Original URL: http://www.theregister.co.uk/2006/06/07/intel_x86_2010/
Intel together forever – or at least 2010 – with x86
EPIC battles put on hold
Intel seems to have learned its lessons from AMD's quick embrace of x86-64-bit extensions and the Itanium debacle. It's sticking with x86 chips for the foreseeable future.
"We pretty much have x86 as the going assumption," said Intel's CTO Justin Rattner, speaking here today at an Intel Labs press event. "As we talk to our partners and customers, we get tremendous resonance around (x86)."
Intel in tandem with HP tried to move away from the x86 instruction set by creating the EPIC instruction set used in Itanium processors. Rather than mimicking Itanic, AMD moved forward with 64-bit versions of x86 chips – a decision that proved to be the better bet. Itanium server sales have been about 96 per cent below IDC's early forecasts, while AMD and now Intel have enjoyed a successful transition to the x86-64-bit market.
Rattner noted that there's a tendency to look at creating new instruction sets when things aren't "going as well as you expect." Ultimately, many of these specialized efforts morph back into general purpose computing parts. So, Intel now "tries to resist the natural urge to invent something new."
It's a lesson that only cost Intel a few billion dollars to learn.
In the future, Intel wants to build x86 chips that consume 10x lower power than today's parts while delivering 10x more performance, Rattner said.
He outlined a vague plan that will see Intel surround multiple processor cores with "reconfigurable caches" and "high bandwidth memory."
"These cores are interconnected with some scalable fabrics," Rattner said. "A lot of our research, in fact, involves what that fabric might look like."
Intel has also put a system in place to test out software workloads of the future. The idea is to "guess what kinds of applications" these large, multi-core systems will be running. Intel has poured "substantial time and money" into this process.
When will Intel's true chips of the future arrive?
"We nominally target basically the end of the decade for the initial introduction of these technologies," Rattner said. "We may beat that in a particular instance." ®