I’ve been working on a project which simplifies launching powerful GPU instances in the cloud. It is still in very early stages, so any feedback would be greatly appreciated
Main features:
Pre-installed deep learning libraries - Keras, Tensorflow, OpenCV, etc.
Root SSH access for each instance
Data is persisted across all your instances through the /shared-data directory
Jupyter Notebook URL for each launched instance
Multiple powerful GPU options
Small instance option (ideal for uploading / downloading data)
Are you familiar with Crestle? I like that you have some hugely powerful instances, and root SSH sounds much better than what Crestle offers (a Jupyter notebook/terminal which is pretty limited). The pricing comparison is quite interesting:
So CPU you’re better, data you’re better, but GPU you’re charging a lot more - and for anything serious that’s going to dominate the cost. And Crestle billing is per second, yours is a minimum of an hour (then second? minute? only hour increments?) - that’s a big difference in my experience. Would/could you consider reducing that minimum?
The Crestle free usage tier has dropped dramatically to an hour CPU/GPU use and 1GB data, it used to be more generous when I signed up - do you have any free tier at all?
I’m probably going to sign up to play because root SSH might be worth the extra cost, but 3x as much on the k80 GPU definitely makes this one a tough sell.
I actually found out about Crestle for the first time today. I saw the GPU price difference and agree that Crestle has me beat there.
You are charged a minimum of 1 hour, then it is per-second. Sorry for not being clear here. I will update the site.
I do plan on reducing the cost. I’m currently limited by funding, so I cannot purchase instances in advance. This significantly limits how low I can go. As soon as I verify that there is a need for this, I will focus heavily on price reduction.
Unfortunately, I currently do not. I figured that having a $0.01 per-hour CPU instance will let people explore the product and that would suffice. I will keep this in mind and work on providing a free-tier option.