A hybrid-core computing architecture integrating computer architecture and compiler technology with commercial, off-the-shelf hardware – an Intel Xeon processor and Xilinx FPGAs –' is providing a speed boost for life sciences research.
Convey Computer Corp. (Richardson, Texas) says it has gained a 172x faster implementation of the Smith-Waterman algorithm, widely used in life sciences applications for aligning DNA and protein sequences, compared to conventional methods and represents the fastest Smith-Waterman implementation to date.
The company's own benchmarks show its hybrid-core computer can process 688 billion cell updates per second as compared to four GCUPS for SSEARCH in FASTA on an Intel Nehalem core using the SIMD SSE2 instruction set.
Convey focuses on increasing the computer processing productivity required by life sciences applications where analyzing 340 terabytes of data is the norm.
The systems help customers reduce energy costs associated with high-performance computing, while increasing performance over industry standard servers. Additionally, Convey systems are provide full support of an ANSI standard C, C++ and FORTRAN development environment to ease programming.
Dr. Jason D. Bakos, assistant professor at the University of South Carolina with the Department of Computer Science and Engineering, said the mission of his Heterogeneous and Reconfigurable Computing Group is to "use reconfigurable coprocessor technology to accelerate applications that have never been accelerated before."
Such applications range from computational phylogenetics and sparse linear algebra to data mining and logic minimization. Additionally, the research group focuses on uncovering new design methodologies for high-performance computing ranging from developing new automatic partitioning tools to improving system architecture with multi-FPGA interconnects.
Meanwhile Dr. Harold "Skip" Garner, executive director of Virginia Bioinformatics Institute (VBI) and a professor in the Department of Biological Sciences at Virginia Tech, explains that VBI will use Convey's hybrid-core platform for its data analysis work for the 1000 Genomes project (an international effort to sequence the genomes of approximately 2,500 people from about 20 populations around the world).
The team has looked at 340 terabytes of data so far and much more is anticipated. The Convey systems will also be used to support text data mining as well as decision and policy informatics work at the Institute.
VBI is also working with Virginia Tech to develop a new high-performance computing hub for the university.
Dr. Garner said "We are intrigued by the concept of putting together traditional processors and FPGAs such that it would reduce the 'threshold of pain' for using these very high-throughput, very efficient processors. Also, there is the promise of having certain codes that will run on this architecture unlike anything else and anywhere else."
Convey Computer Corp. was established in December of 2006 and its investors include Braemar Energy Ventures, CenterPoint Ventures, Intel Capital, InterWest Partners, Rho Ventures, and Xilinx.
Its hybrid-core computing model supports multiple instruction sets in a single address space. Within this common space, the off-the-shelf x86 processors execute 'normal' (x86_64) instructions, while the coprocessor executes the application-specific instructions.
It uses reloadable personalities, application-specific instruction sets implemented in hardware that run on the Convey FPGA coprocessor . These are bundles of functions and functionalities that are not found on your garden-variety x86 processor. Personalities are flexible. And they are blank slates. They can range from providing instruction-level acceleration – for example, emulating a standard vector processing programming model – to highly complex algorithms for a wide variety of applications.
Convey has a number of personalities already available and others in the works. For example, one of Convey personalities implements a vector processing instruction set similar to those found on vector supercomputers. The Convey compiler detects opportunities for vectorization within a source file and generates vector instructions that will execute on the coprocessor. This personality substantially accelerates the processing of long pieces of code with nested loops.
Another Convey personality has been designed specifically for the financial industry. The FAP (financial analytics personality) significantly enhances key financial algorithms such as random number generation and math intrinsics specific to financial analytics.
A personality has also been developed to assist researchers in the field of proteomics – the study of an organism’s complete complement of proteins, one of the most important field of research in the life sciences.