With Intel’s release of the Xeon Phi coprocessor cards, HPC users must ask themselves how much performance they need and how they plan to achieve it. Will resources be devoted towards fast new hardware, re-writing/optimizing software or some balance of the two?
Diverse Software Support for Phi Coprocessors
The resurgence of coprocessor and acceleration cards over the last few years has caused many HPC users to re-evaluate their software applications. Supporting a hardware architecture other than traditional x86 CPUs takes some amount of effort. Intel hopes that their Xeon Phi coprocessors, which use x86 cores, will be very attractive.
There are many options for those considering the transition to Xeon Phi. They might almost be thought of as tiers of increasing effort with corresponding gains in performance. In short, you can recompile your code, link to an accelerated library or dig into the application source code for further tuning.
Start on Intel’s Many Integrated Core architecture guide to explore Xeon Phi. Intel maintains this common landing for new developers with architecture overviews, quick start guides, system software developer’s guide, and workshops.
Many compiler and library products already support Xeon Phi:
- Intel Math Kernel Library (MKL) 11.0 now supports Xeon Phi Coprocessors. See more information in the release webinar. The Intel MKL is bundled with most of Intel’s compiler products.
- Intel Parallel Studio XE 2013 supports developing and optimizing for Xeon Phi Coprocessors.
- Intel Cluster Studio XE 2013 with Intel MPI Library 4.1 now supports Xeon Phi Coprocessors.
- OpenACC support for Xeon Phi will be coming in 2013 from the Portland Group (PGI). The press release from SC12 has further information, and future newsletters will cover the release.
- Additionally, many 3rd-party cluster software products now support Xeon Phi:
- MPICH MPI
- Bright Cluster Manager from Bright Computing
- Adaptive Computing MOAB
- PBS Professional
- Univa Grid Engine
- IBM Spectrum LSF
Intel Xeon Phi coprocessors Released!
For those that haven’t seen the specs of Intel’s many-core x86 MIC architecture product, here are the highlights:
- 60 cores @ 1.053 GHz in a ring architecture
- Over 1 TFLOP double precision floating point performance
- 8 GB GDDR5 memory and 320 GB/sec of memory bandwidth
- 225W TDP and passive cooling for Xeon 5110P
- Compatible with most of Microway’s servers designed for GPUs
Intel will offer passively-cooled and actively-cooled Phi coprocessors for servers and workstations, respectively.
Learn more about accelerating your applications with Intel Xeon Phi
Learn more about Microway’s Xeon Phi-enabled workstations, servers and clusters.