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

MiscellaneousStorageGraphics CardsMotherboardsProcessors
Advertise on BeHardware.com

News of the day

  • True fake pictures of the 7900 GT
  • G80 will support unified shaders
  • Archives

    Février 2006
    1 2 3 4 5
    6 7 8 9 10 11 12
    13 14 15 16 17 18 19
    20 21 22 23 24 25 26
    27 28

    You can read previous news using our archives.

    << Previous day's news
    February 20, 2006
    Next day of news >>
    February 22, 2006

     True fake pictures of the 7900 GT
      Posted on 21/02/2006 at 15:50 by Marc
    Imprimer cette news Envoyer cette news par e-mail

    DailyTech published the first pictures of the future GeForce 7900 GT and many of you have already probably already seen them.

    The only problem, if we take a closer look, is that the GPU is much too small to be a 7900. It isn't even big enough to hide all soldering points made for the GPU on the PCB of the card. In fact, the chip is strictly identical to the one presented by VR-Zone a few days ago like a 7600.

    It would be at most the first pictures of a 7900 GT PCB including a 7600 chip. We all can't wait to see the first pictures of the G80 memory…

     G80 will support unified shaders
      Posted on 21/02/2006 at 12:42 by Damien
    Imprimer cette news Envoyer cette news par e-mail

    DirectX 10 that will arrive with Windows Vista will include several innovations. A new type of shaders will be released, geometry shaders. They will work on primitive functions (dots, lines or triangles) and will be able to create new ones. Geometry shaders will work on data provided by the vertex shader and their first useful function could be to act as tessellation units to increase geometric details (TruForm, Displacement Mapping etc.). This is however only a small part of what they could do and this possibility won't even be used at first because of a lack of performances. It will rather be an improvement of the particle system (with the obvious possibility to add particles), fur type effects, rendering optimisation in a cubemap and calculation of data by primitive that will help the pixel shader to be more efficient. In brief, it is most probable that the first use of these geometry shader would be more abstract than concrete like the tessellation is.

    Let's get back to the G80 and the discussion on unified shaders. With DirectX 10 and the Shaders 4.0, the vertex shader, geometry shader and pixel shader will use a unified set of instructions that will have the same characteristics. It wasn't the case before. To support the DX10, the G80 will obligatory have to support the unified Shader 4.0 (unlimited number of instructions, full support of whole numbers additionally to floating points, logical operators support etc.). How the Shaders 4.0 will be implemented in hardware is entirely at the discretion of the GPU designer. It can else use separated units to process vertices then primitives then pixels or a single type of units that is capable of processing all of it, or a mixed vertices/primitives unit and a unit for pixels etc. DX10 imposes the unification at the programming model level and opens the possibility for the manufacturers to also do it on the hardware level but only if they find it useful.

    In a GPU, pixel shading pipelines or pixel shading units are the most complex because they are optimised to mask texture access latency. Other units aren’t and are in consequence much simpler. With the Shaders 4.0, vertex shader and geometry shader units have some restrictions on texture accesses because LOD can't be automatically calculated. Geometry shaders have several additional instruction available dedicated to primitive function creation. There are slight differences but most of all there are different performances requirements. An efficient vertex shader unit isn't conceived like an efficient pixel shader processing unit.

    Unifying at the calculation unit level makes possible to maximise the use of each unit because if the vertex shader are simple and the pixel shader complicated, the first one won't be used at 100% of their possibilities. The counterpart is a complex management system. Let's admit that this management system represents more transistors than adding dedicated units to vertex shaders additionally to the ones that are supposed unified but restricted to pixels shaders. Or that x unified units and their management system represent more transistors than x pixel shaders units and y vertex shader units. What would be the interest of unification? This is the question that graphic processors manufacturers must have asked themselves after looking at their architecture and taking in account the addition of geometry shaders. Sometime the unification will be interesting and sometime it won't but for the final user this choice won't make any differences.

    << Previous day's news
    February 20, 2006
    Next day of news >>
    February 22, 2006

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