CHIPMAKER Intel has seen a cluster powered by its Xeon Phi chips debut on the Top 500 list at number 150.
Intel's Xeon Phi, the firm's rebrand of its Many Integrated Core (MIC) architecture, has finally been deployed in a cluster and shown that it can at least run the Linpack benchmark. The firm's Discovery cluster has 9,298 cores, with Intel claiming that each Xeon Phi accelerator delivers 1TFLOP of double-precision floating point compute performance.
While Intel bought Cray for its Aries interconnect, at this time all the firm would publicly say is that its Discovery cluster uses Infiniband to connect the 72 racks full of Xeon E5-2670 processors and Xeon Phi accelerators.
Rajeeb Hazra, VP of Intel's Architecture Group said its Discovery cluster is a prototype that still needs hardware and software tuning. Hazra said Intel is "on track to have Xeon Phi introduction on the market by end of 2012".
For Intel, all its Xeon Phi powered Discovery cluster shows is that the firm actually has working Xeon Phi boards. While the firm has seen Xeon Phi enter at 150 on the Top 500, it is a position that is hardly going to set the world on fire, especially as Nvidia's Tesla accelerators continue to dominate the upper echelons of the Top 500 list, despite the next generation Tesla K20 accelerator being still a few months off.
Intel needs two things to happen in the coming six months. The firm needs not just more Xeon Phi powered clusters in six months but ones that are in the top 30, and it needs to show that hardware and software tuning can produce tangible improvements in Xeon Phi performance if it's going to compete with Nvidia's upcoming Tesla K20 boards. µ
Sign up for INQbot – a weekly roundup of the best from the INQ