A KVM guest runs NVIDIA GPUs within 2% of bare metal
Virtio-nvgpu: Near-native Nvidia GPU access inside a KVM guest
virtio-nvgpu forwards NVIDIA kernel driver ioctls between a Linux guest and host at the driver ABI level, letting the guest run NVIDIA's own user-mode drivers unmodified. Benchmarks on an RTX 3060 show a guest renders within 2% of bare metal for frames above 2 ms, uses the same CPU, and four guests share one card evenly at 103.7 fps combined. The target is headless streaming: render, composite, and encode entirely on the GPU inside the guest.
A guest costs what the host costs.
- vsviridov
Can this be used with proxmox?
- mjg59
Hrm, something of a lack of discussion about what level of access the card has to the host in the absence of IOMMU-restricted passthrough.
- mcmatterson
Amazing project! But man, that README is just a textbook example of LLM word salad. It's wild how these tools are so incredibly capable at many things, but their writing sticks out like a sore thumb
- orphereus
Why not use normal GPU passthrough? I don't see how you can use this to share a GPU between multiple VMs, so what is the benefit of using this software over normal GPU passthrough with vfio-pci drivers?
- markasoftware
How is this different than gVisor's nvproxy? https://gvisor.dev/docs/user_guide/gpu/#compatibility
Edit: nvproxy is mentioned as the "direct inspiration" in the readme without mention of how this is different or why it doesn't use nvproxy as a backend.