Biomatters is partnering with ActiveMotif's TimeLogic to allow users of its Geneious Server software to run bioinformatics jobs on Time Logic's accelerated DeCypher system, a field programmable gate array-based technology.
The companies plan to make their partnership public during next week's Plant and Animal Genome Conference in San Diego, Calif.
The partners will create a module that will allow Geneious Server users to run compute-intensive jobs on the TimeLogic accelerated hardware instead of a standard server, Peter Meintjes, Biomatters' market development manager, told BioInform.
"The end deliverable is tool that provides multiple options for running algorithms on datasets, depending on the acceleration required," Meintjes explained.
For example, the system will let a user run Blast, Smith-Waterman, or HMMer on a laptop, "or if it is a larger job, you can just pick from a drop-down list and say 'Run on the Geneious Server' or 'Run on the Decypher card,'" he said.
The partnership "is about extending the functionality of Geneious Server to people who are interested in really high-end acceleration in addition to expanding the existing customer base of TimeLogic to researchers who want to use the hardware but don’t have the necessary expertise to do so," he said.
TimeLogic’s DeCypher system combines PCIe cards embedded with FPGA chips, a host server, and implementations of Blast, Smith-Waterman, hidden Markov model algorithms, and gene-prediction algorithms. According to the company, a single DeCypher card can provide performance equivalent to several hundred CPUs.
When the integration is complete, the planned DeCypher module will let users offload large sequence-comparison jobs onto the cards in order to process data in a fraction of the time of standard compute clusters.
Michael Murray, sales and marketing manager for TimeLogic's products, told BioInform that a single card from the latest version of DeCypher provides compute power that is equivalent to 280 CPU cores. Furthermore, he said, adding a second card to an existing server results in a 97 percent to 98 percent increase in performance.
TimeLogic currently provides two host servers for its DeCypher systems: a Dell R610, which can accommodate three FPGA cards; and a DellR900, which holds up to seven cards. Pricing for a system, which includes a card, software, and server, starts at $25,000.
In internal benchmarks, one DeCypher card running TimeLogic's implementation of Smith-Waterman, DeCypherSW, compared 500 human protein sequences to SwissProt 379 times faster than a single CPU core running Ssearch. Two and three cards compared the same sequences 748 times and 1,099 times faster, respectively.
In another benchmark, one card running DeCypherSW compared 1,000 Arabidopsis EST sequences to the reference genome 236 times faster than a single core using Ssearch. Two and three FPGA cards performed the same task 444 times and 631 times faster, respectively.
Meintjes said that the partners plan to "provide some functionality" for customers by the second half of 2011.
An Increasingly Competitive Market
Murray noted that while historically TimeLogic’s competition has come from standard compute clusters, the rise of alternative options like cloud computing and graphical processing units could in the long run pose a significant threat to his company’s offerings.
However, he said that cloud computing isn’t a major threat at present because most cloud users don’t have “a chronic demand” for analysis services and only access bioinformatics resources intermittently, whereas TimeLogic’s system is targeted at core facilities with researchers “that are doing genome annotation 24/7.”
For these groups, the cost of accessing and utilizing cloud-based resources “appreciates significantly” and concerns about data transfer and security make it a less appealing option, Murray said. He did concede, however, that as these matters are addressed, “the cloud is going to be more of an issue.”
Meanwhile, GPUs are a more immediate threat because they offer good performance at a much lower cost than FPGAs, Murray said. Furthermore, it’s much quicker to develop applications for GPUs while FPGA applications take time and require significant programming expertise.
Although it is possible to use a generic compiler to write C code compiled for an FPGA, the algorithms must be "tweaked" to match the specific card design in order to optimize their performance, Murray explained. “I think going forward … what companies like ours will be forced to do to compete is to continue to come up with faster FPGA card designs that will continue to have that performance advantage,” he said. “And maybe most importantly, [we will have] to develop a new algorithm [and] new tools for our hardware.”
Aside from growing threats from the cloud and GPU markets, TimeLogic also faces competition from other hardware companies that offer FPGA-based technology to the life sciences market. For example, Convey Computer also provides FPGA-based systems that some bioinformatics groups are adopting.
Last May, the Virginia Bioinformatics Institute purchased a computing system from Convey that it planned to use to analyze data generated by the 1000 Genomes Project in addition to other projects (BI 05/28/2010).
The Richardson, Texas-based company has also sold its "hybrid-core" systems to Lawrence Berkeley National Laboratory, Stanford University's Center of Computational Earth and Environmental Science, and Oak Ridge National Laboratory.
Still, Murray said, Active Motif's TimeLogic is holding its own and "enjoyed a good year" in 2010. Last year, the company sold two DeCypher systems priced at more than $1 million and two other systems priced at more than $500,000. Although he could not mention specific customers, Murray did tell BioInform that the systems were purchased by two academic laboratories and two government agencies in North America and Asia.
Currently, he said, the company is looking to form additional partnerships with other software providers similar to the one it has formed with Biomatters.
TimeLogic is also working to develop a new algorithm that will be implemented on its hardware as well as to create an updated version of an existing application that’s similar to the publicly available HMMer algorithm, Murray said. The firm is also working on the next incarnation of its FPGA card design, which it expects to release in the next 12 to 18 months.
ActiveMotif currently has 45 employees about 10 of whom — mostly software developers and engineers — are part of the TimeLogic group.
Have topics you'd like to see covered in BioInform? Contact the editor at uthomas [at] genomeweb [.] com.