3 ms·
AMD was but focused on hpc instead on ml. They have very large installations, but the card is tuned for fp32+ calculations not lower precisions like NVIDIA.
by gpapilion 3y ago
AMD was but focused on hpc instead on ml. They have very large installations, but the card is tuned for fp32+ calculations not lower precisions like NVIDIA.
- ethbr1 3y agoThe other side effect of HPC, I assume, is that compatibility and dev tooling are non-concerns. You provide a BLAS implementation, and your primary customer can take it from there. As opposed to Nvidia's current market, which spans HPC to AWS/Azure/GCP to corporate datacenter to "I followed the directions on Reddit" personal machines.