Hi,
I'm looking for advice on constructing a desktop rig for open source S/W development and testing.
The software in question isn't large at all (~100K SLOC), and can be built quickly, even in a fairly modest VM - CPU isn't an issue. I'm currently thinking either i7 Alder Lake or Ryzen 7000 65W CPU would be more than plenty, with no overclocking.
The difficulty is a requirement for encoding H.264 and AV1 on NVidia/Intel hardware, using NVENV and VA-API. The AV1 encoding in particular will need access to a recent NVidia card and an ARC Alchemist card.
I'm not planning on using these GPUs for anything other than encoding/decoding, and I'll only be using one at a time. I don't think each one will need 16 PCI lanes assigned - 4 or 8 should be fine. So I'm not looking at EPYC/TR/Xeon solutions. I'd like a single box solution if possible, as the home office is somewhat cramped.
My current workflows use a lot of VMs running on KVM. This gives me flexibility to build/test and triage user problems on a variety of platforms (not just Linux). To keep the same kind of workflows, I'll need to pass both GPUs through to various VMs, as the need arises, while using an integrated GPU capability on the workstation desktop for development. From what I've read so far, I'll need to get up to speed on VFIO and IOMMU groups to do this.
Given that I'm not a large corporation, and I don't want to make an expensive mistake, what pitfalls do I need to avoid when specifying a CPU and motherboard which will support two heterogenous cards and PCI pass-through?
I'm looking for advice on constructing a desktop rig for open source S/W development and testing.
The software in question isn't large at all (~100K SLOC), and can be built quickly, even in a fairly modest VM - CPU isn't an issue. I'm currently thinking either i7 Alder Lake or Ryzen 7000 65W CPU would be more than plenty, with no overclocking.
The difficulty is a requirement for encoding H.264 and AV1 on NVidia/Intel hardware, using NVENV and VA-API. The AV1 encoding in particular will need access to a recent NVidia card and an ARC Alchemist card.
I'm not planning on using these GPUs for anything other than encoding/decoding, and I'll only be using one at a time. I don't think each one will need 16 PCI lanes assigned - 4 or 8 should be fine. So I'm not looking at EPYC/TR/Xeon solutions. I'd like a single box solution if possible, as the home office is somewhat cramped.
My current workflows use a lot of VMs running on KVM. This gives me flexibility to build/test and triage user problems on a variety of platforms (not just Linux). To keep the same kind of workflows, I'll need to pass both GPUs through to various VMs, as the need arises, while using an integrated GPU capability on the workstation desktop for development. From what I've read so far, I'll need to get up to speed on VFIO and IOMMU groups to do this.
Given that I'm not a large corporation, and I don't want to make an expensive mistake, what pitfalls do I need to avoid when specifying a CPU and motherboard which will support two heterogenous cards and PCI pass-through?