Currently we build with cuda toolkit 10.2. The PTX assembly is included so the code should run on more recent GPUs. But it would be nice to offer more versions.
I would suggest we add in or replace it with 12.3 as a more recent version. Adding an additional build is only possible if we automate the build process.
If we add this to the conda-feedstock I believe every minor version is compiled.
Currently we build with cuda toolkit 10.2. The PTX assembly is included so the code should run on more recent GPUs. But it would be nice to offer more versions.
I would suggest we add in or replace it with 12.3 as a more recent version. Adding an additional build is only possible if we automate the build process.
If we add this to the conda-feedstock I believe every minor version is compiled.