sglang/ GB200 perf

dsr1-fp8-1k1k-max-tpt

concurrency 1,024 · 8 days ago

cron
passed
re-run
Commit

Reenable MNNVL backend for FlashInfer allreduce fusion (#23402)

wenscarl·8 days ago
PR #23402 · Reenable MNNVL backend for FlashInfer allreduce fusion
Run
GitHub Actions27591824405-1
Slurm job5311
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

10,240

duration

259

max_concurrency

1,024

mean_e2el_ms

24,839ms

mean_itl_ms

1,236ms

mean_tpot_ms

25.30ms

mean_ttft_ms

1,542ms

median_e2el_ms

24,633ms

median_itl_ms

1,269ms

median_tpot_ms

25.33ms

median_ttft_ms

950ms

num_prompts

10,240

output_throughput

36,449tok/s

p99_e2el_ms

31,507ms

p99_itl_ms

1,625ms

p99_tpot_ms

26.09ms

p99_ttft_ms

7,235ms

peak_output_tokens_per_s

43,910s

request_throughput

39.54tok/s

std_e2el_ms

2,158ms

std_itl_ms

191ms

std_tpot_ms

0.82ms

std_ttft_ms

1,714ms

total_input_tokens

9,440,068

total_output_tokens

9,440,624

total_token_throughput

72,897tok/s