Running in Local GPUs

Hi Everyone, and thanks Jermey for the the lovely course…

Thought of sharing the following with you:
I spent sometime searching on how to run the course in Local GPU (Quadro M1000M), and found the following article:
1- How to install fastai v1 on Windows 10

Things did not work, so I tested the following:

2- How to check your pytorch / keras is using the GPU?

I showed I don’t have GPU:

3- I replaced the MS windows driver with NVIDIA manually (automatic update did not work)

4- I repeated (2), the GPU showed up

5- Tried Resnet34 exercise and failed complaining about Memory, so I reduced batch size (bs=8), it took sometime but worked…

Feeling the need for speed, will try the cloud platform…

For the second day, I’m still running in the local machine and working on Fruits 360 datasets.
I searched a bit to see if I can do the task a bit faster in my laptop and feel good about the results if I use ResNet18. So I googled the difference, and while I did not expect impressive accuracy about ResNet18, result was not bad at all in this dataset:

However, with only 2GB of memory in my workstation, this is not sustainable :slight_smile:
Waiting for the weekend to try the cloud option, but most likely going for SageMaker…

1 Like