Nvidia details next-gen GPUs

Nvidia CEO Jen-Hsun Huang has revealed preliminary details of two future GPUs: Kepler (2011) and Maxwell (2013).

According to Huang, the 28 nanometer Kepler will offer a 3-4x performance per watt improvement compared to Fermi, while Maxwell is expected to boast an approximate 16x efficiency leap.

“Fermi is only the beginning. The arrival of Maxwell will mean 16 gigaflops per watt, which represents a 16x improvement over Fermi,” Huang said during the company’s 2010 GPU conference keynote speech. 

“We are constantly learning about barriers and walls and striving to achieve what we at Nvidia refer to as speed of light parallel computing.

“You could say we are are engaged in a fanatical drive to maximize performance per watt. Clearly, Kepler and Maxwell represent a huge step in that direction.”

Huang also noted that “between now and Maxwell,” Nvidia remained on track to introduce a number of GPU-based optimizations, including virtual memory and pre-emption.

“We’ll enhance the ability of the GPU to autonomously process so it isn’t blocking or bottlenecking the CPU. This will [help] take GPU computing to the next level, while providing a significant [boost] in performance,” he added.