Over the past year, barrels of ink have been used to describe all the things being doing wrong in the quest to reach exascale-levels of computation. Disagreement on what direction to take, the evolution vs. revolution argument, undoubtedly has impacted exascale progress – but thankfully has not stifled it completely. Today, we finally have a number of reasons to actually feel encouraged about exascale progress as the dark clouds of confusion and skepticism give way to glimmering rays of technological hope.
The first area of promise, a foundational step toward achieving exascale, is being researched and discussed by both Intel and Nvidia, but our feature article on this topic comes from Intel. We interviewed Intel Fellow, Vivek De, to discuss Near Threshold Voltage processing, or NTV. NTV processing is a research area that holds tremendous promise for more efficient power management, and is applicable to numerous future computing applications ranging from mobile applications to HPC, and is likely to be one of the critical technologies required to enable power-efficient exascale systems. This power management approach, which has been demonstrated with functioning research prototypes, including a solar-powered processor, still has a long way to go. But, in our opinion, NTV is indeed a glimmering ray of hope.
The other ray of sunshine comes from Nvidia with some recent announcements made at their GPU Technology Conference. We were impressed with two innovative components of the Keplar GPU called Hyper-Q and Dynamic Parallelism. We tend to refer to these two as one, as they work hand-in-hand to improve the Keplar GPU’s performance by a factor of 3 to 4x per watt from the previous generation of Fermi GPUs.