Scaling ML meeting

US/Central

Strong scaling seems to degrade performance, i.e., it takes longer to get the same model performance although more GPUs are being used. 

Mini-batch on even on GPU convergences significantly faster.

 

There are minutes attached to this event. Show them.