Help for doing fastai courses with Cuda 5.0 (xps 15 9550)


(Javier) #1

Good day friends,
I apologize if this is not the right subforum to post this but I was not sure where to put it,
I’m enthusiastic to do the fast.ai courses,
my issue is that I want to do them with my Dell XPS 15 9550 laptop, which has a GPU GeForce GTX 960M which has CUDA 5.0 , when I did the windows installation I got the message that my GPU was too old, then I read somewhere that pytorch may bring back support for 5.0 later on,
I tried to compile pytorch from source but didn’t work for me, got different errors during the process,
do you have any suggestions, tips etc so that I can do the fast.ai courses with my Cuda 5.0 GeForce GTX 960M on a xps 15 9550? thank you soooooooooo much
:wink:


(Javier) #2

I now found
https://colab.research.google.com/notebooks/welcome.ipynb
and Im running lesson 1 there now,
is google collab research a good alternative to do the fast.ai courses if I cannot use my xps 15 9550?
thank you


(Allan Jackson) #3

I think my system uses a GTX 940MX so even if they go back to supporting drivers for out machines, we will get the “out of GPU memory” errors as the next roadblock to using Fastai on our laptops. Disappointing, but that is how life is sometimes, you jump one hurdle to find another further on. The choice to use CoLab will work for “some time” as you progress as an early Fastai student. Later on you will come up against the compute time limits (I think it’s 12 hours now?) and possible memory limits as you apply Fastai and other sw on larger and larger data sets. Eventually in your education you may move on to the larger cloud platforms like GCP and AWS. These both provide slower/lower capacity free levels and introductory periods and user credits.
With some research you will find that GCP will offer you more free credits if you watch their Google Cloud webcasts (as well as free coursera time and credits and Qwiklab course time and credits). I must warn you that GCP and AWS have learning curves of their own. If you consider each of these as a learning gradient (in effect learning GCP, Python, Pytorch, FASTAI-v3, etc.) not to mention that they are ALL actively evolving behind the scenes) you will realize what a herculean challenge lies ahead of you in the dark.
Fortunately there are others feeling there way through this dimly lit maze with you.
We are your “Fastai V3 Fellows” (FVF’s ?) and will wipe the blood off of our respective foreheads and help each other through this as best we can.
I would first try and join a slack channel like “Kaggle Noobs” or something similar - try to avoid info sharing where “trolling” is prevalent, Your brain will be tenderized enough without negative comments.
Try searching on the forum for slack channels and see what pops up!
Your FVF!
Allan