How to get the topology of GPU devices?

Question

I want to write a function that returns the topology of the underlying GPU devices as a graph. I want the connections to indicate where data transfer can occur, and the weights to be the throughput capacity of these connections.

I know that nvidia-smi topo -m returns an adjacency matrix which the GPUs as vertices, but it doesn't show the throughput capacity of these connections. How would I go about getting this information?

This is the output of the nvidia-smi command.

    GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    CPU Affinity    NUMA Affinity
GPU0     X  PIX PIX SYS SYS SYS 0-11    0
GPU1    PIX  X  PIX SYS SYS SYS 0-11    0
GPU2    PIX PIX  X  SYS SYS SYS 0-11    0
GPU3    SYS SYS SYS  X  PIX PIX 12-23   1
GPU4    SYS SYS SYS PIX  X  PIX 12-23   1
GPU5    SYS SYS SYS PIX PIX  X  12-23   1

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 bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks