Skip to main content

Call for Proposals for Vista GPU Resources

The Center for Generative AI is inviting proposals for allocation of computation resources in the areas of vision, language, bio, and health. The compute resources are for AI workloads, and will be allocated on the new Vista supercomputer consisting of NVIDIA GH200 nodes.

We will allocate resources at multiple tiers. Tier 1a will focus on standard projects, and the allocations will be limited to 5000 SUs (1 SU = 1 GPU-hour on the GH200 cluster). Allocations at the Tier 1a level will expire 3 months after the award. Tier 1b will focus on exploratory projects that can continue for a longer duration of six months. Tier 1b is limited to 1000 SUs.

Tier 2 allocations (greater than 5000 SU requests) should focus on large projects where there is a track record of promising prior work. These are expected to be large SU requests for projects that have been validated at a smaller scale, and these allocations will expire within 3 months of the award.

Approved grants which have SUs left over at the end of the allocation period will have the possibility to apply for consideration for a 3-month time extension, with a reason why the SUs were under-utilized.

Submission Deadline: We will accept submissions in a rolling manner, and periodically (roughly monthly) evaluate new submissions. Quicker reviews may be possible on a case-by-case basis for urgent projects. We will start accepting submissions on September 16, 2024. We expect to make the first batch of allocation by mid October.

Proposal Areas: The proposals will focus on one of these four primary areas: vision, language, bio, and health.

Proposal Format

Tier 1 (Standard / Exploratory) Proposal Format (2 pages total):

  1. Project title
  2. Area of research (vision, language, bio, health)
  3. Investigator team and TACC usernames
  4. Specific aims, significance, approach, related work, and timeline
  5. Models, datasets description, benchmarks, and validation
  6. Anticipated outcomes
  7. Indicate if Tier 1a or 1b, and the SUs requested (within the applicable caps)

Tier 2 (Large) Proposal Format (3 pages total):

  1. Project title
  2. Area of research (vision, language, bio, health)
  3. Investigator team and TACC usernames
  4. Specific aims, significance, approach, related work, and timeline
  5. Models, datasets description, benchmarks, and validation
  6. Anticipated outcomes
  7. Description of prior smaller scale proof of concept studies
  8. SU allocation budget

a. In either of the Tiers above, our goal is to allocate SUs to projects that can immediately start using Vista resources, so please provide details on the availability of models and datasets for downloading to Vista. We recommend providing specifics here, e.g., dataset is available at [link].

b. In your proposal, please indicate if the project has a significant CPU (as opposed to GPU) component.

c. We will limit each PI to have at most two active projects at any one time.

Submission: Please submit a single pdf (proposal document + short CV of PI) through this link: https://bit.ly/cgai-proposal

Selection Process: The proposals will be evaluated by the scientific board of the Center for Generative AI based on technical merit, feasibility (immediate availability of models and datasets), and the potential impact of the project.