cuda.bindings.nvml.device_get_gpu_fabric_info_v4#

cuda.bindings.nvml.device_get_gpu_fabric_info_v4(intptr_t device)#

Retrieves GPU fabric information including per-type clique assignments.

Returns fabric clique data via nvmlGpuFabricInfo_v4_t. Each entry in the cliques array is a (type, id) pair representing a single clique assignment. The number of valid entries is given by numCliques. Entries are sorted by ascending type (NVML_GPU_FABRIC_CLIQUE_TYPE_*), then by ascending clique id within each type. On Hopper systems, the driver reports Unicast Pointer and Multicast Pointer cliques. On Blackwell and Rubin, Unicast Logical Endpoint and Multicast Logical Endpoint are additionally reported. **View CUDA Toolkit Documentation for a C++ code example**. For Hopper ™️ or newer fully supported devices.

Parameters:

device (intptr_t) – The identifier of the target device.

Returns:

Information about GPU fabric state: including per-type cliques.

Return type:

nvmlGpuFabricInfo_v4_t