SAN FRANCISCO: CHIPMAKER Intel said it expects the Xeon Phi architecture to be used in a general computation role, effectively taking the place of existing Xeon processors.
Intel finally showed off its Xeon Phi accelerator board at the International Supercomputing Conference back in June as an 50+ core accelerator card to be deployed with its Xeon processors. Intel then revealed to The INQUIRER that it expected Xeon Phi to be a coprocessor part, and now the firm has said it expects the Xeon Phi to eventually become a general purpose processor.
Steve Pawlowski, CTO of Intel's Datacenter and Connected Systems Group told The INQUIRER, "Effectively the Xeon Phi core evolves to the point where it does the general purpose computation, and that maybe the bigger core with Xeon Phi today. I don't really look at it as Xeon and Xeon Phi, I look at it as IA [Intel Architecture] and we've got a lot of them. They are either on the same chip or on the same substrate and they are sharing work amongst themselves."
Pawlowski also warned that despite the chip industry banging on about multithreaded performance, Intel still has to work in single thread performance. "We can't forget about single thread performance, it is important. Every computational code has an element of Amdahl's Law so you can't just say I'm going to focus on throughput at the total detriment of single core performance."
Intel also revealed that it has been working on single thread performance in its upcoming Haswell architecture, with the firm saying that it is making "more resources available to single threads", though surprisingly it didn't go into more detail at this point.
While Intel's Xeon Phi is effectively cut down x86 cores, the plumbing to feed 50 or more cores is something that will be very useful to Intel. As for Intel's plans to scale up the Xeon Phi core count, Pawlowski simply said, "Fifty cores is nothing, we're looking at thousands." µ