Low GEMM Performance on Hopper GPU with Small M Shapes

Hi,

Thank you for the great library! I’m observing some unexpected performance with GEMM on Hopper GPUs when using small M dimensions. I followed the example in example14_autotune.py.

Compared to the PyTorch implementation, the performance is significantly lower — around 30% of the expected TFLOPS and memory bandwidth utilization.

Not sure I am correctly using the API — I would greatly appreciate any suggestions.

Environment:
• GPU: H200
• CUDA: 12.8
• PyTorch: 2.6.0
• nvmath-python: 0.3.0

Benchmark code:

import argparse

import torch
import nvmath
from triton.testing import do_bench


def profile(m, n, k, dtype):
    device = torch.device("cuda")
    assert isinstance(device, torch.device)

    X = torch.randn(m, k, device=device, dtype=dtype)
    Y = torch.randn(n, k, device=device, dtype=dtype)
    
    _torch_gemm = lambda: torch.matmul(X, Y)
    
    mm = nvmath.linalg.advanced.Matmul(X, Y)
    
    mm.plan(preferences={"limit":1000})
    mm.autotune(iterations=1000)
    
    # print(mm.algorithms[0].capabilities)
    _nvmath_gemm = lambda: mm.execute()
    
    t_torch = do_bench(_torch_gemm)
    t_nvmath = do_bench(_nvmath_gemm)
    
    return t_torch, t_nvmath


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="GEMM profile")
    parser.add_argument("--m", type=int, default=4096)
    parser.add_argument("--n", type=int, default=4096)
    parser.add_argument("--k", type=int, default=4096)
    args = parser.parse_args()

    print("Provider,Operation,dtype,m,n,k,Runtime,GB/s,GFLOPs")
    
    for dtype in [torch.float16, torch.bfloat16]:
        t_torch, t_nvmath = profile(args.m, args.n, args.k, dtype)

        m = args.m
        n = args.n
        k = args.k

        torch_mem_bd = 2 * (m * n + n * k + m * k) * 1e3 / t_torch / 1e9
        torch_gflops = 2 * m * n * k * 1e3 / t_torch / 1e9
        nv_mem_bd = 2 * (m * n + n * k + m * k) * 1e3 / t_nvmath / 1e9
        nv_gflops = 2 * m * n * k * 1e3 / t_nvmath / 1e9
        
        print(f"TORCH,0,{dtype},{args.m},{args.n},{args.k},{t_torch},{torch_mem_bd},{torch_gflops}")
        print(f"NVMATH,0,{dtype},{args.m},{args.n},{args.k},{t_nvmath},{nv_mem_bd},{nv_gflops}")

Riproduci il confronto usando examples/linalg/advanced/matmul/example14_autotune.py e il benchmark di profiling fornito nell’ambiente H200 e CUDA 12.8 indicato. Confronta la pianificazione e l’autotuning di nvmath Matmul con torch.matmul per forme M piccole; il lavoro è completato quando il divario di prestazioni è spiegato e viene documentato l’uso corretto dell’API o il fix richiesto.