When I check GPUDirect RDMA capability on uni-processor environment, nvidia-smi shows a strange result.
[root@magro ~]# nvidia-smi topo -mp
GPU0 mlx5_0 mlx5_1 CPU Affinity
GPU0 X SYS SYS 0-23
mlx5_0 SYS X PIX
mlx5_1 SYS PIX X
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe switches (without traversing the PCIe Host Bridge)
PIX = Connection traversing a single PCIe switch
I use SYS-1029GP-TT that support 1x Skylake-SP processor which also controls the entire PCIe-bus by its built-in controller.
https://www.supermicro.com/products/system/1U/1019/SYS-1019GP-TT.cfm
So, it should not have any chance for QPI traversal, however, nvidia-smi says the connection between GPU0 and mlx5_X traverse SMP connection.
Any ideas?