🛠️ CUDA/graphics in QEMU-KVM VMs without passing the Nvidia card to them
Summary
This project allows QEMU-KVM virtual machines to run CUDA and graphics by forwarding the NVIDIA driver ioctl surface from the guest to the host without using VFIO or vGPU. It enables stock libcuda and real Vulkan ICD to run in the guest while the host desktop continues using the same GPU.
Why it’s interesting
It enables running real CUDA and Vulkan workloads in a VM while sharing a single NVIDIA GPU with the host desktop without VFIO or vGPU.
Target user
Virtual machine users who need CUDA and graphics support without dedicating a physical GPU.
Source metrics: Points 5 · Comments 2
HN discussion · Project
Source: #HackerNews / Show HN
Summary
This project allows QEMU-KVM virtual machines to run CUDA and graphics by forwarding the NVIDIA driver ioctl surface from the guest to the host without using VFIO or vGPU. It enables stock libcuda and real Vulkan ICD to run in the guest while the host desktop continues using the same GPU.
Why it’s interesting
It enables running real CUDA and Vulkan workloads in a VM while sharing a single NVIDIA GPU with the host desktop without VFIO or vGPU.
Target user
Virtual machine users who need CUDA and graphics support without dedicating a physical GPU.
Source metrics: Points 5 · Comments 2
HN discussion · Project
Source: #HackerNews / Show HN