4 ms·
The language itself? In broad strokes, not really. The typical tooling? Definitely yes, it's usually very HPC centric (MPI, OpenMP, batch job schedulers, ...).
by reikonomusha 3y ago
The language itself? In broad strokes, not really. The typical tooling? Definitely yes, it's usually very HPC centric (MPI, OpenMP, batch job schedulers, ...).
- soco 3y agoI assume all this specialized tooling will be a hurdle in the learning process. It's not like you can spin up a docker image and hack on.
- sampo 3y agoPerhaps in the future, learning how to run native binaries on bare metal, will be viewed as learning specialized tooling.
- 8organicbits 3y agoI haven't used it, but shifter looks like an MLI on docker toolset. https://docs.nersc.gov/development/shifter/examples/ https://docs.nersc.gov/development/shifter/examples/
- 8organicbits 3y agoPeople are reasonably expected to learn k8s or React on the job, and those have a learning curve as well. Are MPI/OpenMPI much more difficult than those? I suppose that's a hard comparison to make. At first glance it doesn't look too crazy to me. https://www.eecis.udel.edu/~youse/openmpi/#/7 https://www.eecis.udel.edu/~youse/openmpi/#/7
- mafuy 3y agoMPI is trivial to learn. CS students here are expected to use it for typical scientific tasks within a few hours of lessons. Complex use cases take some time to understand backgrounds (e.g. multi partition configurations, profiling, infiniband, ...) and it's possible to become a specialist, but that's very rarely needed. But MPI, in most use cases, comprises far less than React does, so it's not a fair comparison.