MAKER OF EXPENSIVE PRINTER INK HP sees Intel's Xeon Phi accelerator card as a complementary product alongside Nvidia's Tesla GPGPU in its servers.
An IBM representative told The INQUIRER at the International Supercomputing Conference (ISC) that Intel's Xeon Phi is unlikely to gain much favour among its Bluegene cluster customers. However HP has given the Xeon Phi a far warmer reception, saying it has already seen interest in Xeon Phi and it expects Intel's HPC accelerator card to be a complementary product to the Nvidia Tesla GPGPU it already sells in Proliant SL series servers.
Ed Turkel, worldwide marketing manager of HPC at HP quipped at IBM's response to Xeon Phi saying, "That's like asking someone who manufacturer hammers if screwdrivers have a future." While Turkel might have made the comment in jest, he does have a point, as IBM's highest performing Bluegene clusters do not use accelerators even though the firm was showing Intel Xeon Phi and Nvidia Tesla K10 boards on its ISC stand.
While IBM said its customers won't be going all out on Xeon Phi, it looks like HP is more upbeat, with Turkel saying, "There is a great deal of interest in MIC technology, it very clearly has a future. The future is going to be very interesting, Nvidia is doing some very interesting things with GPUs and Intel MIC can be a very interesting alternative and we are not necessarily betting on one or the other, we are very happy to collaborate on both and let our customers decide. We kind of expect that people will find that certain applications run themselves better on one or the other, just as like now not every application scales well with GPUs. We are already talking to customers who are contemplating clusters they are going to build next year and trying to decide whether they want to do GPUs or MIC."
It is interesting to see Turkel saying that even HP, one of Intel's biggest and most trusted customers is hedging its bets with Xeon Phi and the MIC architecture. However given that Nvidia has been in the accelerator business for a number of years and enjoyed considerable success in pushing its GPUs and the CUDA programming language, it doesn't look like Intel's Xeon Phi will waltz in and steal Nvidia's business in one fell swoop, and HP won't make the same mistake it did betting a major part of its business on a single chip after the multi-decade fiasco that has been Itanium.
With Intel being very vague over Xeon Phi performance figures and Rajeeb Hazra, VP of Intel's Architecture Group telling The INQUIRER that the firm is still working on final configurations for Xeon Phi, Turkel said HP has not only specifications but products in the labs that it can share with its customers. "We definitely have more concrete numbers, we actually have samples in the labs and we're running on here in the [ISC] booth, and we have more concrete information and we can share that information with customers under non-disclosure agreements," said Turkel, adding that he could not share those details due to a confidentiality agreement between Intel and HP.
It is interesting to see HP take a softly, softly approach to Intel's Xeon Phi, with users and their workloads making all the decisions. While the firm's approach is quite understandable it could mean that Intel might have to wait a while before it sees significant take-up of its Xeon Phi accelerator card. µ