CPU/GPU configuration for training

Hello Fellow users,

I am currently on Lesson 8 and soon will start with Part 2 of the course. While training models (as part of further research or even the course notebook), I noticed that per epoch time (in course video) for Jeremy is under 12 min (Ex: Language model in notebook 10), while it takes me ~1hr per epoch. I am using Paperspace with 8 CPU and 16GB RTX5000 GPU. What configuration should I use to achieve similar speed?

Thank you in advance.

Depends on what type of GPUs Jeremy was using at the time of the course. He currently uses a 15000 Lambda machine which is probably got multiple GPUs. His 2020 machine was probably multiple GPUs with large amounts of RAM.

According to this page rtx5000 and 2080super are equivalent performance wise. If you’re looking at 2020 videos, then Jeremy was probably using a 2080ti or maybe a pair of 3080s? though I’m not sure. He did mention during lectures that he had access to some beefy machines through the University at San Fran.