Home  |  News  |  Reviews  | About Search :  HardWare.fr 

MiscellaneousStorageGraphics CardsMotherboardsProcessors
Advertise on BeHardware.com
Review index:
NVIDIA GeForce 8800 GTX & 8800 GTS
by Damien Triolet
Published on November 13, 2006

NVIDIA is the first company to release a DirectX 10 compatible GPU on the market. This is an important strategic move as this will naturally become the reference for developments based on this API. What kind of improvements will this architecture bring? What about efficiency? And more importantly for us, is this new GPU, the GeForce 8800, really interesting for current games?

The first DirectX 10 GPU
Each new version of DirectX, the programming interface used by most computer games, is an opportunity for GPU manufacturers to develop a new architecture to support new capabilities while noticeably improving performances. The release of DirectX 10 (we already discussed this API in this article, is an important step as there are numerous modifications. It’s incompatible with previous graphic cards and Windows XP, and DirectX 10 will only work with new generations of cards and Windows Vista.

Innovations mainly concern shaders (small programs that allow the rendering of complex 3D images) and developers who will have a "cleaner" DirectX. Just to remind you, 3D rendering roughly consists of calculating the position of vertices (summits of polygons that form objects), gathering these vertices in triangles, cutting the triangles in pixels, and finally applying textures and other effects on the pixels. DirectX 9 allows the execution of these operations (shaders) on vertices and pixels, whereas DirectX 10 also allows it on triangles (and every other primitives). Why? At the beginning, it will probably be mainly used to optimise performances, with more efficient rendering techniques but it also allows the division of a triangle into smaller triangles to increase geometrical details of an object.

These small programs executed by the GPU are becoming more and more complex with the evolution of image quality. With DirectX 10, the shaders move from a 3.0 to a 4.0 version. The latter allows much longer programs with more flexibility.

To support these 4.0 shaders, NVIDIA developed a new architecture different from the GeForce 6 and 7. These previous GPUs had shader processing units for vertices or pixels. The GeForce 8 ends this process and only has “unified units”, capable of processing all types of data. The objective is to avoid units that run uselessly or not at all. For example, with the GeForce 7, if geometrical calculations are numerous, the eight units devoted to this operation are saturated whereas the twenty-four intended for pixels do nothing. With the GeForce 8, all calculation units can work on these calculations. It is in this way that GPU rendering can progress.
How many pipelines?
Stupid question for some, but a technical detail that a lot of people look for. Bringing an answer to this question is very difficult, while for NVIDIA’s marketing department it’s quite easy. Their answer to this question is 128 shader processing pipelines. However, it isn't possible to compare these pipelines to previous ones.

Roughly, these 128 pipelines, or shader or stream processors, correspond in terms of calculation capacity to 32 current pipelines or 8 vertex shaders and 24 pixel shaders of the GeForce 7800 and 7900.

Does it mean there is nothing more to hope from the GeForce 8800 than efficiency coming from the unification? Of course not, since NVIDIA succeeded in increasing the GPU frequency in this area to 1350 MHz! This is more than twice as much as the GeForce 7900 GTX. That's not all...

Page index
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24
Next page >>
A brief history  

Copyright © 1997- Hardware.fr SARL. All rights reserved.
Read our privacy guidelines.