What is the point of Sagemaker Studio Lab?

I was actually excited about getting my account approved and wanted to try out a few things in the Sagemaker Studio Lab account. I was able to get a fastai learner going using the CPU instance, but try as I might, I simply cannot select a GPU. It just says “runtime is not available, try again later” .

If this is a competitor to google colab, but users can’t get a GPU instance for days (I’ve been trying for 4 days straight now), then what is the point of having a CPU based notebook server?

I was actually planning to do this year’s course using Sagemaker Studio Lab but after this, I think I’ll stick to either my local setup or go back to Paperspace or equivalent.

I understand I can fire up my own AWS GPU instance and hook it up as a GPU backend to my studio lab setup, but then it’s really not a Google Colab replacement (from a product perspective). I’m really having trouble understanding where Amazon is going with this offering. My experience for one, has been underwhelming.

1 Like

The GPU availability has been woeful, if they could rectify that it would be awesome. The CPU 12 hour sessions are pretty good though if you have bulk inference work to do. (generally don’t need GPU for inference). Having it maintain your environment, notebooks, data etc between sessions is an advantage over Colab. But yeah, the lack of GPU when you want one is a killjoy.

1 Like

This may be of interest to you …

3 Likes

Yes I saw that, and things like persistent storage, inference potential etc were the attractive points for me, but for a colab ‘competitor’ it needs to have some kind of GPU offering. I’ve never had an instance with Google Colab where it said to me that the GPU wasn’t available. I guess it’s getting too popular and Amazon can make more money selling GPU than offering it to freeloaders like me but still … :slight_smile:

At the very least the GPU unavailability message could point me to the ‘pro’ / paid version (if any) lol.

2 Likes

Out of curiosity, which region are you running in?

I don’t think I can select a region in SMSL unless of course I choose to go with my own GPU enabled AWS instance. Or maybe I didn’t look hard enough?

1 Like

I misread. I was looking at Sagemaker Studio. Hey new feature for me to learn about.

As you run into errors, can you post screenshots in this thread? It will allow me to escalate your tickets and concerns on the AWS side.

@mindtrinket are you working at AWS? I can confirm the experience described by @mike.moloch of never being able to find any GPU capacity. As a result at this stage we won’t be able to recommend it for the course.

6 Likes

hi @mindtrinket , when I click on “Start runtime” with GPU selected it tells me no runtime available.

2 Likes

I do. Let me see what I can do. Same Error?

2 Likes

Yup same.

1 Like

I updated this post a couple days ago mentioning that Studio Lab GPUs are generally unavailable.

1 Like

I’m working on getting this issue in front of someone who can do something about it at AWS

4 Likes