sglang/ GB200 perf

dsr1-fp4-1k1k-mid-curve

concurrency 4,096 · 16 days ago

cron
passed
re-run
Commit

[XPU CI] Pull prebuilt nightly image instead of building per-stage (#27282)

arathi-hlab·16 days ago
PR #27282 · [XPU CI] Pull prebuilt nightly image instead of building per-stage
Run
GitHub Actions27114126202-1
Slurm job5127
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

40,960

duration

314

max_concurrency

4,096

mean_e2el_ms

30,622ms

mean_itl_ms

1,180ms

mean_tpot_ms

24.08ms

mean_ttft_ms

8,460ms

median_e2el_ms

30,307ms

median_itl_ms

1,159ms

median_tpot_ms

24.18ms

median_ttft_ms

8,972ms

num_prompts

40,960

output_throughput

120,152tok/s

p99_e2el_ms

49,792ms

p99_itl_ms

2,357ms

p99_tpot_ms

29.53ms

p99_ttft_ms

23,337ms

peak_output_tokens_per_s

156,420s

request_throughput

130tok/s

std_e2el_ms

4,579ms

std_itl_ms

349ms

std_tpot_ms

3.38ms

std_ttft_ms

4,123ms

total_input_tokens

37,769,666

total_output_tokens

37,742,239

total_token_throughput

240,392tok/s