As powerful as today's super computers are, it's widely believed that we'll see a tectonic shift in supercomputing in this decade - a shift that will leave the world's current super computers far behind. That change is exascale computing technology, which will be thousands of times faster than petaflop super computers. Today, NVIDIA announced that it has a $12.4-million contract to further its exascale computing research.
Several U.S. government agencies have an interest in this kind of computing power, for reasons as varied as national defense, medical research, and engine improvement. The Department of Energy has been one of the most visible proponents of making exascale computing a reality and it's the agency that awarded NVIDIA the two-year contract, via its FastForward program.
Kepler SMX technology provides more cores than the venerable Fermi shader multiprocessor (SM).
Also, control logic is reduced, making for more room and less power consumption.
So, if exascale computing is so important to achieve, what's the holdup? Electrical power, for one thing. NVIDIA chief scientist Bill Dally described it this way in a blog post today: "One of the great challenges in developing such systems is in making them energy efficient. Theoretically, an exascale system could be built with x86 processors today, but it would require as much as 2 gigawatts of power — the entire output of the Hoover Dam."
Dynamic Parallelism lets the GPU create new threads without going back to the CPU,
It makes parallel processing more accessible to developers.