Hi,all
I want to speed up for training time.
Now I’m using only one GPU.
Here is my nvidia-smi info.
Memory-Usage: 1143MiB / 12189MiB
Volatile GPU-Util: 45~60%
In this case, using multi-gpu can speed up my training time?
Training speed depends on many factors including the input loading time (due to number of threads, RAM, CPU speed etc), Batch size used etc.
With the nvidia-smi info,I could see that the GPU is under-utilized. Atleast memory-wise.
First step would be to increase the batch size and see if the speed is sufficient enough.