sglang/ GB200 perf

dsr1-fp4-1k1k-mid-curve

concurrency 8,192 · 2 days ago

cron
passed
re-run
Commit

[AMD] Register DP attention test (#28495)

bingxche·2 days ago
PR #28495 · [AMD] Register DP attention test
Run
GitHub Actions27927465630-1
Slurm job5332
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

81,920

duration

705

max_concurrency

8,192

mean_e2el_ms

68,241ms

mean_itl_ms

1,186ms

mean_tpot_ms

24.21ms

mean_ttft_ms

45,955ms

median_e2el_ms

71,216ms

median_itl_ms

1,011ms

median_tpot_ms

21.97ms

median_ttft_ms

51,161ms

num_prompts

81,920

output_throughput

107,092tok/s

p99_e2el_ms

83,581ms

p99_itl_ms

3,413ms

p99_tpot_ms

42.66ms

p99_ttft_ms

59,022ms

peak_output_tokens_per_s

157,103s

request_throughput

116tok/s

std_e2el_ms

9,053ms

std_itl_ms

664ms

std_tpot_ms

7.39ms

std_ttft_ms

11,865ms

total_input_tokens

75,511,905

total_output_tokens

75,496,082

total_token_throughput

214,206tok/s