4 ms·
Taichi, benchmarked in the article, claims to be able to outperform CUDA at some GPU tasks, although their benchmarks look to be a few years old: https://githu
by bloaf 7mo ago
Taichi, benchmarked in the article, claims to be able to outperform CUDA at some GPU tasks, although their benchmarks look to be a few years old:
https://github.com/taichi-dev/taichi_benchmark https://github.com/taichi-dev/taichi_benchmark
- pjmlp 7mo agoAnd doesn't account for cuTitle, NVidia's new API infrastructure that supports writing CUDA directly in Python via a JIT that is based on MLIR.