Intel's MIC Launch Part I: Xeon Phi Many Core Architecture Product Family

After months and months of information leaks, vague promises and deliberate hints, Intel has finally launched the Knights Corner many integrated core coprocessor, as well as a new brand.

Many Integrated Core (MIC) is a multi-core computer architecture developed by Intel, based on the Teraflops Research Chip multicore research project, the Intel Single-chip Cloud Computer multicore microprocessor and, last but not least, the Larrabee graphics computing accelerator that never was. Larrabee was supposed to be a discrete graphics adapter based on the x86 architecture. Unfortunately, whether because of flawed design philosophy or other reasons, the whole project crumbled before anything became of it. The work was shelved though, not scrapped. In fact, Intel applied the lessons learned from it while creating the many-core architecture. Up until last week, we, like most everyone else, expected Intel to employ “MIC” (Many Integrated Core) as the name of the brand.

During the press briefing regarding the new Top500 list though, Chipzilla surprised us by announcing a new brand: Xeon Phi. It will be used for all Intel MIC products, present and future. Knights Corner is the first product in the line, but we should say right now that it hasn't truly been released yet. Intel has only launched the brand. Even though KC is real and working and ready for HPC implementation, the product release, and a tried and true supercomputer, will only come later in the year (2012). However, the Santa Clara, California-based company did provide us with the general specifications. One standard PCI Express card will have over 50 cores with 3D tri-gate transistors, built on the 22nm manufacturing process. 8 GB of GDDR5 or more will assist said cores.

The company claims a 1 TFLOP double precision performance, which is a lot but not really that close to the 4.58 teraflops that NVIDIA is bragging about (although, admittedly, that's the single-precision rating). Then again, while GPUs have always had a higher parallel computing prowess than x86, they are only useful when programs know how to use them. Many don't. Thus, users of Intel MIC products can merrily skip the tedious process of learning new programming languages like OpenCL or CUDA, thus reducing the time it takes to build and launch a new HPC solution / supercomputer.

All in all, Intel claims to have made a product that can ensure the best price-performance ratio, simplicity of application and time to market, if not the absolute best data processing capability. Cray argued in favor of that claim, and even announced a supercomputer that they were building and had already received orders for. Wait for part II of story to read what we found out about it.

Posted in: Hardware, News, Preview