sglang/ GB200 perf

dsr1-fp8-1k1k-max-tpt

concurrency 6,144 · 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

61,440

duration

447

max_concurrency

6,144

mean_e2el_ms

43,251ms

mean_itl_ms

1,837ms

mean_tpot_ms

37.63ms

mean_ttft_ms

8,615ms

median_e2el_ms

42,466ms

median_itl_ms

1,798ms

median_tpot_ms

38.60ms

median_ttft_ms

8,097ms

num_prompts

61,440

output_throughput

126,614tok/s

p99_e2el_ms

70,996ms

p99_itl_ms

3,351ms

p99_tpot_ms

42.34ms

p99_ttft_ms

33,079ms

peak_output_tokens_per_s

175,432s

request_throughput

137tok/s

std_e2el_ms

6,913ms

std_itl_ms

554ms

std_tpot_ms

4.36ms

std_ttft_ms

6,016ms

total_input_tokens

56,636,934

total_output_tokens

56,621,450

total_token_throughput

253,263tok/s