sglang/ GB200 perf

dsr1-fp8-1k1k-max-tpt

concurrency 4,096 · 8 days ago

cron
passed
re-run
Commit

Reenable MNNVL backend for FlashInfer allreduce fusion (#23402)

wenscarl·8 days ago
PR #23402 · Reenable MNNVL backend for FlashInfer allreduce fusion
Run
GitHub Actions27591824405-1
Slurm job5311
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

40,960

duration

350

max_concurrency

4,096

mean_e2el_ms

33,609ms

mean_itl_ms

1,633ms

mean_tpot_ms

33.45ms

mean_ttft_ms

2,824ms

median_e2el_ms

32,705ms

median_itl_ms

1,666ms

median_tpot_ms

33.88ms

median_ttft_ms

1,196ms

num_prompts

40,960

output_throughput

107,828tok/s

p99_e2el_ms

53,872ms

p99_itl_ms

2,633ms

p99_tpot_ms

35.19ms

p99_ttft_ms

22,889ms

peak_output_tokens_per_s

128,476s

request_throughput

117tok/s

std_e2el_ms

5,062ms

std_itl_ms

364ms

std_tpot_ms

1.64ms

std_ttft_ms

4,726ms

total_input_tokens

37,769,666

total_output_tokens

37,742,239

total_token_throughput

215,734tok/s