otop

Monitor NVIDIA GPUs, and the processes on them, in the terminal, like nvtop.

GPUs are read through NVML, loaded at run time from NVIDIA's driver (libnvidia-ml.so), and drawn with ratatui. Nothing in C is built and no CUDA toolkit is needed.

Each GPU gets a panel with utilisation and memory gauges, temperature, performance state, power, fan, clocks, PCIe throughput and encoder/decoder use. Below that, each GPU's utilisation and memory over time are charted: in two charts for each GPU, in one chart for each GPU, or in one chart of every GPU's utilisation and one of their memory, switched with v. Then a table of the processes on them: PID, GPU, compute or graphics, user, GPU memory, SM/encoder/decoder use, and command line.

Install

cargo install --git https://github.com/YPARK/otop
# or, from a clone
cargo install --path .

Use

otop             # sample every second
otop -d 500      # every 500 ms
otop --once      # print one sample as text, and exit