huixiancheng / CENet

[ICME 2022] CENet: Toward Concise and Efficient LiDAR Semantic Segmentation for Autonomous Driving
MIT License
99 stars 13 forks source link

number of gpu and How much cuda memory is needed #27

Closed mooncakehub closed 7 months ago

mooncakehub commented 1 year ago

You mentioned in the article that RTX 3060 and RTX3090,Are they used at the same time. how much cuda memory is needed

huixiancheng commented 7 months ago

No. Use 3060 to train model with 64x512. Use 3090 to train model with 64x1024 & 64x2048.
For both 64x512 and 64x2048, our batch size make full use of all GPU memory.