Closed jcoffland closed 4 months ago
There is CUDA_VISIBLE_DEVICES
which can control which devices are visible to CUDA but it also affects OpenCL and so does not help in this case.
https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars
Support for passing extra options directly to the core is not planned. However, v8.4.0 alpha implements an option to disable CUDA.
v7 allowed for this. May or may not be needed for core testing.
Specifically
--disable-cuda
may be needed for core 0x23. But can this be handled instead with an environment variable?