search

GANWANSHUI / SimpleOccupancy

(IEEE TIV) A Comprehensive Framework for 3D Occupancy Estimation in Autonomous Driving
203 stars 11 forks source link

Question about GPU memory requiments for training and evaluation respectively? #9

Open CaptainEven opened 10 months ago

GANWANSHUI commented 10 months ago

We use 8 * A100(40 GB) for both training and evaluation. If you use Resnet 50 as the encoder, then V 100 (32 GB) is enough.

CaptainEven commented 10 months ago

We use 8 * A100(40 GB) for both training and evaluation. If you use Resnet 50 as the encoder, then V 100 (32 GB) is enough.

May i ask the minimum GPU memory requiments(both single GPU and multi GPU) to launch the training process?

GANWANSHUI commented 10 months ago

I have not tested the minimum memory, but on my side, 32GB on V100 is enough for resnet 50 encoder. One thing is about shm memory which you may increase this memory. You can share the detailed issues encountered.

CaptainEven commented 10 months ago

I have not tested the minimum memory, but on my side, 32GB on V100 is enough for resnet 50 encoder. One thing is about shm memory which you may increase this memory. You can share the detailed issues encountered.

OK, i got it, thanks for the excellent work and reply!