DARPA describes its Ubiquitous High Performance Computing (UHPC) – 70 pages
Yes, it would exaflop supercomputers on the supercomputer end but it would also mean superfast and robust future smartphones. It will mean selfware operating systems and more hack resistent computers and networks.
The UHPC program will develop the architectures and technologies that will provide the framework and underpinnings for the resolution of the power consumption, cyber resiliency, and productivity problems. The UHPC program will develop computer systems, from the embedded to cabinet system levels, that have extremely high-energy efficiency. These systems will have dramatically reduced power consumption while delivering a thousand fold increases in capabilities.
• Efficiency: New system-wide (hardware and software) technology approaches to minimize energy dissipation per operation and maximize energy efficiency, without sacrificing scalability to ultra-high performance DoD applications.
• Programmability: Develop new scalable system architectures and technologies that do not require application programmers to explicitly manage system complexity, in terms of architectural attributes with respect to data locality and concurrency, to achieve performance, time to solution and other goals.
• Dependability: Develop a system-wide approach to achieve dependability through fault management techniques enabling an application to execute correctly through both failures and attacks, and to protect the confidentiality and integrity of information, while achieving the user’s goals. These goals could include performance, time to solution, energy efficiency or power consumption. Dependability across all levels of a UHPC System Design that is nearly transparent to the user and hardware performance.
* To realize this vision requires reinventing how computers process and manage data and how applications are developed and executed. UHPC System Designs that merely pursue evolutionary development will not be considered.
To reach the energy goals alone will require reducing energy per operation from thousands of pJ/Op (representative of current processors), to tens of pJ/Op. As an example, achieving 50 GFLOPS/W is equivalent to expending only 20 pJ per floating point operation – a budget that must encompass far more than just the floating point operation: leakage losses, operand accesses, and operand transport, along with instruction issue and concurrency control. This will require energy-optimized solutions from the basic functional elements through subsystems and systems
The goals of the UHPC program are as follows:
• Single cabinet system that achieves the UHPC program vision and goals. The hardware system goals areas described in the document.
• Self-aware OS and the resulting new system software stack. • Prototype compiler for the new programming model, that enables ease of programming for an ExtremeScale system.
• Dynamic system that adapts to achieve optimal application execution goals, without the direct involvement of the application developer.
• UHPC System Design that supports modern high performance for scientific and engineering applications.
• UHPC System Design based on a multi-level model of dependability.
• Processor module that is capable of being used within terascale embedded and multiple cabinet systems.
• Development efforts must utilize open innovation and software and hardware co-design throughout the life of the program.
The UHPC software effort spans operating systems; runtime systems for scheduling and lower level resource management; memory management; communication; performance monitoring; power management; self-aware operation; and prototype compilers. It is anticipated that a new system software stack will be developed for a UHPC System. The UHPC program will not provide funding for research and development of large-scale parallel file systems, high bandwidth I/O and storage technologies. However, it is a requirement that a UHPC System include these components.
UHPC Challenge Problems
There are five UHPC Challenge Problems:
• Massive streaming sensor data problem resulting in actionable knowledge,
• Large dynamic graph-based informatics problem,
• Decision class problem that encompasses search, hypothesis testing, and planning,
• Two challenge problems drawn from DoD applications and to be selected after UHPC program initiation.
Initial primary metrics
Metrics that will be used through the life of the program are:
• Energy efficiency: 50 GFLOPS/W for the HPL benchmark
• System Performance: 1 PFLOPS for the HPL benchmark
• Programmability: TBD
• Dependability: TBD
• Cabinet Power Requirement
The two phases covered under this solicitation will have period of performances and funding splits as follows;
Phase 1 will be 24 months:
• Each TA1 team receiving up to 3.25 million US dollars the first year and up to 5.25 million dollars the second year;
• One TA2 team receiving up to 1.75 million US dollars each year;
Phase 2 will be 24 months long:
• Each TA1 team receiving up to 8.65 million US dollars per 12 month period;
• One TA2 team receiving up to 2 million US dollars per 12 month period;
If you liked this article, please give it a quick review on Reddit, or StumbleUpon. Thanks