermaster TR4 cooler that is supposedly compatible with SP3 socket) but not cheap enough to get two of them at once in a single month :)
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,250
16,108
136
I will get to building it, eventually. I want it to be fancy with AIO cooling and I found something cheap (Coolermaster TR4 cooler that is supposedly compatible with SP3 socket) but not cheap enough to get two of them at once in a single month :)
If TR4 looks like SP3 I am pretty sure I have used the same cooler on both. My threadrippers (ols, were 1950x I think they were) are in fact the same cooler as my early EPYC's.

Here is a pic of that system. I still have it ! @igor_kavinski
 

Attachments

  • 20250924_085658[1].jpg
    20250924_085658[1].jpg
    433.7 KB · Views: 6
Last edited:
  • Like
Reactions: igor_kavinski