GPUM
Fork of GPU Monitoring Service by Minato3000, redesigned for machines with many GPUs.
Real-time NVIDIA GPU stats in VS Code. Works locally and over Remote-SSH.
Features
- Fits many GPUs on screen: one table row per GPU in a wide panel, compact three-line items in the sidebar
- Utilization, memory, memory bandwidth, power draw and temperature for every GPU
- Totals across all GPUs (memory, power); the model is shown once when all GPUs are identical
- Compute processes mapped to the GPU they run on
- Live clock in the top-left corner
- Native VS Code look that follows your color theme; warning and error colors only when memory, power or temperature run high
- Light on resources:
nvidia-smi runs asynchronously, unchanged data is not re-rendered, and polling stops while the view is hidden
Requirements
- NVIDIA GPU with
nvidia-smi installed and on PATH
- Linux or Windows with NVIDIA drivers
When connected to a remote machine, the extension runs on the remote host and shows that host's GPUs.
Settings
| Setting |
Default |
Description |
gpum.refreshInterval |
2000 |
Poll interval in ms |
gpum.nvidiaSmiPath |
nvidia-smi |
Path to the nvidia-smi executable |
Credits
Forked from GPU Monitoring Service by Minato3000 (MIT).
| |
