MLPerf v6.1 MangoBoost heterogeneous 2x8 MI300X + 2x8 MI355X (LLMBoost P/D): gpt-oss-120b Offline 285454 tok/s
285454.0 tok/s aggregate Offline
Fields
id
benchmarks/mlperf-v6-1-mangoboost-het-mi300x-mi355x-gptoss-120b-offline-tokensMetric
tps_aggregate
Value
285,454
Unit
tok/s aggregate Offline
Model
gpt-oss-120b
Methodology
MLPerf Inference v6.1 published result (Heterogeneous 4-cluster: 2x8 MI300X 192GB + 2x8 MI355X 288GB, LLMBoost; software: ROCm / LLMBoost). System-AGGREGATE output tokens/s for the named scenario, submitter-measured under MLCommons audit. 32-GPU prefill/decode-disaggregated cluster; largest single gpt-oss-120b Offline figure in v6.1.
Measured by
vendor
Reproducible
yes
notes
ENGINE_ID NOTE: pinned to 'vllm' as the closest existing engine record; the submission's actual stack is named in this methodology string.
Notes
ENGINE_ID NOTE: pinned to 'vllm' as the closest existing engine record; the submission's actual stack is named in this methodology string.