Intel Xeon Phi co-processor simplifies supercomputing

Intel announces 1 TeraFLOP performance from a single PCIe card

Intel today unveiled the new name and production schedule details for their Xeon family Many Integrated Cores Architecture (MIC) co-processor that was previously shown under the codename "Knight's Corner" which is now known as the Xeon Phi. The new coprocessors are available in a PCIe form factor, taking advantage of the latest generation of Intel's Xeon family processors and motherboards direct support of PCIe 3.0.

Last year Intel showcased a single pre-production Phi unit that was able to deliver just over 1 TeraFLOP double precision floating point performance. They really like this number, using the graphic below to demonstrate the difference between 2012 and their first TeraFLOP capable computer, in 1997.

The single connection, two slot wide PCIe card holds the 50+ core Phi processor and a minimum of 8 GB of GDDR5 memory. The MIC coprocessor is produced using Intel's 22nm, 3-D tri-gate transistor technology. While the Phi was used in a new supercomputer that was ranked as the 149th fastest in the world today, production quantities of the co-processor won't be available until sometime later this year.

The Xeon processor is currently found in 70% of the top 500 supercomputers in the world and the first generation Xeon Phi is complementary to that, being optimized for highly parallel supercomputing tasks.  Highly parallel applications are in use in the business world in forms of weather prediction, oil exploration, pharma modeling, and a number of other surprising common applications. Future generations of the coprocessor will likely be targeted at other current Xeon markets; traditional datacenter and high-end workstation workloads.

At their announcement today at the International Supercomputing conference, Intel had more than 40 partners on board with commitments to make use of the Xeon Phi in their supercomputer roadmaps. And note that these are commercial partners, not the academic "build your own" one-offs that are some of the best known supercomputing efforts. High performance computing is still under the radar for most users, but the benefits that it brings, as it expands beyond the current application market, driven in part by products like the Xeon Phi, will begin to get pushed down to the more traditional server-based applications in the near future.