Skip to main content
Premium Trial:

Request an Annual Quote

DRC Computer Claims 9.4 TCUPS Implementation of Smith-Waterman

Premium

This week, DRC Computer announced that an implementation of Smith-Waterman on its massively parallel DRC processor achieved 9.4 trillion cell updates per second, which it claims is the first implementation of the sequence-alignment algorithm to achieve several trillion cell updates per second

The company ran Smith-Waterman using the Affine gap model on the latest version of its Accelium coprocessors running Microsoft Windows HPC Server 2009 R2. DRC's coprocessors each incorporate a Xilinx field-programmable gate array and software elements that allow the module to plug in to an Opteron socket in a multi-processor system.

DRC said it achieved the 9.4 TCUPS benchmark running 200 base-pair reads against a 650 million nucleotide database on clustered, standard servers incorporating "multiple" DRC processors operating as a cloud computing environment.

Furthermore, the company said that a single DRC Accelium processor implemented the algorithm at 530 billion CUPS. The firm used the SSEARCH35 tool within the FASTA genomics tool kit to benchmark the performance.

Specific details about benchmarks in both cases were not provided.

Other companies are looking at various ways to speed the sequence-analysis workhorse. Last October, Pervasive Software said that its implementation of Smith-Waterman — based on its Pervasive DataRush Platform — analyzed 10 million combinations of protein sequences in 81.1 seconds on an SGI Altix UV 1000 with 384 cores with a sustained throughput of 986 GCUPS, which, it claimed at the time, surpassed other implementations by 43 percent. (BI 10/01/2010)

Convey Computer, meantime, has said that its HC-1 system, which integrates multi-core x86 processors with FPGAs, has run a modified version of Smith-Waterman at 64 GCUPS.

DRC said that running its processors on standard Intel-based servers can reduce the time and cost of analyzing gene sequences by a factor of 20 as well as reduce computing cost, power, real estate, and infrastructure costs by more than 90 percent.