
53:58
Is there interest/effort into making a code base portable to alternative CPUs, GPUs, FPGAs? Easily - no changes to source?

59:55
For CPUs and GPUs at least that’s RAJA and Kokkos (somewhat like Mike says)

01:03:53
e.g., RAJA is making it easy to move LLNL codes from NVIDIA to AMD GPUs

01:03:56
The Programming Models and Runtimes SDK would be another place to look.

01:04:11
(or at least easier)

01:04:17
^^ what Tim said

01:04:40
FPGAs are more iffy. For instance, on Intel there is some OneAPI support (which Kokkos could in theory build on) but it isn’t complete.

01:05:23
FPGAs have always been a tricky beast because they are so different.

01:05:54
It can be much simpler…

01:13:45
Still specific to nersc — and local nersc uids

01:14:04
https://docs.nersc.gov/services/sfapi/

01:14:22
Condor can do that sso long as compute providers are willing to run the services.

01:14:22
Through no fault of NERSC’s — they can’t make everyone use their logins

01:14:27
DOE could though

01:15:10
it would make things a lot easier if we could come up with a “lowest common denominator” across doe for such an API — and grow this from there