hi all, i have 46 cpu cores and 96gb ram. it seems like fastai run my model on single cpu which result in slow training. how do i make use of all my resources for faster training.
Why are you training on CPU cores and not on GPUs?
i only have CPUs on my machine
Hey @mphakaj, CPUs aren’t great for training deep learning models. GPUs allow far more simultaneous operations and are much more efficient. If you really are restricted to using only CPUs, there are some cases where it might make sense for you to run multiple processes (e.g. if you’re doing a hyperparameter search).
One area where multiple CPUs can help is in data loading/preprocessing. E.g. when you initialise a DataLoader, you can specify num_workers=46. (info here: Title | fastai)