F5F Stay Refreshed Hardware Desktop Consider building AdviceAMD or Intel for GPU pass-through?

Consider building AdviceAMD or Intel for GPU pass-through?

Consider building AdviceAMD or Intel for GPU pass-through?

R
RepoRizer
Posting Freak
Find
872
04-19-2025, 02:22 AM
#1
Hi,
I’m seeking guidance on setting up a desktop setup for open source software development and testing. The project size is small (~100K SLOC) and can be developed quickly in a modest VM; CPU power isn’t a concern. I’m considering either an i7 Alder Lake or a Ryzen 7000 65W processor, without overclocking.

The main challenge involves encoding H.264 and AV1 on both NVidia and Intel platforms, using NVENV and VA-API. The AV1 encoding will require access to a recent NVidia card and an ARC Alchemist card. I plan to use only one GPU at a time for encoding/decoding, not for general tasks. I suspect each GPU may need only 4 or 8 PCI lanes, so I won’t be targeting EPYC, TR, or Xeon systems. A single box solution would be ideal, given the limited space in my home office.

My current workflow relies heavily on running multiple VMs via KVM for flexibility across platforms. To maintain consistency, I’ll need to route both GPUs through different VMs as needed, while using an integrated GPU feature on the workstation for development. From what I understand, I’ll have to familiarize myself with VFIO and IOMMU groups to manage this setup effectively.

Since I’m not a large company and want to avoid costly errors, what potential issues should I watch out for when choosing a CPU and motherboard that can support two different cards and PCI pass-through?
Reply
Reply
R
RepoRizer
04-19-2025, 02:22 AM #1

Hi,
I’m seeking guidance on setting up a desktop setup for open source software development and testing. The project size is small (~100K SLOC) and can be developed quickly in a modest VM; CPU power isn’t a concern. I’m considering either an i7 Alder Lake or a Ryzen 7000 65W processor, without overclocking.

The main challenge involves encoding H.264 and AV1 on both NVidia and Intel platforms, using NVENV and VA-API. The AV1 encoding will require access to a recent NVidia card and an ARC Alchemist card. I plan to use only one GPU at a time for encoding/decoding, not for general tasks. I suspect each GPU may need only 4 or 8 PCI lanes, so I won’t be targeting EPYC, TR, or Xeon systems. A single box solution would be ideal, given the limited space in my home office.

My current workflow relies heavily on running multiple VMs via KVM for flexibility across platforms. To maintain consistency, I’ll need to route both GPUs through different VMs as needed, while using an integrated GPU feature on the workstation for development. From what I understand, I’ll have to familiarize myself with VFIO and IOMMU groups to manage this setup effectively.

Since I’m not a large company and want to avoid costly errors, what potential issues should I watch out for when choosing a CPU and motherboard that can support two different cards and PCI pass-through?

Reply
Reply