Compute for a paper deadline, without a procurement cycle.
Most research compute is spiky: nothing for three weeks, then everything at once before a submission. Paying by the second suits that pattern far better than a reserved cluster that idles most of the term.
The V100 and CUDA 11.8 images exist specifically so that older published results can be reproduced without fighting driver incompatibilities. If a paper you are building on was written against an older stack, pin to it rather than porting.
Verified academic accounts receive additional credit on each top-up. Write to us from an institutional address with a link to your group page and we will apply it to your account.