Phobos
A Rust kernel language that compiles its own GPU code and serves LLMs with it
Emerged finds new products worth a look, daily — what’s new today
nvkvm-pv
Share one NVIDIA GPU between the host and several KVM guests, no passthrough
If you run local models or CUDA work in VMs, this lets guests use the GPU at near bare-metal speed while your desktop keeps it too.
nvkvm-pv is a paravirtualised NVIDIA GPU for QEMU-KVM. A guest kernel module forwards the NVIDIA driver's control calls over virtio to a host process, while the CUDA hot path maps device memory directly, so unmodified CUDA, PyTorch and Vulkan run at 98–99.9% of bare metal. No VFIO or vGPU licence is needed, and the host and multiple VMs share the card. Tested from Turing to Blackwell. Experimental and not a hardened multi-tenant sandbox. Apache-2.0, guest module GPL-2.0.
More like this
Goethe
Turn a Linux box with NVIDIA GPUs into a private, OpenAI-compatible endpoint
rGPU
Write PyTorch on your laptop while the CUDA work runs on a remote NVIDIA box
try-omarchy
Run the Omarchy Linux desktop as a native app on Apple Silicon
ZSvirt
Open-source virtualisation platform from ZStack, with VMware migration built in
Try Omarchy for Windows
Run the Omarchy Linux desktop in a Windows window without a dual boot