title: the same kit run at one user with graphs off
source: manuals/sglang-101/capture/gpu/out-gb10/no-cuda-graph-bench-random-c1.txt
lines: 34-51
lang: text
note: Compare the median TPOT and the P99 ITL with the run before.
---
---------------Time to First Token----------------
Mean TTFT (ms):                          263.80    
Median TTFT (ms):                        291.11    
P90 TTFT (ms):                           354.70    
P95 TTFT (ms):                           373.75    
P99 TTFT (ms):                           413.63    
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          31.29     
Median TPOT (ms):                        31.29     
P90 TPOT (ms):                           31.37     
P95 TPOT (ms):                           31.46     
P99 TPOT (ms):                           31.60     
---------------Inter-Token Latency----------------
Mean ITL (ms):                           31.29     
Median ITL (ms):                         31.40     
P90 ITL (ms):                            32.73     
P95 ITL (ms):                            33.31     
P99 ITL (ms):                            41.72     
