List your company
ResearchFormerly WizWand

Sparse GEMM Performance on Matrix Dimensions (M, N, K) H20 GPU

0.135Runtime (ms)RT-Lynx-Kernel
-0.33762.852456.04259.23255May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
RT-Lynx-KernelM=N=2048, K=3072, GPU=H202026.050.1351.478.33
Pytorch-GEMMM=N=2048, K=3072, GPU=H202026.050.199——
CUTLASSM=N=2048, K=3072, GPU=H202026.050.2160.92—
Pytorch-SpMMM=N=2048, K=3072, GPU=H202026.050.2210.9—
cuSparseLtM=N=2048, K=3072, GPU=H202026.050.2820.71—
RT-Lynx-KernelM=N=2048, K=12288, GPU=H202026.050.4541.729.39
RT-Lynx-KernelM=N=4096, K=3072, GPU=H202026.050.4651.684.6
CUTLASSM=N=4096, K=3072, GPU=H202026.050.6161.26—
cuSparseLtM=N=4096, K=3072, GPU=H202026.050.7091.1—
Pytorch-SpMMM=N=4096, K=3072, GPU=H202026.050.7151.09—
Pytorch-GEMMM=N=4096, K=3072, GPU=H202026.050.781——
Pytorch-GEMMM=N=2048, K=12288, GPU=H202026.050.781——
CUTLASSM=N=2048, K=12288, GPU=H202026.050.8010.97—
cuSparseLtM=N=2048, K=12288, GPU=H202026.051.1370.69—
Pytorch-SpMMM=N=2048, K=12288, GPU=H202026.051.2590.62—
RT-Lynx-KernelM=N=4096, K=12288, GPU=H202026.051.6521.884.83
RT-Lynx-KernelM=N=8192, K=3072, GPU=H202026.051.8021.662.28
CUTLASSM=N=8192, K=3072, GPU=H202026.052.0441.46—
cuSparseLtM=N=8192, K=3072, GPU=H202026.052.0661.5—
Pytorch-SpMMM=N=8192, K=3072, GPU=H202026.052.0871.43—
CUTLASSM=N=4096, K=12288, GPU=H202026.052.2871.35—
Pytorch-SpMMM=N=4096, K=12288, GPU=H202026.052.6481.17—
cuSparseLtM=N=4096, K=12288, GPU=H202026.052.6691.16—
Pytorch-GEMMM=N=8192, K=3072, GPU=H202026.052.994——
Pytorch-GEMMM=N=4096, K=12288, GPU=H202026.053.099——
RT-Lynx-KernelM=N=8192, K=12288, GPU=H202026.056.7541.772.37
CUTLASSM=N=8192, K=12288, GPU=H202026.057.4971.59—
cuSparseLtM=N=8192, K=12288, GPU=H202026.058.2021.45—
Pytorch-SpMMM=N=8192, K=12288, GPU=H202026.058.2171.45—
Pytorch-GEMMM=N=8192, K=12288, GPU=H202026.0511.95——