Argonne LLM-Inference-Bench: SambaNova SN40L x8, SambaFlow, Mistral-7B BF16, 128in/128out, batch 1
1226.4 tok/s aggregate (batch 1 single request on 8-RDU node)
Fields
id
benchmarks/argonne-sn40l-sambaflow-mistral-7b-8x-b1-128io-tpsMetric
tps_aggregate
Value
1,226.4
Unit
tok/s aggregate (batch 1 single request on 8-RDU node)
Model
mistralai/Mistral-7B-v0.1
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench, arXiv:2411.00136, All_results.csv). FP16/BF16 weights, 128 in / 128 out. Throughput = batch*(in+out)/end-to-end latency = AGGREGATE over the batch, NOT per-request decode tok/s. Batch 1, 8x SN40L RDU node (512GB aggregate DDR). Latency 0.209s.
Measured by
third_party
Reproducible
no
notes
SN40L runs models compiled through SambaFlow; batch-1 latency 0.21s for 128-out is the dataflow architecture's signature — ~610 tok/s effective single-request decode.
Notes
SN40L runs models compiled through SambaFlow; batch-1 latency 0.21s for 128-out is the dataflow architecture's signature — ~610 tok/s effective single-request decode.