Most of the frameworks have been build with a specific CUDA version. SOL requires the CUDA 11.1 builds, as they are the only ones supporting the NVIDIA A100.
| Question | Answer |
|---|---|
| SOL tells me that support for NVIDIA is not available. | This is usually caused by a version of SOL without NVIDIA support. Please check if pip3 list installed | grep nec-sol-device-nvidia shows that it is installed, and has the same version as pip3 list installed | grep nec-sol. |
| Which GPUs are supported? | In general all CUDA capable GPU starting from the Kepler architecture (i.e. Tesla K40) are supported. |
I get “CUDA Out of Memory” errors when running sol.optimize(...) with autotuning enabled? |
This is cause by the fact that frameworks usually preallocate huge portions of the GPUs memory for themselves, leaving only little memory left for SOL to perform the autotuning. We are looking in hooking up into the framework’s memory allocation to solve this issue in a future release. |
| SOL does not load CUDNN. | We do not bundle CUDNN with SOL. Usually it is already installed within your system. If not, please run: nec-sol install cudnn or install v8.0.5 manually from https://developer.nvidia.com/cudnn (requires a free CUDA developer account). |