Goethe
Turn a Linux box with NVIDIA GPUs into a private, OpenAI-compatible endpoint
Emerged finds new products worth a look, daily — what’s new today
A Mac developer can run CUDA PyTorch code against a rented GPU without moving their whole environment onto that machine.
rGPU keeps your Python process local and sends GPU work to a remote NVIDIA machine. It offers two routes: a custom PyTorch device whose operations travel over TCP, and shims that intercept libcuda, the CUDA runtime, cuBLAS, cuBLASLt and cuDNN so unmodified Linux CUDA programs, including stock PyTorch, run against the remote card. Connections are neither authenticated nor encrypted, so the server belongs on localhost behind an SSH tunnel. No benchmarks yet.
More like this
nvkvm-pv
Share one NVIDIA GPU between the host and several KVM guests, no passthrough
Petal
Share single windows and drive them together, from CoScreen's own co-founder
Phobos
A Rust kernel language that compiles its own GPU code and serves LLMs with it
XConnect
SSH, SFTP and RDP in one client for phone and desktop, no account needed
OpenRod
Give each coding agent its own VM and its own network policy