search

bytedance / lightseq

LightSeq: A High Performance Library for Sequence Processing and Generation
Other
3.22k stars 329 forks source link

Is it normal that A10 inference speed is lower than 2080ti? #523

Open qinbo23 opened 1 year ago

qinbo23 commented 1 year ago

hello？I tested the Transformer-base inference speed on different devices. It's weird that A10 speed is lower than 2080ti speed.

MODEL: Transformer-base DATA: fp16 SPEED: (number of src characters / second) 3090 7.5k/s 2080 4.5k/s A10 2.0K/s

Mr-wang2016 commented 11 months ago

me too. can anyone help?