sglang/ GB200 perf

dsr1-fp8-1k1k-ultra-tpt

concurrency 4,096 · 22 hours ago

cron
passed
re-run
Commit

[AMD] Register DP attention test (#28495)

bingxche·23 hours ago
PR #28495 · [AMD] Register DP attention test
Run
GitHub Actions27927465630-1
Slurm job5330
GPUs16·prefill 8 / decode 8
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

40,960

duration

616

max_concurrency

4,096

mean_e2el_ms

59,553ms

mean_itl_ms

2,953ms

mean_tpot_ms

60.49ms

mean_ttft_ms

3,879ms

median_e2el_ms

58,077ms

median_itl_ms

3,064ms

median_tpot_ms

61.64ms

median_ttft_ms

1,101ms

num_prompts

40,960

output_throughput

61,235tok/s

p99_e2el_ms

98,087ms

p99_itl_ms

3,748ms

p99_tpot_ms

62.60ms

p99_ttft_ms

41,286ms

peak_output_tokens_per_s

71,270s

request_throughput

66.46tok/s

std_e2el_ms

9,073ms

std_itl_ms

495ms

std_tpot_ms

3.53ms

std_ttft_ms

8,455ms

total_input_tokens

37,769,666

total_output_tokens

37,742,239

total_token_throughput

122,514tok/s