I have a Ubuntu Plex VM running under an ESXi host with an Intel iGPU passed through. I’ve had a few buffering issues with 4k transcodes, and it has me researching potential implications using the iGPU passed through to a VM and how VRAM is handled.
Since the iGPU leverages system RAM and the iGPU is passed through to the VM, my understanding is the VRAM will be taken from whatever RAM is allocated to that VM (and unused by the rest of the system). So RAM must be assigned with this in mind.
More importantly, I reviewed my motherboard manual and under the Onboard Graphics section, it lists the maximum shared amount of memory as 512MB! I have another MB with the same socket and it says ‘Graphics specifications may vary depending on the CPU installed’.
As this is PCI passthrough on ESXi, I used the lspci commands to review the memory shown by the GPU within the Ubuntu VM. It shows 256MB.
Everyone seems to praise Intel Quick Sync and for good reason, but the info above has me wondering if things aren’t quite as clear when passing through the iGPU to a VM without understanding the RAM requirements for the VM and potential limitations of the motherboard.
Very interested in hearing other’s opinions here! Ultimately I need to decipher this so I can proceed with a new build. I considered Intel a given for QSV, but now I think a dedicated GPU passthrough would be a more guaranteed approach. Although I’m on ESXi, I plan on migrating to Proxmox in the next year or two, and passing through a dedicated GPU seems easier there.
Advantage of AM5 + dGPU on the ESXi side is that it avoids the big little Intel incompatibility with ESXi. Disadvantage is power consumption would be higher going this route, although something like an Arc a310 or 380 wouldn’t be terrible.
Another advantage of AM5 is the ProArt motherboards that have x8x8x4 PCIe lanes. The B650 model is quite affordable and this would give me x8 HBA, x8 GPU, x4 10G NIC. The x670E model has 10G but it is pricey and the 10G NIC is Marvell which is not supported by ESXi.
I appreciate any input!