Performance
In the early years of HPC, the raw power of technology made our hearts race. Breakthrough supercomputing performance opened up mind-boggling research and discovery opportunities.
Today's ordinary technology offers only incremental performance increases. We don't want ordinary; we want extraordinary. We want the thrill of adrenaline rushing, order of magnitude, anything's possible performance increases.
It's here. And it's Convey.
Adrenaline rushing performance increases
With the installation of HC-1 servers at the University of California, San Diego (UCSD), the engineering team is achieving substantial performance gains, cost savings, and a reduction in the environmental impact. On a widely used bioinformatics application, UCSD is reducing the time to completion by two orders of magnitude.
“Current methods are slow and generally computationally prohibitive, taking up too much time and resources. Even the fastest blind search tools could run for months while analyzing millions of spectra against a large protein database. With the HC-1 we are predicting the same search could be accomplished in less than a day. This will benefit processes sent to the web server at the Mass Spectrometry Center from around the world.”
-Pavel Pevzner, director of the Center for Computational Mass Spectrometry in UCSD's Jacobs School of Engineering
Order of magnitude performance increases
The College of Computing at Georgia Tech (Atlanta, Georgia) is using the HC-1ex to accelerate data mining and analysis computations.
“As part of our involvement with the DARPA Computer Science Study Panel (CSSP), we are studying how to use heterogeneous computing systems to speed up statistical data analysis and mining applications. These computations can be slow because of hard-to-predict patterns of data access, for which the Convey memory subsystem has shown significant advantages over comparable platforms. The capability of the Convey HC-1ex to create customized operations using the FPGAs is also extremely attractive, since it will allow us to study opportunities available to extend conventional instruction sets for this class of computations.”
-Rich Vuduc, assistant professor in the School of Computational Science and Engineering (CSE) within the College of Computing at Georgia Tech
Anything's possible performance increases
Convey’s HC-1ex achieves more than two times the performance of the Convey HC-1 running the Smith-Waterman algorithm, a widely used life sciences application. The HC-1 performance of this algorithm was already the fastest implementation to date—172 times faster than conventional methods.1 For the HC-1ex, a highly-optimized version of the Smith-Waterman algorithm is 401 times faster than what is typically achieved on an x86 processor.2
Convey Computer Corporation: The leader in hybrid-core computing
1 According to Convey's internal benchmarking, the Smith-Waterman algorithm runs 172 times faster on the HC-1 than a software implementation on an Intel Nehalem core.
2 According to Convey's internal benchmarking, the Smith-Waterman algorithm runs 401 times faster on the HC-1ex than a software implementation on an Intel Nehalem core.
