GPU compiler could "turn desktops into supercomputers"

Researchers plug into graphics chips with novel compiler program

Stewart Mitchell
18 May 2010

A software team at North Carolina State University claims it could turn a desktop PC into a supercomputer by harnessing the power of its graphics processing unit (GPU).

An average CPU boasts between 20-60 gigaflops of computing power, or 20 to 60 billion operations per second, but that pales compared to some modern GPUs that can run 1 teraflop (a thousand billion operations per second).

The North Carolina researchers have written a software compiler that they say can fully exploit the power of the GPU.

“We have developed a software tool that takes computer program A and translates it into computer program B – which ultimately does the same thing that program A does, but does it more efficiently on a GPU,” said Dr Huiyang Zhou, an associate professor of electrical and computer engineering at North Carolina.

The researchers found that programs designed for conventional processors ran 30% faster using the compiler and a GPU, and that could be the tip of the iceberg. “Tapping into your GPU can turn your personal computer into a supercomputer,” Zhou claimed.

According to the programmers, the main problem for GPU computation is that the architecture of the GPU is designed to process graphics and focuses on turning data into millions of pixels on a screen, with those operations taking place in isolation from each other.

Zhou says his team's compiler effectively translates generic software into a program that runs efficiently on this architecture.

Read more about: