The decision to use GDDR3 memory has cured any potential memory bottlenecks, allowing frequencies up to 800MHz (1.6GHz effective). The need for faster memory isnâ€™t something thatâ€™s escaped nVidia either, which is why youâ€™ll see GDDR3 memory on the GeForce 6800 cards as well, running similarly high frequencies.
But it wasnâ€™t just hardware advances that ATI was showing off in Toronto, it was also very proud of a new compression method called 3Dc. 3Dc allows efficient compression of normal maps. Now, normal maps are used extensively in films and the latest games to provide very lifelike and detailed models. Basically, a normal determines how a model reacts to light sources and pretty much acts like an arrow, telling the light what direction to reflect in. Itâ€™s this definition of surface reflection that produces a three-dimensional illusion. A normal map is a texture that stores the normal information for a surface, which can then be wrapped around a bunch of polygons and provide a convincingly lifelike 3D model.
What makes normal maps so useful is that by wrapping the polygons up in this way, far fewer triangles are needed. Traditionally, higher polygon count equalled higher detail, but higher polygon count also equalled higher system load. So, by wrapping your polygons up in a normal map, you can significantly reduce the amount of polygons required and consequently reduce system load.
Unfortunately there are a couple of problems with normal maps. The biggest problem is that they take up a huge amount of texture memory, and even with todayâ€™s cards shipping with 256MB of RAM, itâ€™s simply not enough to store detailed normal maps. The obvious answer is to employ traditional compression techniques to the normal map textures, to reduce their size and increase storage capacity. But this is where problem number two comes in. Unfortunately normal maps donâ€™t react well to traditional texture compression methods like DXTC and S3TC, resulting in a significant quality degradation. This is where 3Dc comes in. 3Dc has the ability to compress normal maps to a ratio of 4:1 with little or no quality degradation, effectively quadrupling the available texture memory.
ATI insists that 3Dc is easy to implement in existing games that use normal maps, and that it incurs very little system overhead. So, with more texture memory available to games developers, the result should be far more detailed and lifelike characters and models, with very little performance hit.
Now Iâ€™m always wary when graphics manufacturers develop new tools that allow games to be tailored to their specific hardware. Thankfully though, ATI has made 3Dc an open standard that any hardware or software vendor can take advantage of.
To demonstrate 3Dc, ATI showed a demo from the forthcoming Serious Sam 2, where one model employed 3Dc and the other didnâ€™t. The results were impressive, but as with any demo, itâ€™s there to show off the hardware in the best light. Iâ€™ll therefore wait until I see 3Dc employed in actual games rather than demos before making a judgement.