sglang/ GB200 perf

dsr1-fp8-1k1k-max-tpt

concurrency 1,024 · 1 day ago

cron
passed
re-run
Commit

[AMD] Register DP attention test (#28495)

bingxche·1 day ago
PR #28495 · [AMD] Register DP attention test
Run
GitHub Actions27927465630-1
Slurm job5331
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

10,240

duration

260

max_concurrency

1,024

mean_e2el_ms

24,933ms

mean_itl_ms

1,240ms

mean_tpot_ms

25.39ms

mean_ttft_ms

1,552ms

median_e2el_ms

24,529ms

median_itl_ms

1,271ms

median_tpot_ms

25.53ms

median_ttft_ms

937ms

num_prompts

10,240

output_throughput

36,323tok/s

p99_e2el_ms

33,738ms

p99_itl_ms

1,578ms

p99_tpot_ms

26.24ms

p99_ttft_ms

8,377ms

peak_output_tokens_per_s

43,198s

request_throughput

39.40tok/s

std_e2el_ms

2,572ms

std_itl_ms

211ms

std_tpot_ms

0.76ms

std_ttft_ms

1,939ms

total_input_tokens

9,440,068

total_output_tokens

9,440,624

total_token_throughput

72,644tok/s