Intel has this week released the first production samples of its brand new supercomputer on a card, the Xeon Phi 5110P. Destined for the HPC (High Performance Computing) market the Phi cards will help Intel push into a market it has struggled with in the past.

HPC is the world of big computers, super computers so large they are given names by the people that build them, often quite catchy names such as ‘Sequioa’, ‘Stampede’ and who could forget ‘Big Blue’, IBM’s Jeopardy playing supercomputer.

The Phi release has been on the cards for a while now. After being announced in June they are now in the hands of a select few with the cards becoming available in early 2013. The Xeon Phi 5110P will cost $2699USD if you are contemplating building your own bedroom supercomputer.

Intel is again attempting to assail the supercomputer ivory tower, a market it has never been able to dominate as they do on the PC desktop. Previous attempts such as the Itanium server processor have cost Intel billions and have had very little impact in the world of big tin, servers and supercomputers.

The Phi 5100P is this new weapon, a silicon design that takes multiple cores to a whole new level. The new head of the Phi family (95110P) can contain up to 62 individual cores, each a 64bit x86 processor that is roughly equivalent in design to the original Pentium processor. The chip logic has been upgraded since the original Pentium days of course, with new hyper-threading and 64bit updates amongst the changes.

5 billion transistors are used to produce each chip that runs at 1.1Ghz, cranking out 1TFlop (trillion floating point operations per second) of processing power.  Built using Intel’s latest 22nm 3D Tri-Gate technology as seen in the Ivy Bridge generation of CPU’s, the manufacturing processor does give Intel’s new Phi a power advantage over the current kings of HPC; IBM, AMD, and NVIDIA.

While it may seem a little unusual to try to leverage the old x86 architecture into the world of supercomputers, where 640k has never been enough, it does make development easy. With many people growing up coding x86 there is a wealth of talent already available. AMD and NVIDIA are both struggling with the task of getting the world used to yet another programming model and language, required to develop for their supercomputer co-processors.

Unlike most other Intel CPU’s the Phi’s chips don’t plug into a socket on a motherboard, instead like a graphics processor the Phi are built on to a PCIe card with memory and interfaces included. Technically described as a co-processor the Phi is placed in a system for a sole reason, to crunch through massive calculations.

The HPC market is not an easy direction for Intel to take, dominated by big name tech companies such as IBM, SGI, HP and Dell, even the co-processor market is crowded with AMD and NVIDIA dominating.

AMD and NVIDIA both derive their chips from the latest generation graphics processors, which are both exceptionally good at crunching immense calculations without taking a breath, achieving 2-3 TFlops (trillions of floating point operations per second).

The Intel Phi also started life as a graphics processor, Intel’s attempt to take on AMD in the graphics market. The original chip design named Larrabee never cut the mustard as a graphics processor, early on though engineers at Intel did note the chips number crunching abilities. After cancelling Larrabee Intel went back to the drawing boards and Phi was born, now internally named Knights Corner. As a graphics processor Intel’s Xeon Phi makes a great super computer.

The humble everyday x86 Xeons have been Intel’s most successful entrant into the server market, by accident more than design though. Now Intel has a new chip to jam in the many blade slots of today’s supercomputers, a new weapon with which to assault the Top 500 supercomputer list.

One of the first supercomputers to use the Phi in a full blown installation will be the Stampede Supercomputer, capable of 10 petaFlops when complete and requiring 6 megawatts of power to run the beast. IBM’s currently top ranked supercomputer ‘Sequoia’ is capable of 16.32 petaFLOPS (16,320 tFLOPS), for comparison purposes.

Stephen Hawking and Cambridge University have already taken delivery of the first sample 5110P chips, integrated into the latest SGI UV2000 mini supercomputer. Hawking said in a statement ‘the chip gives us the ability to focus on discovery and continue to lead the worldwide efforts to advance the understandings of our universe.’

While the Phi family won’t see the light of day on many desktops it will make an important contribution behind the scenes, helping to answer the big questions. What will the weather be tomorrow? Who makes the fastest chip? Whatever the answer at the very least Intel will put pressure on the competition and keep the evolution of the supercomputer moving at pace.

Reference: Intel Xeon Phi
Reference: Toms Hardware