Mitrionics’ hybrid computing speeds genomics apps.
By Ryan DeBeasi
September 15, 2009 | We use CPUs for general computing, and GPUs are for more than just graphics, but Mike Calise, executive VP of Mitrionics, is hoping that next-generation sequencing companies will add a third type of chip to their arsenal: the Field Programmable Gate Array, or FPGA.
An FPGA is a processor in which the logic gates can be rearranged to create a chip that’s specialized for a particular application. Mitrionics sells FPGA-based servers along with software that makes them easier to program. Calise anticipates a $500 full genome sequence within the next five years, but in order to get there, sequencing companies will need to crunch masses more data for less money. This trend, he says, will drive adoption of his company’s hybrid computing platform.
Calise defines a “hybrid computer” as “a heterogeneous mix of processors in a single system with working software… That could be a cell processor from IBM; that could be a GPU from Nvidia or others; that could be an FPGA; or a combination of those all in the future.” Of crucial importance is the software. “A heterogeneous computer is simply a computer with multiple processor types, but a full hybrid computer is a heterogeneous computer with all the software worked out to get the maximum benefit for any algorithm that goes through it.”
For example, to do more efficient calculations in programs such as BLAST the FPGA can be configured as a 2-bit processor, rather than a 64-bit processor. Each base can be represented in two bits, so an FPGA configured this way would process bases individually rather than in groups of 32, improving accuracy, and if running calculations in parallel, speed.
Jag Boleria, a senior analyst at the Linley Group, explains the importance of bit width with a multiplication problem. If you’re multiplying two and two, he says, each of the twos can be represented in two bits: “10.” In a 32-bit processor, however, those numbers will be represented as “10” preceded by 30 zeroes. That’s a waste of 30 bits for each number. A two-bit processor, on the other hand, would only use two bits for each number. “Architecturally, that’s much more efficient for the functionality than a general-purpose processor would be,” says Boleria. According to a Mitrionics whitepaper, not all algorithms can be sped up by using the processors, and most FPGAs top out in the hundreds of megahertz, compared to 2 or 3 gigahertz in consumer-level CPUs.
Super Soft
Founded in 2001, Mitrionics differentiates itself by its software applications are written in Mitrion-C, not a chip design language, and are run in the company’s Mitrion Virtual Processor software, which runs on top of the physical FPGA. This “Mitrion” platform differs from what companies such as Celoxica offer, says Boleria: Celoxica provides C-programmable FPGA hardware, but it can only be used for a specific set of financial calculations.
By contrast, the Mitrion platform is general-purpose. “It’s a complete reconfigurable solution,” says Calise. “Hardware stays constant and the MVP [is] what changes. It’s all virtual changes so you can run a job on BLAST, you can run an SSearch, you can reprogram for HMMer, you can do all of this stuff flexibly with no hardware change and no being pigeonholed into a closed ‘black box’ system.”
Mitrionics has about 30 customers, some of which bought the platform from SGI, which sold servers that included Mitrionics software. In April, SGI filed for bankruptcy and was purchased by Rackable for $25 million. “Mitrionics is working with a number of premier FPGA accelerated systems suppliers,” says Calise. The company now sells Mitrion-based hybrid computing servers itself. “I’d say our sweet spot in deal size is between $20,000 and $200,000.”
Currently, fewer than half of the company’s customers are biotechnology companies. “The challenge we have,” says Calise, “is that the people within [genome] centers are trying to figure out how to do more genomes; they’re not necessarily the people who know how to take an accelerator and do something exciting to solve that problem with [it].” Potential customers in genomics centers will “wait until they have a fast-running application on an accelerated system. Then it’s like, ‘Great, whatever’s under the hood [is] OK by me, but we won’t tinker under the hood.’ Once we see that, there’ll be more tinkerers and it’ll be a self-fulfilling growth. But it’s not quite there yet.”
To attract more customers in the sequencing space, Mitrionics has ported the BLAST-N application to the Mitrion-C language and made the program open source. Also in the pipeline are a Smith-Waterman program, Hidden Markov Models code, and generic DNA kernels for next-gen sequencing as well. Calise says that in tests using the Mitrion platform on SGI-supplied servers, BLAST-N ran 60 times faster than it would have on a non-accelerated server. Calise also notes that FPGAs allow users to boost performance without increasing power usage or size.
Calise says, “[FPGA] technology did not come to be standardized and cost effective until just about over a year ago.” He expects units to become smaller and more energy-efficient, and predicts the number of logic gates in the processors will increase over time.
Mitrionics could face competition from pay-per-hour cloud computing services such as Amazon. While Calise sees the cloud as a competitor in the short term, he calls it a “phenomenal opportunity” in the long term.
“We actually don’t care where... that FPGA hybrid computer, exists,” he says. “Frankly, if it’s in the cloud, that makes it more pervasive and more interesting, and potentially people could charge different subscription rates for different results, so all of that is exciting for us. Should we be talking to Amazon? Absolutely. Should we be talking to the large pharmas about the problems they’ll see in pharmacogenomics? Absolutely. Do we care where that data center is? ... Nope. And if it’s in the cloud and that generates sales, even better.”
This article also appeared in the September-October 2009 issue of Bio-IT World Magazine.
Subscriptions are free for qualifying individuals. Apply today.