domains / families / devops / doca-gpunetio-ib-write-bw

doca-gpunetio-ib-write-bw

separate

Use this skill when the user is building or interpreting the doca/tools/gpunetio_ib_write_bw tool pair — a CUDA kernel posts RDMA WRITE work requests through the doca-gpunetio device-side surface to measure sustained GPU-driven WRITE BW on a GPU+IB-device pair. Trigger even without the tool name — typical phrasings: "measure WRITE BW when the GPU posts the WRs", "BW swings run to run on the same flags", "is the NIC saturated or am I bound on the CUDA kernel", "meson compile fails for the GPUNetIO bw tool", "nvidia_peermem isn't picking up my GPU buffer", or "GPU-posted WRITE BW vs the CPU path". Refuse and route elsewhere for general doca-gpunetio library work, DOCA setup, the GPU-posted WRITE latency analog, CPU-posted RDMA benchmarking (doca-perftest), and app-level end-to-end rates.

When to use

• measuring sustained RDMA WRITE bandwidth with GPU-posted work requests
• building the gpunetio_ib_write_bw tool pair
• interpreting GPU-driven WRITE BW results
• diagnosing bandwidth swings between runs
• identifying bottlenecks in GPU occupancy, NIC issue rate, or link saturation
• comparing GPU-posted WRITE BW against CPU-initiated perftest results
• debugging meson compile failures for the GPUNetIO bw tool
• troubleshooting nvidia_peermem GPU buffer issues

When not to use

• general doca-gpunetio library work
• DOCA setup
• GPU-posted WRITE latency analog
• CPU-posted RDMA benchmarking (doca-perftest)
• app-level end-to-end rates

Skill page

Source