Intel and Cray: Pushing a Divergent Architecture?
- 0
- Add a Comment
Over at Ars Technica, an article describing the upcoming marriage of Intel processors, graphics technology, and Cray knowledge shows that it is quite possible that the future of high performance computing just may change standard personal computing as we know it, and in the process, threaten the existence of AMD.
Intel is pursuing yet another shrink of the circuitry, down to 32nm, and adding the Quick Path Interconnect, essentially putting the most used control right on the chip, as AMD had done with K8. The design will add something Intel is calling Advanced Vector Extensions, described as a 256-bit version of the current SSE registers, which are 128 bits wide. The AVX extensions will also increase the number of operands with each clock cycle to 3. These extensions will be a step away from the planned SSE 5 from AMD.
The system will incorporate the Intel Larrabee graphics solution, which, although slammed by that company of the green logo, seems to promise incredible video performance. Intel seems to be moving into an era of having so many cores on a single die that, unlike nVidia designs, it really doesn’t matter how much work each individual core does. Intel seems to be fairly good at scaling cores on a chip, so if nVidia has X performance on their chip, and Intel has only 0.1X performance per core ( a good estimate, by all accounts), it will be semi-trivial to ramp up 12 cores on a chip to leapfrog the performance of nVidia. Each Larrabee core is estimated to have twice the number of FLOPS as the core CPU, with the path being 512 bits wide. nVidia isn’t out on this one yet, but it certainly can’t afford to be sitting on any laurels, or simply continuing at current speed. It is time to pick up the pace.
What becomes the harbinger of doom for AMD is that the commitment to this structure is not trivial, and is so different from today’s offerings that it would seem to necessitate major changes outside the core CPU and graphics. Perhaps it will finally see the GPU becoming a pin package, with a socket on the motherboard, and, to the casual observer, seeming like a second physical CPU.
Since AMD is usually a half-step to a step behind of Intel, simply due to size of delivered products, it will need to get positively gonzo about the speed with which it proceeds when the change occurs.
Are we at another place in time like 1980, just before the original IBM PC was released? Will a new architecture for personal computing make the current one seem as quaint as the PC made the Apple II?
-
Quote of the day:
Read, every day, something no one else is reading. Think, every day, something no one else is thinking. Do, every day, something no one else would be silly enough to do. It is bad for the mind to be always part of unanimity. - Christopher Morley
