Activation Function GPU Kernel Synthesis on KernelBench (test)
1.01ReLU PerformanceCuTeGen
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CuTeGenGPU=NVIDIA GeForce RTX 4090 (24 GB), CUDA version=13.0, PyTorch version=2.8.0, CUTLASS version=v4.3.0 (commit acb4593), LLM Engine=GPT-5, Evaluation Protocol=Relative to PyTorch reference implementation2026.04 | 1.01 | 1 | 1 | 1 | 0.88 | 0.87 | 2.45 | 1.01 | 0.99 | 0.94 | 1.01 | 3.45 | 0.99 | 0.99 |