CUDA Center of Excellence (CCOE) Program

 

Preamble

The CUDA Center of Excellence (CCOE) program by NVIDIA recognizes, rewards, and fosters collaboration with universities at the forefront of massively parallel manycore computing research. Schools identified as CUDA Centers of Excellence have proposed a unique vision for improving the technology and application of parallel computing. Through the CUDA Centers of Excellence NVIDIA intends to:

  • Empower academic researchers to do world-changing research, by dramatically increasing the computing power available to scientists and engineers - on the desktop, in the laboratory, and in the data centre.
  • Improve the state of parallel computing education, to train the next generation of computer scientists and computational scientists for a world of massively parallel computing.
  • Establish research, educational, and recruiting relationships with the foremost academic institutions in the world.

The CUDA Center of Excellence program is competitive and prestigious. Institutions meeting the requirements below are invited to submit proposals to NVIDIA at any time; proposals will be evaluated on a rolling basis.

A full description of the program follows. For quick answers to common questions, see the CCOE Frequently Asked Questions.

Eligibility

Any academic institution whose primary goals include both research and education may apply for CUDA Center of Excellence status. This includes both US and international universities. Non-university research institutions with strong educational missions will be considered on a case-by-case basis.

Merit criteria

Proposals to form a CUDA Center of Excellence will be evaluated on:

  • The vision presented by the proposal for improving the technology and application of massively parallel computing.
  • The quality, quantity, and diversity of ongoing research at the institution.
  • The commitment to, and depth of, educational plans for teaching CUDA and massively parallel computing.
  • The likelihood of broad impact in NVIDIA's view, including importance of the application areas, likelihood of success, plans for deployment of research results, and the caliber of researchers and institutions involved.

NVIDIA explicitly invites proposals to apply CUDA and GPU computing (e.g. to various scientific and engineering problems), or to improve CUDA and GPU computing (e.g. with higher-level libraries or languages, compiler research, etc.), or both.

Requirements

A university proposing a CUDA Center of Excellence will describe a vision for massively parallel computing research built around NVIDIA CUDA technology. NVIDIA welcomes research visions that advance the state of both application domain research such as computational science & engineering, finance, etc., and "core" computer science and engineering topics such as parallel architecture, programming languages and compilers, numerics, etc.

Universities applying for CCOE status must already satisfy two minimal requirements:

  • Demonstrate promising research results involving CUDA in at least 3 separate research laboratories or projects.
  • Teach one or more courses on a recurring basis, ideally accessible to domain computational scientists as well as computer scientists that introduce students to CUDA and provide them the background to try CUDA in their own projects.

Other possible elements

In addition to the above requirements, a CUDA Center of Excellence may involve many other specific elements. Universities proposing a CCOE are encouraged to think creatively about various ways that the institution and NVIDIA can work together. For example, a CCOE proposal may request that NVIDIA:

  • Provide educational materials for use in teaching CUDA.
  • Donate equipment to support courses teaching CUDA.
  • Donate equipment to support a small research cluster
  • Partner with other hardware vendors, for example to provide host machines for teaching labs and research clusters.
  • Provide CUDA training, either generic CUDA tutorials or advanced domain-specific workshops.
  • Provide access to prerelease hardware and software to enable research on next-generation technology
  • Designate specific NVIDIA researchers and engineers to serve as a "point of contact" for the CUDA Center of Excellence institution.
  • Provide discounted hardware for large-scale purchases and in-kind cost matching purposes.
  • Give cash in the form of a grant, research contract, or unrestricted gift.

While the CCOE University may:

  • Make CUDA courses available online.
  • Develop and release domain-specific libraries and packages.
  • Host domain-specific workshops or conferences relevant to CUDA (such as 'AstroGPU', see [1]).
  • Host a CUDA cluster for use within the university and available, within reasonable restrictions, to a specialized external research community.
  • Cooperate in joint publicity efforts (such as press releases, launch events, etc.) showcasing work at the university.

The above items are just intended as examples and are neither required nor necessarily representative.

Duration

Each CUDA Center of Excellence has an intended duration of up to three years, subject to NVIDIA receipt and approval of an annual report. Renewal each year is at the sole discretion of NVIDIA. Centers are renewable beyond the three-year duration upon invitation by NVIDIA; if invited, a proposal to renew a CCOE will be reviewed competitively and upon its own merit. The annual report should highlight publications, results, courses, and projects supported by the CCOE. The contents of the report will be used in joint publicity efforts by NVIDIA and the CCOE institution.

Application Process

Universities proposing a CUDA Center of Excellence should submit a short proposal to CCOEproposal@nvidia.com . Proposals should be in Adobe PDF form and should consist of no more than 5 pages motivation and description including budget, up to 5 pages supplementary material detailing previous or envisioned research including references, and a biographical sketch or condensed curriculum vita, which must be no more than 2 pages per principal investigator.

CCOE Frequently Asked Questions

Contents

Intent

Is the CUDA Center of Excellence program intended to motivate institutions to incorporate CUDA into their research and teaching, or ***to recognize*** schools that already do so?
A: The CCOE program recognizes schools pushing the frontier of massively parallel computing using CUDA. A CCOE also establishes a formal relationship and fosters future collaboration between the institution and NVIDIA. Schools that do not yet meet the requirements for a CCOE are welcome to apply to the NVIDIA Professor Partnership Program . We also encourage Ph.D. students working on relevant topics to consider applying for the NVIDIA Fellowship Program .

Does the CUDA Center of Excellence come with a monetary grant?
A: A CCOE is not itself a large monetary grant; it is a recognition of excellence and a vehicle for closer collaboration, including the possibility of future financial support.

Does the CUDA Center of Excellence carry a public relations component?
A: Yes, CCOEs provide positive publicity both to NVIDIA and our partner institutions. NVIDIA will work with the university on joint publicity efforts (such as press releases, launch events, etc.) and on public relations surrounding the establishment and activities of the CCOE.

Eligibility

What institutions are eligible?
A: The CCOE program targets academic institutions whose primary goals include both research and education. Candidate institutions need not be located in the United States.

I am on the faculty of a teaching college (e.g. a four-year liberal arts college); is my institution eligible for a CUDA Center of Excellence?
A: No, schools whose primary mission does not include graduate-level research would normally be considered ineligible for a CCOE. Exceptions for unique circumstances will be considered on a case-by-case basis.

I am a researcher at a non-university research organization with a strong educational mission, or affiliated with an academic institution; is my institution eligible for a CUDA Center of Excellence?
A: While the CCOE program targets academic universities, exceptions will be considered for such research organizations on a case-by-case basis.

My University is large and diverse. Can an institution have more than one CUDA Center of Excellence?
A: No, only one CCOE would normally be considered per institution. If a CCOE already exists at your institution we will be happy to connect you to those researchers.

Teaching requirement

We don't yet teach a course in CUDA but would be happy to once we get a CUDA Center of Excellence award, is this okay?
A: No, the teaching requirement is a precondition for a CCOE. Candidate institutions must have already taught CUDA in some fashion, and must demonstrate a commitment to continue teaching CUDA. NVIDIA views teaching as a vital component to fueling and expanding continued research, since courses such as these are how students will learn about and gain experience with CUDA to take back to their own disciplines and research programs.

I teach a class on parallel programming that uses CUDA but not exclusively, since we also cover other types of parallelism (e.g. MPI, OpenMP). Does this meet the teaching requirement?
A: Potentially! We are looking for evidence that students who take this class will be able to try CUDA for their own research projects.

I teach a class on computer graphics but the students learn about CUDA and use it in their projects. Does this meet the teaching requirement?
A: Possibly, if you can demonstrate that the course is accessible to and taken by a broad audience (i.e., not just advanced computer graphics grad students), ideally including students from disciplines other than Computer Science.

Research requirement

We don't yet have multiple research projects using CUDA, but have some good ideas for research we will be able to do once we get a CUDA Center of Excellence award. Is this okay?
A: No, the research requirement is a precondition for a CCOE. We encourage schools just getting CUDA research efforts underway to consider a proposal to the NVIDIA Professor Partnership Program.

Applying for a CUDA Center of Excellence

How long should a proposal be?
A: To reduce the proposal preparation and reviewing effort we prefer to receive short, direct, informal proposals. As an upper limit, a CCOE proposal should consist of no more than 5 pages motivation and description including budget, up to 5 pages supplementary material detailing previous or envisioned research (including references), and NSF-style 2-page biographic sketches of the principal investigators.
We invite institutions considering applying for CCOE status to discuss the possibility informally first at CCOEProposal@nvidia.com.

When are proposals due and how long is the review process?
A: Proposals are accepted on a rolling basis and reviewed roughly quarterly.

General

What CUDA Centers of Excellence have been established to date?
A: As of this writing, NVIDIA has recognized the following schools as CUDA Centers of Excellence:

  • The University of Illinois at Urbana-Champaign (PI: Wen-Mei Hwu)
  • The University of Utah (PI: Chris Johnson and Chuck Hansen)
  • University of Cambridge (PI: Graham Pullen)
  • National Taiwan University (PI: Ji-Wang Chern)
  • Harvard University (PI: Hanspeter Pfister)

How many CUDA Centers of Excellence does NVIDIA intend to establish?
A: The CCOE designation is competitive and prestigious, and represents a significant commitment on the part of NVIDIA to collaborate with and support the designated institution. The bar is thus quite high. However, NVIDIA intends to recognize and reward true excellence wherever it emerges, and has set no fixed limit on the number of CCOEs.

What is the duration of a CUDA Center of Excellence award?
A: Each CCOE has an intended duration of up to three years, subject to NVIDIA receipt and approval of an annual report. Renewal each year is at the sole discretion of NVIDIA. Thus the actual CCOE awards are made for one year, conditionally renewable each year for up to three years total.