NSF/Intel Partnership on Computer Assisted Programming for Heterogeneous Architectures | Apply Online

FAQs: NSF/Intel Partnership on Computer Assisted Programming for Heterogeneous Architectures (CAPA)

What is the CAPA grant opportunity?

The NSF/Intel Partnership on Computer Assisted Programming for Heterogeneous Architectures (CAPA) is a National Science Foundation discretionary research grant opportunity focused on making it dramatically easier to develop efficient software for heterogeneous computing environments (systems that combine multiple kinds of processors and accelerators).

What kinds of computing systems is CAPA focused on?

CAPA is motivated by the shift from relatively uniform platforms to heterogeneous architectures. Examples described include systems that combine CPUs with GPUs, DSPs, programmable accelerators, and FPGAs, as well as increasingly diverse CPU designs that vary in core counts, memory hierarchies, communication mechanisms, and vector instruction sets.

Does CAPA address data center computing as well as single machines?

Yes. CAPA explicitly recognizes that data centers add additional heterogeneity and scale, where applications may run across thousands of distributed compute elements, with non-uniform networking, complex storage tiers, and diverse underlying hardware.

What problem is CAPA trying to solve?

CAPA targets the growing mismatch between who writes software and what it takes to make software run well on heterogeneous systems. Software development is widespread across many fields, but achieving high performance and energy efficiency on mixed hardware often requires specialized expertise in architecture-specific programming models and tuning. This gap can cause hardware capabilities to go underused and can limit performance and energy savings.

What is the overarching aim of the program?

The overarching aim is to fund groundbreaking university research that produces a significant and measurable leap in software development productivity by partially or fully automating development tasks that are currently done manually.

What are the three core research objectives emphasized by CAPA?

CAPA emphasizes three core research objectives:

• Improving programmer effectiveness: reducing the burden and expertise required to write high-quality code.
• Enabling performance portability: allowing software to run efficiently across different hardware targets without extensive rewrites.
• Supporting performance predictability: enabling developers to anticipate and reason about performance outcomes on heterogeneous systems instead of relying on trial-and-error tuning.

What does CAPA mean by improving programmer effectiveness?

Within CAPA, improving programmer effectiveness refers to reducing how much specialized, hardware-specific expertise is needed to produce high-quality and efficient software for heterogeneous systems, including reducing manual effort currently required for optimization and tuning.

What does CAPA mean by performance portability?

Performance portability refers to running software efficiently across different hardware targets without having to do extensive rewrites for each specific architecture or device combination.

What does CAPA mean by performance predictability?

Performance predictability refers to making performance outcomes on heterogeneous systems more understandable and anticipatable, reducing dependence on trial-and-error approaches to performance tuning.

What types of research directions does CAPA invite?

CAPA invites proposals across several interrelated technical directions described in the opportunity, including:

• New programming abstractions and methodologies that separate performance-related decisions from a program's functional design.
• Program synthesis and machine learning-driven approaches for automatic software construction, with explicit emphasis on demonstrable correctness (not only heuristic generation).
• Advanced hardware-based cost models and abstractions to support multi-target code generation and improve predictability for specified heterogeneous architectures.
• Integration of research outcomes with principled and usable software engineering practices, rather than isolated prototypes.

What are "programming abstractions and methodologies" in CAPA's context?

In this solicitation, these are approaches that allow developers to express intent and correctness while tools and compilation strategies manage hardware-specific optimization details. A key idea is separating performance-related decisions from the functional design of the program.

How does CAPA describe the role of program synthesis and machine learning?

CAPA includes program synthesis and machine learning-driven approaches aimed at automatic software construction, and it specifically emphasizes demonstrable correctness rather than relying only on heuristic code generation.

What are hardware-based cost models supposed to accomplish under CAPA?

The opportunity calls for advanced hardware-based cost models and abstractions that can bridge high-level code and low-level architectural realities, support multi-target code generation, and make performance behavior more predictable for specified heterogeneous architectures.

Why does CAPA emphasize integration with software engineering practices?

CAPA stresses that research outcomes should connect to principled and usable software engineering practices rather than remaining isolated prototypes, to increase the likelihood that results can influence real development workflows.

What type of funding mechanism is CAPA?

CAPA is described as a National Science Foundation discretionary grant opportunity.

What is the Funding Opportunity Number (FON) for this solicitation?

The Funding Opportunity Number listed is 16-606.

What activity category is associated with this opportunity?

The opportunity is under NSF's research and development activity category.

What CFDA number is associated with CAPA?

The opportunity is listed under CFDA 47.070.

What is the maximum award amount (award ceiling)?

The listed award ceiling is $3,000,000.

How many awards were expected?

The opportunity indicated an expectation of making about 3 awards.

What was the original closing date for proposals?

The original closing date listed was December 15, 2016.

When was the opportunity created?

The opportunity was created on September 19, 2016.

Who was eligible to apply?

Eligibility was described broadly as "Others", with clarification referenced in the full eligibility text. Applicants were expected to consult the solicitation details for precise eligibility rules.

Is this opportunity specifically intended for universities?

The program's stated aim is to fund groundbreaking university research focused on automating software development tasks and improving productivity for heterogeneous programming.

What does CAPA mean by "automation" of development tasks?

CAPA frames automation as partially or fully automating development tasks that are currently done manually, in order to achieve a significant and measurable leap in software development productivity.

What motivates CAPA's focus on heterogeneous architectures?

The opportunity cites the rapid shift from relatively uniform platforms to heterogeneous architectures, along with the increasing diversity in CPU designs and the added heterogeneity and scale found in data centers. It also notes that underutilization of hardware capabilities can limit performance and energy savings needed for emerging computing demands.