I just set up a fresh install of Proxmox VE on a 7950X3D and intend to use the 3D V-cache half of it for a gaming/workstation VM with the other CCD running various selfhosted services in LXCs. I gave the VM 16 vCPUs with the host type and set their affinity to "0-7,16-23", then prevented the host from running anything on those cores with these commands:
The performance in Cinebench R23 is on par with a 7800X3D which is as expected, but commands like lstopo or lscpu -C all report only 16MB of total L3 cache instead of 96MB, 512KB of per-core L2 instead of 1024KB, and 64KB of L1d and L1i instead of the expected 32KB.
Why does this happen, and is there any way to fix it? I haven't tested any cache-intensive programs to find out if this is only the reported info being incorrect or if it also affects performance, but either way it isn't ideal.
I'm using Proxmox VE 8.0.4 with kernel 6.2.16-10-pve, ran Cinebench on Windows 11 and lscpu/lstopo on Arch.
code_language.shell:
systemctl set-property --runtime -- user.slice AllowedCPUs=8-15,24-31
systemctl set-property --runtime -- system.slice AllowedCPUs=8-15,24-31
systemctl set-property --runtime -- init.scope AllowedCPUs=8-15,24-31The performance in Cinebench R23 is on par with a 7800X3D which is as expected, but commands like lstopo or lscpu -C all report only 16MB of total L3 cache instead of 96MB, 512KB of per-core L2 instead of 1024KB, and 64KB of L1d and L1i instead of the expected 32KB.
Why does this happen, and is there any way to fix it? I haven't tested any cache-intensive programs to find out if this is only the reported info being incorrect or if it also affects performance, but either way it isn't ideal.
I'm using Proxmox VE 8.0.4 with kernel 6.2.16-10-pve, ran Cinebench on Windows 11 and lscpu/lstopo on Arch.
Last edited: