Published on September 29, 2026, PTXBench evaluates model-generated matrix multiplication and attention kernels in architecture-specific PTX for H100 and B200 GPUs.
Published on September 29, 2026, PTXBench tests models that write matrix multiplication and attention kernels in architecture-specific PTX for H100 and B200 GPUs. Its evaluation measures correctness, speed, and whether the requested instructions are actually executed.
The benchmark aims to help engineers identify where AI-generated low-level GPU kernels fall short of specialized libraries. Consult the original publication for benchmark details and findings; also check its methodology and test conditions before comparing performance or applying conclusions to other environments.