Graphic ExtremeVery often we ask ourselves does ATI or NVIDIA dominate the 3D accelerator market. This isn’t the right question, because the answer is Intel!
With more than 50% of market shares, the processor manufacturer is the undisputed leader. This domination is due to integrated chipsets, which are far from being unanimously accepted because of their reduced performances. The previous graphic core name was even ironic, Graphic Extreme. Despite their slowness, their technological delay and the number of games that were not supported, they weren’t exactly extreme. Of course, this only concerns 3D and not everybody needs it. Why use a complex GPU for a computer only dedicated to office use? Integrated graphic cores fit this role perfectly.
But Intel doesn’t only attack the "2D" card market and clearly aims for more by targeting polyvalent machines or even game computers by increasing its prices in adding millions of transistors for 3D support. Today Intel integrated cores are called GMA (for Graphic Media Accelerator), they entirely support DirectX 9 and Shaders 2.0. Of course, this support will be useful for the office when Longhorn will be available, because the new interface will use 3D and will need shaders 2.0. But for the time being, it’s clear. When Intel sells this chipset it clearly indicates that gaming is possible. So, is this really the case? We will see with the latest generation of Intel chipsets, the GMA 900 and 950.
Entry level evolution ?
Here’s a graph with the evolution of fillrate of entry level and high end products. We included all NVIDIA chips sold since 1999 from the TNT2 to the GeForce 7800 GTX. Results would have been similar with ATI graphic cards.
The difference between the two extremes of a GPU range is always accentuated. It’s important to know that these graphs only report raw data without taking into account architectural optimisations (and possible reductions due to the filtering quality via drivers) which can significantly increase performances. It’s also important to specify that high end GPUs include more improvements than their entry level equivalents.
Bandwidth doesn’t evolve much over the years. Even if it increases with the GeForce 6200 TurboCache, less memory is available at full speed. On the other hand, bandwidth went through the ceiling in 2003 with high end products.
Fillrate follows a similar path. If differences remained 1.5 or 3 times more between two extremes for a long a time, with the massive parallelisation of tasks of the last high end GPUs it increased to 6 – 7. In short, if high end GPUs would have evolved at the same rate as entry level GPUs, the current high end product would be a 6600 GT. If entry level products would have evolved at the same rate as the high end, the current entry level would also be a 6600 GT.
What does this tendency mean? That reducing entry level products costs is very difficult and slows down evolution. To compete with Intel’s integrated solutions, and to try and place a GPU in every computer, ATI and NVIDIA released products which compared to high end graphic cards produce lower and lower performances. In the end, the two GPU manufacturers probably helped integrated solutions to gain market shares. With the continuous reduction in performance of entry level products, the gap between entry level graphic cards and integrated solution reduces. On the one hand, the computer manufacturer has to spend approximately $25 for an entry level graphic card or only $5 to change from a standard chipset to an integrated one. This makes quite a big difference on paper as both products can run games.
Wouldn’t it have been best for ATI and NVIDIA not to release such cheap and low performance 3D solutions? To not try and place a $25 graphic card in every computer, but to convince gamers to spend $100 for a 3D graphic card and that polyvalent computers must be equipped with such a graphic card?
Developers could have pushed game graphic quality and not worried about the 50% of computers equipped with integrated graphic chipset and also the number of computers equipped with low performance graphic cards.
Let’s be clear on this point, ATI and NVIDIA’s solutions are better (you will see that in the next pages) than Intel integrated chipsets. But is the gap big enough? Intel is currently working hard on improving graphic solution performance on the driver and architectural levels. If Intel could have a product capable of directly competing with ATI and NVIDIA for performances and quality/reliability, these two would have something to worry about.
Isn’t it time for ATI and NVIDIA to develop an architecture fully adapted for entry level products? Today the “easy solution” is to avoid the production cost explosion for hardware and software by using the same architecture for the whole product range. The cost of this architecture can, however, be a problem for entry level products. Fewer useless options with higher performances, this is the eternal debate.