3 ms·
I'm no expert, but I think that a lot of the GPU speedup has come through massively parallel execution rather than processor/bus/memory speed. NVidia's Quadro T
by gramie 6y ago
I'm no expert, but I think that a lot of the GPU speedup has come through massively parallel execution rather than processor/bus/memory speed. NVidia's Quadro TRX 8000, for example, has 4,608 CUDA cores, 576 Tensor cores, and 72 RT cores. I don't know what all those do, but they are certainly doing a lot of it in parallel!
- CyberDildonics 6y agoFirst, massively parallel execution takes more transistors and moore's law is about more transistors. Second, memory bandwidth has increased significantly.