Run busGrind -p 1 -u 0 -e 0 -d 1, the Concurrent Host/Device Bandwidth Matrix result is confusing.

run busGrind -p 1 -u 0 -e 0 -d 1, I got,
.......

Test Description: Bus bandwidth between the host and a single device

Host/Device Bandwidth Matrix (GB/s), memory=Pinned
Dir\D 0 1 2 3 4 5 6 7
D2H 56.83 57.12 57.14 57.15 55.37 55.47 55.49 53.53
H2D 56.17 56.21 56.21 56.20 56.17 56.17 56.12 56.14
BiDir 101.26 101.38 101.40 101.37 89.20 43.62 7.25 9.71

Test Description: Bus bandwidth between the host and multiple devices concurrently

Concurrent Host/Device Bandwidth Matrix (GB/s), memory=Pinned
Dir\D 0 1 2 3 4 5 6 7 Total
H2D 44.61 44.80 43.68 43.69 25.32 25.49 25.45 25.47 278.51
D2H 15.94 15.91 15.80 15.84 11.57 11.58 11.58 11.60 109.83
BiDir 22.55 22.76 22.62 22.64 10.04 4.40 17.99 18.03 141.02

As we can see, the BiDir result is confusing. why device 5 and device 7 have such low bandwidth? Is this expected?

BusGrind is a cuda demo suite tool. This might not be the right place to ask, but I couldn’t find the demo suite's repository.

Looking forward to your response.

Start by reproducing busGrind -p 1 -u 0 -e 0 -d 1 and capture the Concurrent Host/Device Bandwidth Matrix output. Trace the busGrind implementation that computes or prints the BiDir per-device values, then determine whether the low values for devices 5 and 7 are expected. Done means the behavior is explained or the output is corrected, with a verification case if the result is a bug.