Argonne: PVC x2 vLLM Llama-2-7B FP16 128io b64
1894.5 tok/s aggregate (batch*(in+out)/latency)
Fields
id
benchmarks/argonne-pvc-vllm-llama2-7b-fp16-2x-b64-128io-tpsMetric
tps_aggregate
Value
1,894.5
Unit
tok/s aggregate (batch*(in+out)/latency)
Model
meta-llama/Llama-2-7b-hf
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench arXiv:2411.00136). vLLM 0.6.1 benchmark_throughput.py, FP16, 128in/128out, AGGREGATE over batch. 2 PVC tiles, bs64, lat 8.65s.
Measured by
third_party
Reproducible
yes
notes
CSV column 'Intel PVC GPU' = ALCF Aurora/Sunspot Intel Data Center GPU Max (Ponte Vecchio, filed under xpu-intel-dc-gpu-max-1550; the PVC product line maps to the Max series). vLLM xpu backend, 1/2/4-tile configurations. Two-tile scaling ~4.3x over 1 tile at bs64.
Notes
CSV column 'Intel PVC GPU' = ALCF Aurora/Sunspot Intel Data Center GPU Max (Ponte Vecchio, filed under xpu-intel-dc-gpu-max-1550; the PVC product line maps to the Max series). vLLM xpu backend, 1/2/4-tile configurations. Two-tile scaling ~4.3x over 1 tile at bs64.