August 06, 2012
Intel’s upcoming Xeon Phi processors will be the company’s direct answer to NVIDIA’s Tesla and, to a less extent, AMD’s Firestorm GPUs. For the initial Knights Corner product, Intel has kept most of the critical elements (core count, clock frequency, attached memory capacity) under wraps, but recently, VR-Zone described the specs on some of the pre-production silicon. While much of the story is based on unnamed sources, it nonetheless provides an interesting perspective on the development of the chipmaker’s first manycore architecture.
The architecture, known as Many Integrated Core (MIC), is based on the failed Larrabee chip, which was initially aimed at the visual computing market. In December of 2009, Intel scrapped the effort, citing hardware and software schedule slips. But the design was salvaged, with the Larrabee architecture reportedly coupled with Aubrey Isle silicon and pitched to select partners as a manycore HPC coprocessor.
The original ASIC, manufactured using a 45nm process, became the development platform known Knights Ferry. It incorporated a 32-core chip and 2 GB of on-board GDDR5 memory. The Knights Ferry card delivered 750 gigaflops single precision and sub-400 gigaflops double precision.
In 2011, Intel began to develop Knights Corner, the first MIC product (now know as the Xeon Phi family), based on its new 22-nm tri-gate process. According to the VR-Zone report, there were three A0 Knights Corner iterations with 48, 52, and 60 cores, and between and 2 to 8 GB of on-board memory. These first versions were said to consume roughly 300 Watts.
Supposedly, Intel used the 60-core version to demo the 1-teraflop double precision capability to the public. Despite those improvements, it was likely to fall short of the expected performance of NVIDIA’s upcoming Tesla K20 GPU, which is expected to deliver well north of 1 double precision teraflop.
The next major Knights Corner revision, B0, now includes higher core counts, more memory, and better thermal design. Specifically, this revision comes in 57-, 60- and 61-core versions and 3 to 8 GB of GDDR5 memory. Although three of the new boards still consume up to 300 watts, two have been cranked down to 245 watts. Other features have been modified as well, including a larger on-chip cache, higher memory bandwidth, and a tweaked BIOS. While even this revision may not match the performance of the K20, Intel apparently believes it will eventually win the battle against its GPU competition.
Full story at VR-Zone
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?