> On 26. May 2021, at 12:05, Ole Holm Nielsen <[email protected]> > wrote: > > I'm trying to build TensorFlow-2.4.1-fosscuda-2020b.eb on a GPU-equipped node > using the latest EB 4.3.4, and I get this warning: > > (lines deleted) >> == installing extension TensorFlow 2.4.1 (28/28)... >> == configuring... >> WARNING: No CUDA compute capabilities specified, so using TensorFlow default >> (which may not be optimal for your system). >> You should use the --cuda-compute-capabilities configuration option or the >> cuda_compute_capabilities easyconfig parameter to specify a list of CUDA >> compute capabilities to compile with. > > > So I'm trying to find out the meaning and use of > "--cuda-compute-capabilities", and the documentation at > https://docs.easybuild.io/en/latest/version-specific/help.html says: > >> --cuda-compute-capabilities=CUDA-COMPUTE-CAPABILITIES List of CUDA >> compute capabilities to use when building GPU software; values should be >> specified as digits separated by a dot, for example: 3.5,5.0,7.2 (type >> comma-separated list) > > This makes me none the wiser! Can anyone tell me what these numeric values > are supposed to mean, and how I pick the right values for the GPUs in my > nodes?
Hi Ole! Those numbers are related to which kind of gpu you have. There’s a good table here: https://en.wikipedia.org/wiki/CUDA
signature.asc
Description: Message signed with OpenPGP
