We're very excited to announce that we have started collaborating with a team from Intel (yep, that Intel) to optimize key parts of the GATK code to make it run faster. The first fruits of this collaboration --a set of hardware-based optimizations for the PairHMM algorithm in HaplotypeCaller-- are available as of today in version 3.1 of the GATK. Please see the release notes and version highlights in the Version History section of the Guide for details.
Of course this is only the beginning, and we're looking forward to delivering more performance improvements for various other GATK tools moving forward as part of this collaboration.
What's really cool is that this collaboration extends beyond our little GATK team; the Intel Bio Team is also going to be working with other groups at the Broad Institute to make their software run faster as well, all with the goal of accelerating scientific research and discovery.
For more details and background information, see the Bio-IT World story here: http://www.bio-itworld.com/2014/3/20/broad-intel-announce-speed-improvements-gatk-powered-by-intel-optimizations.html
If it is compatible I plan to use a Xeon Phi host processor to run Red Hat and GATK.
Mainly I see it's floating-point performance with AVX-512 (6 TFLOPS single-precision/ 3 TFLOPS double-precision) extremely advantageous over regular Xeons.