A Richer Gaming Experience

Everyone has been blabbing on about how physics and the graphics cards that accelerate them will change the face of gaming and the graphics card industry. Well, so far, software physics APIs still reign supreme. Even Ageia's PhysX has been an abysmal failure because all it did was make the few games that supported it look prettier, without making it any more immersive. Perhaps it would have been better integrated into a GPU, instead of being sold as a separate chip.

Both NVIDIA's Quantum Physics Technology (QPT) and ATI's competing Close To Metal (CTM) technology use middleware-based SDKs like Havoc FX to offload physics processing to the GPU. They do this by processing physics instructions as they would shader instructions. All thanks to the new unified architecture. Both NVIDIA and ATI will provide a low-level method for programmers to issue commands to the GPU, much like what ASM or Machine Code is to processors. So when they are implemented, we can count on more realistic physics in our games.

But that's not all when it comes to ATI. Take a look at the current slew of ATI-based motherboards. Notice the fact they have three PCI Express X16 slots? Well, ATI has always been planning to introduce their “Triple-Play” concept. Of course, a system with three graphics cards is a really hard sell, much harder than getting people to buy two cards.

Still, ATI has laid the ground work not only for physics offloading to the GPU, but also the offloading of physics processing to a dedicated graphics card. For those who need nothing less than the best, ATI may just have what it takes to offer the ultimate, if ridiculously extravagant, graphics and physics processing solution.

What Does This Mean To Us?

Well, prettier games and the ability to dedicate a graphics card to solely process physics while retaining CrossFire functionality. Cool.

GPGPU Redefined

GPUs are massively-parallel processors, with many processing pipelines compared to the superscalar x86 processors we use in general computing. This massively-parallel architecture can be harnessed to greatly accelerate the processing of certain instructions. In fact, using GPUs to process those instructions can improve processing speed by up to 20 times. This also frees up the CPU to work on other instructions.

ATI's Close To Metal (CTM) technology uses a programming language that allows a programmer to skip the Application Protocol Interface (API) and directly access the GPU. Traditionally, if you want to process something on a GPU, you would have to use an API to do it. For example, you would have to issue a processing instruction via DirectX (via BrookGPU) or OpenGL.

As APIs are very command-centric, they are limited in functionality. With CTM, well, you can skip the APIs and directly control the GPU. This should improve performance levels and allow them to do tricks with the GPU that is normally not possible with APIs. It's generally the same idea behind choosing ASM over CPP.

Stream computing, or the moniker given for processing on massively-parallel processors like GPUs, is currently a big AMD initiative and for good reason too. Stream computing can be applied to a wide variety of tasks that support high levels of parallelism. A good example would be games optimized to run on IBM's Cell microprocessor.

So, the potential is really limitless. A good example of stream computing at work would be running Folding@Home clients on the ATI R5XX-based graphics cards and Playstation 3 clients. They offer far greater processing power than that of traditional x86 processors.

What Does This Mean To Us?

Right now? Not much. Even saving lives by helping cancer research through the the Folding@Home project takes time.

However, in the future, we can look forward to better support for physics processing and acceleration of very parallelized code. Hopefully, we'll even see IEEE-754 conformity.