Skip to main content

Rise Of The (Video Game) Machines

These video game characters are excited that GPUs will soon be able to do more than just bring them to life.

Video games are big business. And the push for faster, flashier graphics has driven not only game developers, but the development of computer hardware to handle those graphics. It’s gotten to the point where computers have a separate “brain” that only handles graphics – and these graphics processing units (GPUs) are usually 100 times more powerful than the central processing units (CPUs) that handle pretty much everything else your computer does (we’re talking teraflops versus gigaflops). Now researchers have come up with a way to put the power of the GPU to use for functions other than video games.

It shouldn’t be that hard to tap into the power of a GPU, right? After all, it’s just another computer processor. Wrong. GPUs are designed to handle graphics data efficiently, which means their hardware was developed to handle lots of separate operations – with each operation corresponding to a single pixel on your computer screen. In other words, performing complex computing functions is difficult for GPUs because they are not good at running operations that need to communicate with each other. Think of GPUs as running isolated computing silos instead of interconnected computing webs.

Enter Huiyang Zhou. Zhou has led a team of computer engineers in developing new compiler software to address the problem. Their compiler effectively takes a program and translates it into a GPU-friendly format. I won’t go into the nerdcore details, but think of the compiler as a mother bird chewing up a program and spitting it out so the GPU can process it more easily.

But does it work? Developers have been trying to do this for a while, and GPU developers have even tried optimizing code for some benchmark programs to get them to run more quickly in GPUs. The compiler, being a computer program, does the optimization a lot faster than a human developer could. And the researchers have found that programs optimized using the compiler run 30 percent faster on GPUs than programs optimized by the GPU developers.

Hopefully that means GPUs can soon get down to work, and stop spending all their time playing video games.