cuda-api-wrappers
cuda-api-wrappers copied to clipboard
Add support for NVML functionality
One NVIDIA's libraries, which we currently ignore completely, is NVML - The NVIDIA Management Library. It allows access to a bunch of meta-data which we currently fully access - neither through the driver nor through the runtime API; and the ability to make some settings which otherwise we can't make. This includes:
- ECC memory: Error and correction statistics and ECC dis/activation.
- GPU resource utilization stats reported by the cards.
- OS processes actively using the GPUs.
- Clocks and "performance-state" info
- Temperature and fan speeds.
- Power draw and limits
- More identification info: Versions, serial numbers, BIOS versions etc.