Scalable Parallelism in the Extreme | 17 600

Scalable Parallelism in the Extreme (SPX) - FAQs

What is the NSF SPX grant opportunity?

Scalable Parallelism in the Extreme (SPX) is a National Science Foundation (NSF) grant opportunity focused on research to improve computing performance in a world where gains can no longer depend mainly on faster single processors. The program targets approaches that exploit parallelism at unprecedented scale, reflecting modern systems built from multicore and many-core processors, specialized accelerators, and large cloud or warehouse-scale platforms.

Why is NSF emphasizing "post-Moore's Law" computing in SPX?

SPX is framed around the reality that semiconductor scaling is hitting physical and economic limits and that single-processor performance improvements have largely plateaued. Because the traditional path of making one processor core faster is no longer delivering the same returns, the program highlights parallelism as the most promising path for continued performance growth.

What does "Scalable Parallelism in the Extreme" mean in practice?

It refers to parallel computing at such large scale that issues that look minor today can become major barriers. When workloads span huge numbers of cores, nodes, and services, small inefficiencies can multiply into serious performance losses. Even short serial portions of code can impose strict limits on overall speedup (reflecting classic limits like Amdahl's Law), and real-world factors like coordination overhead and communication costs can make those limits even more punishing.

How does SPX relate to high-performance computing (HPC)?

SPX is aligned with national priorities for future high-performance computing and is positioned as a response to the need for a long-term path forward for HPC in the post-Moore's Law era. The opportunity emphasizes extreme-scale parallel systems and the research needed to keep HPC performance improving even when hardware improvements alone are not enough.

Is this program only about hardware improvements?

No. SPX emphasizes coordinated advances across the computing stack rather than relying on any single layer. The solicitation calls for collaborative research spanning areas from services and applications down through runtime systems, operating systems, compilers, programming models, and even micro-architecture. The idea is to produce approaches that scale, remain programmable, and can be reused as platforms evolve.

What technical challenges is SPX trying to address?

Based on the opportunity description, SPX targets several classes of problems that become dominant at extreme parallel scale, including:

• Parallel efficiency at scale: small overheads and inefficiencies becoming catastrophic when multiplied across large systems.
• Limits from serial code sections: even brief serial work can significantly restrict overall speedups.
• Heterogeneity: effectively using mixed environments where CPUs, GPUs, and other accelerators work together.
• Complex memory hierarchies: performance sensitivity due to differing memory systems across components.
• System-level overheads: unacceptable latency and coordination costs as architectures and distributed systems become more complex.
• Reliability at scale: reduced mean time between failures as the number of components increases, even if individual components are reliable.
• Performance predictability in distributed environments: challenges from communication, synchronization, contention, and topology.

Why does SPX emphasize heterogeneity (CPUs, GPUs, accelerators)?

The opportunity notes that heterogeneity is becoming the norm. Systems increasingly combine CPUs, GPUs, and other accelerators that each have different performance characteristics and memory behaviors. Efficiently managing these mixed environments is a major challenge, especially when developers need predictable performance and portability instead of one-off tuning for each platform.

What does SPX mean by "coordinated advances across the stack"?

It means the program is looking for research that connects multiple layers of the computing system rather than treating them as isolated. The solicitation explicitly calls out collaboration that can span:

• Services and applications
• Runtime systems
• Operating systems
• Compilers
• Programming models
• Micro-architecture

The intent is to build solutions that improve performance at extreme parallel scale and that can remain useful as platforms change over time.

Does SPX encourage collaboration with industry?

Yes. The program encourages coordination with industry efforts pursuing similar goals. This reflects an emphasis on real-world impact and technology transition in addition to foundational research.

How is SPX aligned with the National Strategic Computing Initiative (NSCI)?

SPX is described as aligned with NSCI and positioned to support the full set of NSCI strategic objectives. The opportunity ties the research to national priorities for future computing, including:

• Foundational research in architectures and software that can drive performance gains beyond traditional semiconductor scaling
• Development and deployment of programmable, scalable, reusable platforms within national HPC and scientific cyberinfrastructure
• Stronger integration between data analytics and modeling and simulation workflows
• Progress toward capable extreme-scale computing

What kinds of outcomes does SPX appear to value?

From the description provided, SPX emphasizes approaches that are practical at extreme scale, scalable, and still programmable. It also highlights reuse as platforms evolve, suggesting an interest in ideas that generalize beyond a single machine or one-off optimization and that can carry forward as architectures change.

What is the funding agency and funding instrument for SPX?

The funding agency is the National Science Foundation (NSF), and the funding instrument is a grant.

What is the activity category for this opportunity?

The activity category is listed as "science and technology and other R and D."

What is the CFDA number for SPX?

The CFDA number provided is 47.070.

What is the opportunity number for SPX?

The opportunity number listed is 17-600.

How many awards were anticipated?

The listing anticipated around 25 awards.

What is the award ceiling for SPX?

The award ceiling field is listed as 0. In many listings, this typically indicates that the ceiling is not specified in that particular field, rather than implying that no funding is available.

What was the original closing date shown for this opportunity?

The original closing date shown was January 9, 2018.

Who is eligible to apply?

Eligibility is listed as "Others," with a note that clarification is referenced in an additional eligibility section. The provided information does not include the text of that additional eligibility section, so specific eligible organization types are not detailed here.

Is SPX focused on any specific computing environments?

The description highlights modern systems dominated by multicore and many-core processors, specialized accelerators, and massive cloud and warehouse-scale platforms. It also emphasizes highly interconnected distributed environments where communication, synchronization, contention, and topology can strongly influence performance.

Why does reliability become a bigger issue at extreme scale?

The opportunity notes that as systems incorporate more components, the overall system can fail more often simply because there are so many parts involved. Even if individual components are highly reliable, the mean time between failures for the entire system can decrease at extreme scale.

What performance issues become more important at extreme scale?

SPX highlights issues such as latency, coordination overhead, and difficulty predicting performance in distributed settings. Factors like communication patterns, synchronization points, contention for shared resources, and system topology can strongly shape performance outcomes at scale.

Is SPX aimed at isolated research ideas or integrated approaches?

The solicitation language emphasizes that the intent is not only to produce isolated ideas, but to build approaches that scale, remain programmable, and can be reused as platforms evolve. It also explicitly calls for collaborative work spanning multiple research areas.