Access to a compute cluster without a university affiliation?

[crashcat]

TL;DR

I want access to a compute cluster same as most universities offer, where you schedule jobs through slurm etc., but I am not at a university this summer or fall.

I am temporarily an orphaned researcher with no university affiliation. Is there any public access compute cluster I can run some DFT on? I know that, for instance, Brookhaven Lab allows free access to its state of the art micro/nanofab facilities if you write and get your proposal accepted. Is there something similar from a university or national lab for computing resources?

 

[jedishrfu]

You could contact the UT Austin TACC center to see if you could get any time. However, I suspect there would need to be some money involved and/or a federal research grant. maybe you could get your old dept to sponsor you with or without pay.

https://www.tacc.utexas.edu/research-development/users

https://portal.tacc.utexas.edu/acco...eateaccountportlet_action=accept_usage_policy

And I found this XSEDE site:

https://www.xsede.org/

You couldn't run your code on a personal computer?

 

[crashcat]

Thanks I'll check them out. I have been running it on my personal computer and got like half the results I need, but I don't have enough RAM for the number of atoms in some of the simulations. The high number of atoms is to deal with it being amorphous material. I was looking and maybe I can run a version that only takes 50 GB, but right now it's like 80 GB.

 

[Vanadium 50]

XSEDE is an uphill battle if you aren't already NSF funded.

The obvious answer is AWS, Google Cloud, Azure, etc. If you don't want to pay for them, recognize that nobody does - that's why there is all the competition/restrictions on university clusters.

 

[crashcat]

I never thought about using AWS. Their pricing is actually reasonable. For a 16-core Xeon, 128 GiB RAM and the discount for letting them run the job whenever there is free capacity, it's 0.62 USD per hour. So I'm estimating $200 a month for a couple of months to run the code I need, or could be much less if their Xeon runs a lot faster than my OC i5 from 2016...

Plus it's scalable for the size of the job I'm running. I'm going to look more into this over the weekend.

 

[jedishrfu]

Of course you could do what these brothers did perhaps with a gaggle of raspberry pi boards:

https://www.newyorker.com/magazine/1992/03/02/the-mountains-of-pi

https://www.zdnet.com/article/build-your-own-supercomputer-out-of-raspberry-pi-boards/

 

[crashcat]

The Pi's are so slow, I think it would be cheaper and faster to run the CUDA version of Quantum Espresso, which I haven't tried but some guy vouched for it.

 

[pbuk]

The Pi's are so slow, I think it would be cheaper and faster to run the CUDA version of Quantum Espresso, which I haven't tried but some guy vouched for it.

Yes, you should be running this on a GPU for efficiency, cloud GPU is also a thing e.g. FluidStack.