How to Request an NSF ACCESS Compute Allocation

Getting time on a national supercomputer through NSF ACCESS — the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support program that replaced XSEDE — does not require holding an NSF award or clearing a subaward process. It requires submitting an allocation request through the ACCESS Allocations system (allocations.access-ci.org) that is scaled to the project’s size and reviewed against criteria that get more demanding as the request gets larger. This guide walks through that request process step by step: which tier to request, what the request needs to justify, how review works, and what happens after an award is made. For the broader question of how ACCESS as a program differs from a campus HPC recharge center, see CASRAI’s companion guide, NSF ACCESS vs. campus recharge models — this page assumes that context and focuses specifically on the mechanics of asking for an allocation.

Pick the right request tier first

ACCESS structures every request into one of four project types, distinguished by resource ceiling and review burden rather than by discipline. Requesting the wrong tier for a project’s actual scale is the most common reason a request stalls or comes back with questions, so this decision comes before anything else:

• Explore — for resource evaluation, benchmarking, code development, small classes, and early graduate-student work. Capped at 400,000 ACCESS Credits, lasts for the supporting grant’s duration or 12 months (whichever is longer), and needs only a one-page proposal. Multiple Explore requests are allowed, and they can be submitted at any time.
• Discover — for research grants with modest computing needs, Campus Champions, larger classes, and gateway development. Capped at 1,500,000 Credits, same 12-month-or-grant-length duration, and needs a proposal of up to three pages.
• Accelerate — for mid-scale, experienced-user projects: multi-grant consolidation, collaborative work, and work explicitly building toward a future Maximize request. Capped at 3,000,000 Credits and needs a proposal of up to ten pages, reviewed by a merit panel rather than checked only for eligibility.
• Maximize — for large-scale research activity that genuinely exceeds Accelerate’s ceiling. Awarded in resource units rather than a flat Credit cap, runs for 12 months, and goes through a full merit-review proposal reviewed by the ACCESS Allocation Review Committee (AARC) on a fixed twice-yearly cycle rather than the rolling, submit-anytime pattern of the smaller tiers.

A project team new to ACCESS should generally start at Explore or Discover even if the eventual computational plan is larger — an Accelerate or Maximize reviewer will want to see that the team has already demonstrated it can use a smaller allocation efficiently before awarding a much larger one.

What the request actually needs to justify

Every tier asks the same two underlying questions — what will this allocation be used for, and can this team use it responsibly — but the depth of justification required scales sharply with tier:

• Proposed use of resources. At Explore/Discover, this is a brief project abstract: the project’s goals, how ACCESS resources specifically fit into the work, and any software the team already knows it will need. At Accelerate and especially Maximize, the same question becomes a full computational plan — the specific codes and workflows to be run, the resource types requested (CPU, GPU, large-memory, storage), a credible estimate of how much of each is actually needed, and evidence the team has scoped that estimate rather than guessed at it.
• Prior computational experience. Every request includes a CV or resume (capped at three pages) for the project lead and any co-leads, so reviewers can see the team’s track record with the compute scale being requested. At the merit-reviewed tiers, this expectation sharpens into supporting evidence inside the proposal itself — prior results, benchmarking data, or a smaller ACCESS allocation already used productively — that the team can execute at the requested scale, not just describe it.
• Funding context. Supporting-grant information is requested where it exists, partly because it helps establish project duration and partly because reviewers use it to gauge scope — but, as covered in the companion comparison guide, an active grant is never a precondition for requesting or holding an ACCESS allocation.
• Role-specific documentation. A signed advisor letter (or NSF GRFP correspondence) is required when the project lead is a graduate student, since undergraduates cannot lead a request and graduate-student leads need a faculty backstop. A course syllabus is required for a classroom-use request. All supporting documents are uploaded as PDFs through the request form.

Where and how you submit

Requests are submitted directly through the ACCESS Allocations site, not through a separate NSF portal or your institution’s sponsored-programs office. The mechanics are the same across tiers: create or log into an ACCESS account, select the project type that matches the scale decision above, and complete that tier’s request form with the abstract or proposal, team CVs, and any role-specific attachments described above. There is no annual fixed deadline for Explore, Discover, or Accelerate — requests are accepted on a rolling basis and reviewed continuously. Maximize is the exception: it is reviewed on a fixed twice-yearly cycle by the AARC, so timing a Maximize submission around that cycle matters in a way it does not for the smaller tiers.

Review timelines and what happens after award

Review speed tracks tier size closely. Explore requests are typically decided by the next business day, with an email confirmation on submission — the check at this level is essentially eligibility and resource fit, not competitive merit review. Discover requests go through the same lightweight eligibility-and-fit check on a similarly fast, rolling basis. Accelerate moves to genuine panel merit review, still submitted anytime but taking longer to turn around than the eligibility-only tiers. Maximize requests are reviewed by the AARC against criteria consistent with NSF’s standard intellectual-merit and resource-efficiency review emphases, on its twice-yearly schedule.

An award of ACCESS Credits is not itself compute time — the team still needs to exchange awarded Credits for a specific allocation at a participating Resource Provider system (the university- and national-lab-operated centers that actually host the hardware), and each Resource Provider approves that exchange based on fit between the workload and its systems. Budget roughly a week for that exchange step to clear before treating the allocation as usable. Once an allocation is active, day-to-day use looks like using any other Slurm-scheduled cluster — see CASRAI’s guides on writing an sbatch job script, requesting a GPU in a Slurm batch job, and getting an interactive session on a compute node for the mechanics of actually running work once the allocation itself is in hand. Watching how a request’s estimated need compares to actual usage also matters for renewal: reviewers on a follow-on request can and do look at whether a prior allocation’s Credits were used efficiently, which is part of why understanding fairshare and queue priority on the awarded system is worth doing early rather than only after jobs start waiting.

Renewing, extending, or moving up a tier

Explore and Discover allocations are designed to be requested repeatedly rather than renewed in the traditional sense — a team that has used its Credits productively simply submits a new request at the same tier, or moves up to Discover or Accelerate once the work has outgrown the smaller ceiling. Accelerate and Maximize projects are explicitly expected to plan ahead for that trajectory: ACCESS documents Accelerate itself as a track record-building step for teams that expect to eventually need a Maximize-scale allocation, so a team’s Accelerate proposal is read partly as evidence for whatever it requests next. Because Maximize runs on a fixed 12-month term reviewed twice a year, a project expecting to need continued Maximize-scale resources should plan its renewal request well before the current award’s term ends, rather than waiting for Credits to run out.

If a request is declined or reduced, the eligibility-only tiers (Explore, Discover) generally allow a prompt resubmission with a revised abstract, since there is no fixed deadline cycle to wait for. At the merit-reviewed tiers, treat reviewer feedback the way a declined NSF proposal is treated elsewhere in the research-administration world: a real substantive response addressing the reviewers’ specific resource-efficiency and methodology concerns, not a resubmission of the same proposal.

How this differs from requesting time on a campus HPC cluster

The request process above is unique to ACCESS’s national, merit-reviewed model. A campus core facility or HPC recharge center works on an entirely different logic — usage is typically requested through a local queue or account-request form rather than a competitive proposal, and cost is recovered through a billing rate charged against a grant or department account rather than awarded as a no-cost Credit. Neither model is a substitute for the other in every case: a lab might use a locally billed campus core facility for day-to-day development and reserve an ACCESS allocation for larger production runs that exceed local capacity. CASRAI’s full comparison of the two models covers the governance, cost-recovery, and grant-budgeting differences in detail; the funding side of who pays to build the local facility in the first place is covered separately in the guide to research infrastructure funding.

Frequently asked questions

Do I need an NSF grant to request an ACCESS allocation?

No. Eligibility is based on institutional affiliation and role — graduate student or higher, at an eligible U.S. organization — not on holding a specific award. A supporting grant can be listed where relevant, but it is never a prerequisite.

How long does an Explore or Discover request take to review?

Explore requests are typically decided by the next business day. Discover follows the same lightweight, rolling eligibility-and-fit review rather than a competitive panel process, so it is similarly fast compared to Accelerate or Maximize.

What’s the difference between an Explore proposal and a Maximize proposal?

Explore asks for a one-page abstract describing the project’s goals and intended use of resources. Maximize requires a comprehensive proposal reviewed by the ACCESS Allocation Review Committee against NSF-style intellectual-merit and resource-efficiency criteria — the same underlying questions, answered at far greater depth and with supporting evidence of prior efficient use.

Can I request more than one allocation at a time?

Yes, at least at the smaller tiers — Explore and Discover both explicitly permit multiple concurrent requests. Maximize is the exception: typically one active request per project is allowed, with only limited exceptions.

What happens immediately after my allocation is approved?

Awarded ACCESS Credits still need to be exchanged for a specific allocation at a participating Resource Provider system, and that Resource Provider approves the exchange based on fit with its own hardware. Plan for roughly a week for that step before counting the allocation as usable.

Can an undergraduate submit an ACCESS allocation request?

Not as project lead. Undergraduates can be added as authorized users under a faculty or otherwise-eligible PI’s allocation, but they cannot hold the lead role on the request itself.

Once an allocation is active, running production work efficiently is its own topic — see CASRAI’s guides to recognizing and running embarrassingly parallel jobs and moving data to and from an allocated system with Globus for the practical next steps.