Intel sets a ten year goal for exascale computing

CHIPMAKER Intel is turning its thoughts and focus to the upcoming world of 'exascale computing'.

Intel CTO Justin Rattner's keynote speech rounded off this year's Intel Developer Forum in San Francisco with a look into the future. He announced that Chipzilla's ten year goal is to achieve a 300 times improvement in energy efficiency.

Five years since Intel introduced the Core Duo, Rattner said, "Our ten year goal here is to achieve, nothing modest, a three hundred times improvement in energy efficiency for computing at the extremes."

"That means the equivalent of 20 [picoJoules/FLOPS]. Trust me we're towards a magnitude away from that goal today."

Compared to today's Xeon system that uses 200W, the aim is to reach an exascale 100 GFLOPS system that uses less than 2W by the year 2018. Intel fellow Shekhar Borkar said, "This is only possible if you look at the entire stack and innovate across all aspects of the system."

The keynote included demonstrations of the practical uses of multi- and many-core technology. Rattner showed off a machine using a 'Parallel Javascript' engine to speed up web browser based service and greatly improve graphics.

Also on stage was a tower PC running a Core i7 processor, which was being used as an LTE 4G base station. Another computer ran a facial recognition system that kept images encrypted to the owner and only showed certain images to other verified users. Last of the four uses of many- and multi-core technology shown was a server capable of processing more than 800,000 transactions per second.

An interesting demo towards to the end of Rattner's keynote was a quick look at a 'Hybrid Memory Cube', which showed off DRAM delivering a speed of over 120GB/s. Rattner said, "So it's not only the fastest DRAM ever demonstrated, but it's probably the most energy efficient DRAM." µ