b78588d163
* CUTLASS 3.7 * clean up changelog --------- Co-authored-by: yuzhai <yuzhai@nvidia.com> Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
2.6 KiB
Distributed GEMM
Requirements
Build
Make sure to set up CUTLASS with
support for Programmatic Dependent Launch (PDL),
that is with the CUTLASS_ENABLE_GDC_FOR_SM90 flag.
cmake $PATH -DCUTLASS_NVCC_ARCHS="90a" -DCUTLASS_ENABLE_GDC_FOR_SM90=1
Minimum software
Like all other CUTLASS examples, the NVIDIA driver, runtime, and CUDA Toolkit are required. This example specifically requires CUDA Toolkit 12.6 or newer, due to some of the necessary CUDA graph APIs.
Hardware / driver settings
This example requires Hopper GPUs with NVLink network.
If you're not sure, first run the following command and make sure your GPU compute capability is 9.0:
nvidia-smi --query-gpu=name,compute_cap --format=csv
Sample output:
name, compute_cap
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
NVIDIA H100 80GB HBM3, 9.0
Then you should make sure there is an NVLink network by checking the GPU network topology,
and making sure there's NV* links between every pair of GPUs:
nvidia-smi topo -m
Sample output:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X
Finally, check if the driver enables peer to peer access, which should usually be the case, but it's good to check anyway:
nvidia-smi topo -p2p r
Sample output:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7
GPU0 X OK OK OK OK OK OK OK
GPU1 OK X OK OK OK OK OK OK
GPU2 OK OK X OK OK OK OK OK
GPU3 OK OK OK X OK OK OK OK
GPU4 OK OK OK OK X OK OK OK
GPU5 OK OK OK OK OK X OK OK
GPU6 OK OK OK OK OK OK X OK
GPU7 OK OK OK OK OK OK OK X