ada_gemm_k64_warp32_async
Ada FP32 GEMM winner with 64x64x64 tiles, 32x8 threads, and cp.async double-buffered shared-memory stages.
This repository contains the standalone CUDA source for the ada_gemm_k64_warp32_async lane from
the PyC kernel lab. It is a source artifact for inspection and benchmarking;
it is not a precompiled binary and the result below is not a universal ranking.
Performance
| Kernel | GPU / architecture | Shape | Best recorded result | Evidence |
|---|---|---|---|---|
ada_gemm_k64_warp32_async |
not recorded | not recorded | Not measured in the published campaign | No published performance receipt was found for this lane. |
The result is reported with the original campaign's timing and correctness context. Compare kernels only when GPU, CUDA version, matrix shape, warmup, repeats, and reference/correctness mode match.
Source
kernel.cu— copied fromkernels/prototypes/ada/gemm_k64_warp32_async/kernel.cu.- Original lane tags:
cuda, matmul, ada, sm89, prototype, fp32, vectorized, winner, k64, warp32, async, cpasync.
Build/run contract
{nvcc} -O3 -std=c++17 -lineinfo -gencode arch=compute_89,code=sm_89 -gencode arch=compute_89,code=compute_89 {source} -o {build_dir}/{name}
{build_dir}/{name} 1024 1024 1024 10 50
- Downloads last month
- -
cuda
kernel
gpu-optimization
hpc
apache-2.0