search

YijinHuang / SSiT

SSiT: Saliency-guided Self-supervised Image Transformer for Diabetic Retinopathy Grading
23 stars 4 forks source link

Are four V100 (16 GB memory per GPU) adequate to run the code from scratch using DDP? #12

Closed osivaz61 closed 8 months ago

YijinHuang commented 8 months ago

Thank you for your question. In our training setting, 64GB GPU is sufficient to train SSiT.

osivaz61 commented 8 months ago

You said that 64GB is sufficient. However, there is a difference between 4 V100(16GB per GPU) and only one GPU that has 64GB GPU memory.