Intel Gaudi 3, Optimum-Habana (sw 1.24): Llama-3.1-8B-Instruct FP8 1HPU bs768 128in/2048out aggregate tok/s
19242 output tok/s AGGREGATE across 1x Gaudi 3 (batch 768; ~19242 tok/s per HPU)
Fields
id
benchmarks/gaudi3-1hpu-llama31-8b-fp8-128-2048Metric
tps_aggregate
Value
19,242
Unit
output tok/s AGGREGATE across 1x Gaudi 3 (batch 768; ~19242 tok/s per HPU)
Model
Llama-3.1-8B-Instruct
Methodology
VENDOR-MEASURED (Intel model-performance table, Gaudi software 1.24, Optimum-Habana + Hugging Face model, maxabs FP8). Llama-3.1-8B-Instruct FP8 on 1 HPU, batch_size=768, input_len=128, output_len=2048. MAX THROUGHPUT under a large batch, NOT a single-user decode figure. ~19242 tok/s per HPU.
Measured by
vendor
Reproducible
yes
notes
Optimum-Habana run_generation.py path, use_hpu_graphs + flash_attention_recompute. Sibling rows from the same table were already ingested; these are the previously unrecorded input/output/HPU combinations.
Notes
Optimum-Habana run_generation.py path, use_hpu_graphs + flash_attention_recompute. Sibling rows from the same table were already ingested; these are the previously unrecorded input/output/HPU combinations.