Cocalc for student project

[Krysalide]

hello,
I hope i post at the right place.
I begin a project as a student, i need at least 800MB of storage for my dataset and access to GPU for deep learning training. I will use it 2 hours per day during 3 months. I am ready to spend 15 dollars per month. Can you give me some advice? Is it a good idea to take a compute server? What can i get for 15 dollars? Best Regards. Christophe D

[williamstein]

What can i get for 15 dollars?

That's $0.50/day. Since you want to use a compute server with a GPU for 2 hours, that's $0.25/hour. You can get a spot instance T4 on Google Cloud via CoCalc for about $0.16/hour

That leaves $0.18/day (50-2*16 = 18). You can get a cocalc license with 2 GB RAM and 3 GB disk for $5.43/month (or about $0.18/day):

https://cocalc.com/store/site-license?user=academic&period=monthly&run_limit=1&cpu=1&ram=2&disk=3&uptime=short&member=true

I would recommend buying both a license and then using a T4 spot instance compute server (pay as you go), in order to fit within your budget.

NOTE: Compute servers are the only way to use GPU's on CoCalc.

-- William

[Krysalide]

Thanks for your answer!! I have another question: Is it possible to do some remote ssh from vscode and run the code on cocalc server?

[williamstein]

You can easily run vscode directly on the compute server, by just clicking the "VS Code" in the "..." dropdown.

[Krysalide]

yes i am going to test that possibility first, although i really prefer to use local IDE.

[Krysalide]

i finally took a license and will use a gpu when needed. I want to say vs code in the browser is really nice. Thanks!!!

[Krysalide]

I took a compute server T4 Colab at 0.23 $. plus 0.03$ per hour when off. I am doing my purchase from Europe France Do you advise me to close server and recreate one when needed?

[williamstein]

Do you advise me to close server and recreate one when needed?

It depends on your use case. If you deprovision it, there is no cost at all, but you may have to reload files, and of course anything on any local scratch disk will be gone. But that may be fine for your use case.

You can also change the disk type to something less expensive to reduce that $0.03. The price is high partly because colab is large.