Strong Compute Call for Research Proposals.

You/your team is solving for useful, trustworthy AI.
You need GPU.

We're offering:
* $10k-$100k grants to access.
* Up to 100x 24GB Ampere GPUs.
* Time on cluster is shared. 5-50%/day over 3 months.
* + dedicated single GPU machines.
~1+TB storage, per researcher.

Must publish a public access demonstration.
* Code + sample of data must be public, ideally full dataset.
* Goal to publish within 3 months.

Applications are open now.

Apply now, hear back within 2 business days.

How to win: be working on something interesting:
* How to deal with hallucination.
* Preventing private data leakage.
* Post transformer architectures.
* Alternative training optimizers/algorithms.
* Explainable AI.
* Something better than any of the above.

Applications are open now.

Submit your proposal in under 20 minutes. 5 questions and 1 one minute video.

Get Started
What is the application process?
Applicants submit proposals via the Google form. Proposals are reviewed and the outcome is decided within 1-2 business days after submission. Successful applicants are notified and on-boarded immediately.
When can I expect to hear back?
1-2 days.
Is this open to individuals or organisations?
Will successful applicants receive cash or GPU credits?
Credits for use with Strong Compute.
When does the three-month access start and finish?
3 months starts within 2 days of being accepted.
What does publishing a public access demonstration mean?
You must be aiming to release the research in an open format, for example code on github under BSD/MIT/Apache licence, research on arXiv or similar. Exceptions may be made for publications in major journals/conferences. Datasets used should be open if possible, or if not possible, an open sample set must be made available (e.g. compatible synthetic data that can be used to run and verify your code).
How is time-on-cluster managed?
Users launch training / computation jobs onto a shared cycling cluster. Jobs are run for a short period of time (minutes to hours) before being paused to allow the next job in the cycle to run. Time-on-cluster is configured per user group by adjusting length of cycle time. Strong Compute includes cluster state saving helpers for this.You can always interrupt running jobs to test something quickly - instant super computer feedback.
What technologies will my team need to be familiar with?
PyTorch. (we use torch run to run distributed jobs).
You must integrate Strong Compute cycling utils to use the cluster (we’ll help). -  cycling_utils Strong Compute has published the ISC Demos GitHub repository to get you started.
Is support available?
Personal onboarding.
2 calls per week.