verifiedmedium · 0.80updated 2026-10-14

MLPerf v5.1 MiTAC 8x MI325X: llama2-70b-99.9 Offline aggregate tok/s

33703 tok/s AGGREGATE over 8x MI325X 256GB

Fields

benchmarks/mlperf-v5-1-mitac-8xmi325x-llama2-70b-offline-tokens

Metric

tps_aggregate

Value

33,703

Unit

tok/s AGGREGATE over 8x MI325X 256GB

Model

llama2-70b-99.9

Methodology

MLPerf Inference closed-division submission. v5.1, MiTAC, 8x AMD Instinct MI325X 256GB HBM3e + 2x EPYC 9755. Offline: 'Tokens per second: 33703', VALID. Values quoted verbatim from the submission's mlperf_log_summary.txt (performance/run_1). AGGREGATE throughput across all accelerators in the system, NOT per-request.

Measured by

vendor

Reproducible

yes

Cross-references

MLPerf v5.1 MiTAC 8x MI325X: llama2-70b-99.9 Offline aggregate tok/s

Fields

Cross-references

Accelerators (1)

Engine (1)

Interconnect (1)