4 ms·
That was no problem, because back then you only really needed a GPU for training. People who only learned about DL through transformers and LLMs might find this
by sigmoid10 2y ago
That was no problem, because back then you only really needed a GPU for training. People who only learned about DL through transformers and LLMs might find this hard to believe, but you could comfortably run inference on AlexNet with 250mb RAM on a CPU and Inception only used like 50mb. No fancy hardware or accelerators necessary.