The high-end graphics card market isn't the straightforward, mano-a-mano battle it used to be. Graphics cards with dual GPUs, systems with multiple cards in CrossFire or SLI implementations, and most recently hybrid graphics have all muddied the picture. Meanwhile, Intel has improved its integrated graphics and has long-term plans in Larrabee to challenge AMD and Nvidia in discrete graphics.
But next month we should get a good, old-fashioned GPU showdown when AMD and Nvidia release entirely new, high-end GPUs. Details of these GPUs, AMD's RV770 and Nvidia's G200, and the graphics cards based on them, are trickling out.
The RV770, or ATI Radeon 4800 series, will launch on June 16, according to several sites. Graphics cards based on the first GPU in the series, the Radeon 4850 with 512MB GDDR3 memory, will be available immediately for $230. This GPU is the follow-up to the current Radeon 3850. Higher-end cards based on the Radeon 4870 with 512MB GDDR5 and the dual-GPU Radeon 4870 X2 with 1,024MB GDDR5--a first--will be announced on June 23. The Radeon 4870 cards won't ship immediately because of the lack of availability of GDDR5 memory, but when they do show up prices should be around $350. The Radeon 4870 X2 cards won't show up until early 4Q and will cost about $550.
One of the more intriguing specs of the Radeon 4800, according to the site TGDaily which has many details on the 4800 series, is "physics processing capabilities." Physics is used to define how objects in a game move and interact. It can deliver a more realistic, compelling experience, but few games are designed to take advantage of physics.
There are really only two players here: Havok, which makes software for game developers and movie editors, and Ageia, a competitor best known for pioneering the PhysX Accelerator boards for gamers. Intel purchased Havok in September 2007, and Nvidia completed the acquisition of Ageia in February and plans to integrate the PhysX Accelerator into its GPUs. ATI previously worked with Havok on physics (and AMD reportedly considered buying Ageia late last year), but it is unclear where AMD is getting the physics technology now. (There have been some reports that AMD borrowed a programming language developed at Stanford University for the GPUs that power the Folding@home research project and extended it to support game physics.)
Nvidia's G200 will reportedly launch on June 18 (some sites put it in early July). This GPU family was previously known at the GeForce 9900 series, but several sites are now reporting that Nvidia will switch to a new naming scheme: the GTX 200 series. Initially this should consist of two products, the GTX 280 and GTX 260, but otherwise there are few details aside from the fact that G200s supposedly contain more than 1 billion transistors (not including memory), use GDDR3 memory with a 512-bit interface, have an updated Unified Shader architecture, and will be in cards that cost $450 or more.
With the massive Computex tradeshow in Taipei coming the first week of June, I suspect we will know a lot more about these new GPUs before the rumored announce dates.