Share one NVIDIA GPU between the host and several KVM guests, no passthrough
Emerged finds new products worth a look, daily — what’s new today
gputaskman
A terminal monitor that names the local models sitting in your video memory
Is this yours? Claim itWhen a local model runs slowly, this tells you which process holds the memory, which model it loaded and how much went to weights versus context cache.
gputaskman is for people running local models who want to know what is inside the card, not only how full it is. It lists memory and engine activity for each process, then works out which runtime and model each one has loaded, reading GGUF metadata and runtime logs to estimate weights and KV cache. It covers Windows with NVIDIA and AMD cards, native Linux, WSL and Apple Silicon, and installs with uv or pipx. Scott Hanselman calls it early and experimental: the breakdowns are estimates and the Windows builds are unsigned.
More like this
GPU specs and live rental prices from 24 clouds, side by side
Fly a camera past a Schwarzschild black hole and watch light bend, in a browser
How long until a local LLM machine pays for itself against API bills
A Linux desktop built on a Vulkan renderer of its own
See the compiled output of the function under your cursor without leaving Neovim