Hypercomputers

---

Hypercomputing: Turning Algorithms into Circuits

Until now, reconfigurable computing has been largely limited to theory because no satisfactory tools have existed to exploit the inherent parallelisms of the underlying programmable hardware, called Field Programmable Gate Arrays (FPGAs). While FPGAs are capable of executing hundreds of thousands of instructions per clock cycle, existing software development tools cannot capture inherent algorithmic parallelisms or complex multi-million gate designs.


Starbridge introduces unique patented and patent-pending technologies that bridge this gap, making the inherent power of reconfigurable computing hardware accessible to High Performance Computing for the first time. Simply put, Starbridge technology allows programmers to intuitively and graphically turn an algorithm into its own set of circuits, rather than forcing the algorithm into existing hardware structures. The capability to create highly efficient and massively parallel circuits on demand, for supercomputing applications, is a revolutionary concept with far reaching implications in many data and computation dependent industries.

We call this new approach Hypercomputing . Hypercomputing delivers the power to boost the performance of complex computational challenges by orders of magnitude, while reducing the maintenance, operating power and air conditioning requirements-all in the footprint of one 4U chassis.

Turning Months into Minutes for Elite Organizations

Hypercomputers provide the immense computing power needed for the most complex scientific, industrial, and military pursuits. Hypercomputers supply the computing power you need to explore new possibilities and solve new problems. And they dramatically compress the timetable for performing advanced analyses, computations, and comparisons-so you can start turning months into minutes.

Hypercomputing provides bioscientists more powerful computational tools to conduct advanced research. It provides geoscientists the potential to run precise seismic algorithms in days or weeks (that used to require up to six months). And it provides the defense industry with powerful new tools for research, munitions, surveillance, and encryption.

Some of the world's most advanced and reputable organizations—from NASA to the US Air Force —are already using Hypercomputers to expand discovery and solve complex new problems.

The Move from Supercomputing to Hypercomputing

Hypercomputing represents a fundamental shift in how computing power is harnessed and applied to advanced algorithms.

Instead of a typical microprocessor, Hypercomputers use FPGAs as the computational engine. Microprocessors are specifically designed and hardwired to perform general tasks, resulting in very poor efficiency-typically only 3 to 5% of the chip is used at any time. In contrast, an FPGA is a very flexible hardware resource-a kind of silicon whiteboard that can be configured, erased, and configured, time and again, creating the most optimal circuitry for the algorithm being run. The on-chip logic of an FPGA can be quickly re-wired, dynamically incorporating new functionality. The resulting efficiency of the FPGA is near 100% silicon utilization, with the entire chip performing critical data processing tasks.

The Starbridge Hypercomputer uses high-end Xilinx Virtex-II FPGAs, each containing more than six million gates. Various Hypercomputer configurations comprise between 7 and 22 Virtex-II FPGAs, providing from 36 million gates to 124 million gates, that can be employed simultaneously to implement massively parallel computational pipelines for powerful algorithm execution.

Most "Grand Challenge" problems are inherently parallel. Once these parallelisms are identified, code can be written (or ported from C, C++, Fortran, or other existing code) to capture these parallel operations, and turn formerly serial operations-or merely distributed operations-into truly parallel algorithms. In most cases, instead of porting an entire program to VIVA, porting only the computational kernels achieves the desired increase in speed.


The Hypercomputer's unique FPGA architecture also makes it possible to concentrate the power of hundreds of clustered microprocessors into a single 4U rack-mounted system and requires only a standard wall plug for power. This exceptionally small footprint supplies computing power sufficient to execute algorithms typically running on several hundred CPUs, while significantly reducing power consumption, air conditioning costs, and space requirements.

---

Reconfigurable Computing – The most widely used computational engine for large problems is cluster computing-an approach to supercomputing that aggregates hundreds or even thousands of individual CPUs. However, cluster technology is reaching a point of diminishing returns. Clusters are expensive to build and difficult to maintain; they require ever increasing power, cooling systems, and physical space; and their architecture places inherent limits on scalability and performance.

A new approach to high performance computing is called Reconfigurable Computing. Researchers and scientists have studied and identified the exciting potential of Reconfigurable Computing: orders of magnitude speed increase by executing algorithms directly in specialized hardware.

NASA – Scientists at NASA's Langley Research Center use a Starbridge Hypercomputer for projects that push traditional computational boundaries, including radiation analyses, digital signal processing, and atmospheric studies. Recently, the Marshall Space Flight Center purchased a Hypercomputer to perform structural analyses, rocket engine plume analyses, and other complex flight research projects. NASA scientists at both facilities are planning future uses for this remarkable technology, including using Hypercomputers as command centers for spacecraft and satellites.

United States Air Force – The Air Force is using Viva®, Starbridge's graphical, high-level FPGA development environment, to program embedded FPGA chips for munitions applications. This includes exploring the possibilities of using Viva to program Automatic Target Recognition systems and other kinds of advanced munitions systems.