CPU pinning logical/physical indices about gpu-passthrough-tutorial HOT 3 CLOSED

I must have made a mistake somewhere while editing the pinning since I can't reproduce the 'worse' results with your configuration anymore. Thanks for helping me out anyway!

from gpu-passthrough-tutorial.

Hey sorry for the delayed response...

First, let me refer you to this section of the arch wiki page on cpu pinning. Here you'll see the use of the command lspcu -e. The cpu threads are numbered on the leftmost "CPU" column and you can see which threads share the same physical core under the "CORE" column. Pretty much every tutorial on the subject will tell you to pin threads together by physical cores and not across different physical cores. In your modified example, you're pinning threads across different physical cores. This is simply not recommended anywhere and I'm genuinely surprised to see better performance numbers.

Let's say based on the CPU topography for your 3950X that I wanted to pass physical cores 8-15 (bottom row of lstopo) to the guest and leave 0-7 (top row of lstopo) to the host. This corresponds to passing 16 out of 32 logical threads to your VM. Here's what I'd expect the correct pinning configuration with threads paired by physical cores to look like:

<vcpu placement="static">16</vcpu>
<cputune>
    <vcpupin vcpu="0" cpuset="8"/>
    <vcpupin vcpu="1" cpuset="24"/>
    <vcpupin vcpu="2" cpuset="9"/>
    <vcpupin vcpu="3" cpuset="25"/>
    <vcpupin vcpu="4" cpuset="10"/>
    <vcpupin vcpu="5" cpuset="26"/>
    <vcpupin vcpu="6" cpuset="11"/>
    <vcpupin vcpu="7" cpuset="27"/>
    <vcpupin vcpu="8" cpuset="12"/>
    <vcpupin vcpu="9" cpuset="28"/>
    <vcpupin vcpu="10" cpuset="13"/>
    <vcpupin vcpu="11" cpuset="29"/>
    <vcpupin vcpu="10" cpuset="14"/>
    <vcpupin vcpu="11" cpuset="30"/>
    <vcpupin vcpu="10" cpuset="15"/>
    <vcpupin vcpu="11" cpuset="31"/>
<cputune>

Please go ahead and confirm whether your current configuration outperforms this one. I would not expect it to. Here's the configuration I suggest you run a comparison with in which threads are paired across different physical cores:

<vcpu placement="static">16</vcpu>
<cputune>
    <vcpupin vcpu="0" cpuset="8"/>
    <vcpupin vcpu="1" cpuset="9"/>
    <vcpupin vcpu="2" cpuset="10"/>
    <vcpupin vcpu="3" cpuset="11"/>
    <vcpupin vcpu="4" cpuset="12"/>
    <vcpupin vcpu="5" cpuset="13"/>
    <vcpupin vcpu="6" cpuset="14"/>
    <vcpupin vcpu="7" cpuset="15"/>
    <vcpupin vcpu="8" cpuset="24"/>
    <vcpupin vcpu="9" cpuset="25"/>
    <vcpupin vcpu="10" cpuset="26"/>
    <vcpupin vcpu="11" cpuset="27"/>
    <vcpupin vcpu="10" cpuset="28"/>
    <vcpupin vcpu="11" cpuset="29"/>
    <vcpupin vcpu="10" cpuset="30"/>
    <vcpupin vcpu="11" cpuset="31"/>
<cputune>

I've also gone ahead and edited the README for the tutorial to make the cpu pinning section a little bit easier to understand. Please report back with your results.

from gpu-passthrough-tutorial.

Closing issue awaiting response

from gpu-passthrough-tutorial.