Further performance scalability only possible through parallel computing

May 3, 2010 13:34 GMT  ·  By

TSMC and Globalfoundries have both scrapped their respective share of processing nodes and decided to jump directly to smaller technologies, such as 22nm or 20nm. This is necessary in order to keep up with Moore's law, which states that the number of transistors on an integrated circuit doubles roughly every two years. Intel and Advanced Micro Devices have been, and still are, both using this law as a guideline. Moving forward, however, NVIDIA claims that the law will die out and parallel computing will take its place.

According to NVIDIA's VP, Bill Dally, Intel and AMD won't be able to stick to Moore's law for much longer and that adding extra cores to the same processor is like building “an airplane by putting wings on a train.” This, he says, is because doubling the number of transistors in a serial CPU only results in ”a very modest increase in performance” at a “tremendous expense in energy.” On the other hand, “doubling the number of processors,” which is the essence of parallel computing, “causes many programs to go twice as fast.” Dally also places a somewhat strong emphasis on power efficiency, saying that cores should be optimized for efficiency, instead of processing speed.

“More importantly, parallel computers, such as graphics processing units, or GPUs, enable continued scaling of computing performance in today's energy-constrained environment. Every three years we can increase the number of transistors (and cores) by a factor of four. By running each core slightly slower, and hence more efficiently, we can more than triple performance at the same total power. This approach returns us to near historical scaling of computing performance,” Dally states.

Unfortunately for NVIDIA, consumers are most likely skeptical about promises of power efficiency after seeing how power-hungry the Fermi is, compared to ATI's Radeon HD 5000 Series at least. Also, seeing how the two aforementioned main foundries still intend to keep advancing their technologies for years to come, AMD's and Intel's approach should be able to stick around for a number of years.