So you've just downloaded that Beyoncé clip from your favorite video site, and now you want the hip-hop queen on your iPod. The trouble is, she's in AVI format and the iPod plays MPEG-4. Unless you've got some serious processing power, the conversion could take the better part of an hour. The right CPU could save countless minutes. The right GPU could save you even more.
By now you know that there's more than one processor driving your PC. In addition to the CPU, or central processing unit, there's the GPU, or graphics processing unit. No doubt, you also know that GPUs play an extremely important role in system performance. These are the chips that actually build the images popping up on your Windows desktop.
What you may not realize is that GPU power is on the rise. Today, even lowly integrated graphics chips-inexpensive parts woven into motherboard chipsets such as Intel's GMA900 and 950-do a pretty good job of rendering 3D images. Meanwhile, the leading GPUs, the sort that drive graphics cards plugged into high-speed PCI x16 slots, are morphing into CPU coprocessors , taking on more tasks than ever before and relieving some of the burden on your main processor.
These powerhouse chips are a far cry from the early, fixed-function 3D cards that ran your PC in the mid-1990s. They offer tremendous floating-point performance, a big boon for high-end graphics apps as well as engineering and scientific software. They're highly programmable. And they're parallelizable . What does that mean? An ATI Radeon X1900 series card includes 48 arithmetic logic units, or ALUs, each capable of handling 32-bit floating-point math. That means it's well suited to tasks easily broken up into small chunks. Each chunk can run simultaneously.
One application is 3D acceleration, the primary purpose of a graphics card. Then there's physics calculation for PC games-the math that determines, among other things, how the dead guys tumble down a flight of stairs during a first-person shoot-'em-up. While companies like Ageia are developing chips dedicated to physics acceleration, the parallel computational engines built into today's graphics cards are ideal for many types of physics calculations. nVidia has been working with Havok, a developer of physics APIs (application programming interfaces) for games, in an effort to bring physics acceleration to the GPU. (nVidia's Havok solution also works on ATI cards, and ATI plans to unveil its own solution by the end of the year.)
GPUs can also handle video transcoding , the conversion from one video format to another. With ATI's Avivo technology built into Radeon X1900 cards, you can take a video stream rendered for one codec and quickly transcode it into another. With, say, a Beyoncé AVI file, you can quickly convert it to an MPEG-4 file for playback on your iPod. Along the same lines, graphics developers are discussing the possibility of using GPU horsepower to offload audio processing from the CPU.
A project called "General-Purpose Computation on GPUs," GPGPU for short, is spearheading the most ambitious use of graphics processors. The goal is to handle a wide range of parallelized tasks, including signal processing, offline 3D rendering, computer vision, and more. You can find out more about GPGPU.
How does running applications on a GPU compare with running them on a standard CPU? It really depends on the task. When developing Avivo, ATI discovered that video transcoding on the GPU can occur anywhere from four to eight times faster than on a high-end, dual-core CPU. Of course, CPUs are always evolving. Next year, we'll see quad-core CPUs, which could close the gap on some applications
Back To Top
