NVIDIA GB300 NVL72 8-GPU slice (2 nodes), MLPerf Inference v6.1 Server: Llama 2 70B under the Server latency constraint

15996 tok/s per GPU under a latency guarantee, still ahead of the full rack's Server figure per GPU (13124). The 2.7% Offline-to-Server discount here sits in the same band as the single-node B300 node's 2.2% and AMD's MI355X node's 3.0%, which is reassuring: at 8-GPU scale the scenario discount IS roughly stable across implementations, and it is only at rack scale that it becomes unstable in both sign and size. INTERCONNECT IDS FILLED 2026-10-04: interconnect_ids was empty; it now carries the interconnect list of the accelerator record(s) this run names (nvidia-b300). BASIS AND ITS LIMITS, stated because this field is easy to over-read: it records the FABRIC PRESENT IN THE HARDWARE CONFIGURATION, which is how every previously-filled record in this repo reads it, not traffic measured over that fabric. The clearest evidence for that convention is that single-GPU records already carry a scale-up fabric here - trtllm-llama31-8b-fp8-h100-1k1k-tps is 1x H100 at TP=1 and carries ['nvlink-4','nvswitch-4'] - and that host-link-only records carry a PCIe value (arc-pro-b70-mlperf-v6-1-llama2-70b-offline-tokens carries ['pcie-gen5'] on 4x Arc Pro B70, which has no GPU-to-GPU fabric at all). So a value here does NOT mean the interconnect was exercised by this particular workload, and must not be cited as evidence that it was. Whether the run actually crossed that fabric is a question about TP/EP/PP size, which this repo records in methodology, not here. NOT INVENTED: nothing was filled where the accelerator record itself carries no interconnect, so this field stays empty for any part whose own record has none - which is the honest state for a single-socket Apple sOC and for parts whose vendor publishes no host interface.