Hardware and software requirements¶
cuFFTMp relies on NVSHMEM for GPU-initiated communications. It inherits many of its requirements.
x86_64 or ppc64le CPU architecture
NVIDIA data center GPUs, of the Volta, Ampere or Hopper architecture
All GPUs should be
or connected via GPUDirect RDMA over Infiniband using a Mellanox adapter (CX-4 or later)
Note that being merely connected via PCIe does not guarantee peer-to-peer access between GPUs. Also note that NVLink is not required for peer-to-peer access, but it is the preferred transport.
cuFFTMp is shipped as part of the HPC SDK and requires
Linux 64 bits
CUDA 11.0 or later