GPU V100 server Provider: DataCrunch.io

Part 1 (2019)
1

Hi!

I’m Ruben from DataCrunch.io. DataCrunch provides low-cost dedicated V100 servers.
We are currently in private beta which means that anyone who wants to try, gets access at a 50% price reduction ($0,475/h for a V100 server).

We intend to have an image for Fastai later which requires 0 setup, currently you can choose to start from a Ubuntu 18.04 image.
In our docs it is explained how you can get Fastai and Jupyter notebooks up and running in a matter of 10 minutes.

Our machines are tailored for AI workloads, in fact the inspiration to start DataCrunch.io came from following the Fast.ai course! The GPU’s in our servers are all linked via up to 100GB/s NVLink, resulting in about 95% efficiency when training Resnet50 on 8 GPU’s.

Please get in touch if you are interested to try!

1 Like
3

The price for Fast.ai members has been set to $0.45/h for a V100 instance with dedicated hardware (ie. you do not share your resources with others).
We’ve also provided a Fastai image so there’s no setup required to get started!

Here you’ll find how to fire up a DataCrunch.io V100 instance:
https://course.fast.ai/start_datacrunch.html

Make sure to use coupon code FastAI30% to receive 30% discount.

2 Likes
4

This link doesn’t work any more, it seems.

5

Thanks for notifying, the url was changed to https://course19.fast.ai/start_datacrunch.html.
The url above should work again when the current course is updated.

1 Like
6

Hi

I’m just getting started. I started the fastai image on a single GPU $0.65 instance, then I tried saying:

git clone https://github.com/fastai/course-v4, from https://course.fast.ai/start_datacrunch

and it gave me a nearly empty repository asking me to clone fastbook instead. Which I did.

Promising looking collection of notebooks, but the first cell has a runtime error. Something about
something_attrs being a string. Hmm…

Any tips on how to get started. Would it be better to start from plain Ubuntu?

I also don’t know what to to with run.sh. Again, any tips.

7

Hi Chris,

It seems our image needs an update.
In the meantime, you can get it to work by executing these commands in the Jupyter terminal:

conda install -c fastai -c pytorch -c anaconda fastai gh anaconda
conda install -c fastai nbdev

Hope this helps!

1 Like
8

Yes. That works. Thankyou. The solve for the first command is a bit slow, but that’s fine for now.

2 Likes
9

The images are now updated, the fix above is no longer needed.