Intel strives to “standardize the base.”
June 10, 2008 | While Intel’s digital health group is developing a new business for the company—creating products and services for more effective home-based care within the health care industry—other groups within the microchip titan are looking to apply Intel’s general computing expertise to optimize performance in industries with growing high performance computing (HPC) needs.
Wilfred Pinfold is general manager of Intel’s Integrated Analytic Solutions group. Originally from Liverpool, Pinfold trained in computational fluid dynamics, moving to United States in 1980 (and losing most of his scouse accent in the process.) Pinfold spoke at a BioTeam-organized pre-con workshop Bio-IT World Conference & Expo on Intel’s interest in life sciences. Kevin Davies asked him to discuss Intel’s growing interest in digital health, bioinformatics, and next-generation sequencing.
Bio-IT World: What is attracting Intel to the bioinformatics/life sciences fields?
Pinfold: There is a significant interest in accelerators and accelerated platform solutions. We’ve released a technology called QuickAssist, which allows you to plug accelerators into the platform directly, and we’re looking at “many-core” architectures—highly accelerated platforms, we’re already shipping quad cores, but we expect the core-count to continue to climb. These systems are extremely valuable for analytic workloads—things like bioinformatics, seismic work, engineering design work, workloads that have been considered HPC workloads…
What we find as we look into these analytic workloads is that the traditional ecosystems with independent software vendors (ISV) are difficult to form. If a user builds their own HPC system, these ISVs can not deliver an executable that will run consistently on every user’s platform. To deal with this, customers exchange source code, compile it, and bring it up on their own system. To do this well requires a high level of expertise and a considerable amount of time.
Understanding that not everyone has the level of expertise or time available, we’re doing the things that are necessary to make a turnkey accelerated platform available. To make this work, we then realized we needed to do something to help the ecosystem develop—for there to become a base of executable codes... We’re trying to do what is necessary to get that ecosystem into place, such that if you want to buy a bioinformatics system, you can buy a highly accelerated platform that will plug into your sequencer and produce good sequence results in a reasonable time frame, without having to learn all about clusters and accelerators.
How did your interest in the next-generation sequencing arena begin?
Our interest in supporting accelerated workloads was high, and we realized, “Oh my god, this is a great place to be right now. This is crying out for a solution.” Whereas the top institutes like Broad and Sanger can build their own solutions with large clusters, for the next tier of users, the people who will buy 1-2 systems to do serious biology, they don’t want to have to buy a machine that’s more costly than the sequencer itself and more complex to run. They want a solution.
So the bioinformatics space really stood out, particularly the task of dealing with sequence data off the high-throughput sequencers and how that was going to change the computational solution. And having the realization that it was becoming fragmented—people were looking at FPGA solutions, GPU solutions, a variety of solutions. If the market fragments in that way, it will make it impossible for software developers and ultimately stall the ability for high-throughput sequencers to be shipped in the quantities warranted.
Is this primarily a hardware or software initiative?
The solution has to combine hardware, software, and services. We want to work with existing providers in the ecosystem to deliver this solution. For example, BioTeam is an excellent bioinformatics services provider and we don’t want to become the content experts in bioinformatics! There are plenty of people at Broad and Sanger who develop excellent alignment and assembly algorithms, and there are commercial entities providing other parts of the solution. We want to work with open-source providers, like Sanger, so whatever solution we put out there will be suitable for running things like MAC and VELVET. We also want to work with [companies like] ABI, Roche, Helicos, and Illumina to make sure their codes like ELAND will run effectively on this solution. We want to provide a solution by working with and influencing, not fragmenting the market. We are talking with and working with many of these ecosystem providers to understand how best to do this. Our intent is to try to do what Intel has done so well in the PC field: to try to get people to standardize the base, to get to a point where the customer has the best of both worlds—they can get the stable platform they need, then on top of that, all the software and services will work. They don’t have to worry about whether their HPC system will have 80 or 120-nodes in the cluster, Gigabit Ethernet or Infiniband, RedHat or Debian operating system, etc. We’ll try and stabilize that.
Vendors ship sequencing instruments with huge clusters. Is this sustainable?
As sequencers become affordable for clinical applications, it will become increasingly important to offer a turnkey bioinformatics solution. It is this customer base we are targeting. As new sequencing technologies like those from Illumina and Helicos produce ever increasing amounts of raw data, there will be pressure to explore exotic computational solutions like FPGAs and GPUs. This will further fragment the market and disrupt the ecosystem. We believe an accelerated solution can help avoid this.
Would your device work with all the next-gen platforms?
We know that customers would like to use multiple sequencers to achieve their goal. For example, they might use 454 for the framework and fill in with short-read data. So the idea is this system could take data from multiple sources and integrate it effectively.
We’re also interested in dealing with the image processing task. There are opportunities not yet taken to improve image quality in that cycle, because it’s very hard. e.g. If you could do work to improve the resolution of the star-field image before you do base calling, then that’s a very interesting workload—it benefits greatly from acceleration on special purpose silicon or an FPGA. So there are significant opportunities if we think about the workload throughout that transition. Clearly that requires working closely with all the instrument vendors and software providers. Our intent is not to bluster in and provide the solution, but to try to work with the community to make it attractive for all the players to come to a solution that then has some commonality, and allows an ISV community to develop services communities, then we get to stabilize the base and become a good provider to our customers.
What do you think about the pace of progress in this field?
The really exciting thing about this field is that it is moving so rapidly. If you can’t get your machine out and generate revenue in the next 2-3 years, then you probably need to be working on your next generation machine! We at Intel are very familiar with rapid improvements—we’ve followed Moore’s Law for many years. It’s forced us to improve all our manufacturing and design techniques very rapidly. The majority of our products are replaced by new products every 12 months.
This article appeared in Bio-IT World Magazine.
Subscriptions are free for qualifying individuals. Apply Today.