Currently, our CUDA tests are rather memory hungry. This is fine for running the test suites using A100 GPUs.
However, we don't need "all that memory" for the purposes of running tests. I have local system with a 1 GB NVIDIA GPU, and the CUDA tests take almost 17 minutes to finish.
So, we should work on paring down the runtimes and/ or sizes of the tests.
Currently, our CUDA tests are rather memory hungry. This is fine for running the test suites using A100 GPUs.
However, we don't need "all that memory" for the purposes of running tests. I have local system with a 1 GB NVIDIA GPU, and the CUDA tests take almost 17 minutes to finish.
So, we should work on paring down the runtimes and/ or sizes of the tests.