sglang/ GB200 perf

dsr1-fp8-1k1k-ultra-tpt

concurrency 4,096 · 4 days ago

cron
passed
re-run
Commit

[cookbook] Laguna-M.1: add PD disaggregation section (#28737)

Jiminator·4 days ago
PR #28737 · [cookbook] Laguna-M.1: add PD disaggregation section
Run
GitHub Actions27858605416-1
Slurm job5325
GPUs16·prefill 8 / decode 8
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

40,960

duration

621

max_concurrency

4,096

mean_e2el_ms

60,000ms

mean_itl_ms

2,972ms

mean_tpot_ms

60.90ms

mean_ttft_ms

3,947ms

median_e2el_ms

58,476ms

median_itl_ms

3,081ms

median_tpot_ms

62.09ms

median_ttft_ms

1,105ms

num_prompts

40,960

output_throughput

60,806tok/s

p99_e2el_ms

99,116ms

p99_itl_ms

3,819ms

p99_tpot_ms

63.03ms

p99_ttft_ms

42,002ms

peak_output_tokens_per_s

71,416s

request_throughput

65.99tok/s

std_e2el_ms

9,239ms

std_itl_ms

508ms

std_tpot_ms

3.57ms

std_ttft_ms

8,683ms

total_input_tokens

37,769,666

total_output_tokens

37,742,239

total_token_throughput

121,656tok/s