Announcement

Collapse
No announcement yet.

Texture Units not all being used?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Texture Units not all being used?

    Ok you have 4 pixel pipes with 4 texturing units each. I don't know much about how the chip hardware works but wouldn't this be better.

    You have a game that says uses 2 textures, so you send 2 texture to the chip, but split it up over all 4 pipes then spit it out to the screen vs just using on pipe to spit out 2 textures per clock.

    Any comments on this?

  • #2
    I think each texturing unit (there are four) can work on one pixel per pass, and that the individual texturing units cannot share the workload of texturing the same pixel. So, yeah, in a game that uses only two textures per pixel there is a waste, and the GF4 with higher clock speeds will outperform it.

    Note that I qualified the above with, "I think".

    -[Ch]amsalot

    Comment


    • #3
      the parhelia can use the extra TMUs for better filtering,
      example dualtexturing uses 2 TMUs and the other 2 TMUs is used for free trilinear filtering.
      This sig is a shameless atempt to make my post look bigger.

      Comment


      • #4
        I didn't mean shareing the work load of one pixel. But instead of having the card use one pipe to process one pixel per clock cycle it uses 4 pipes to process 4 pixels per clock cycle. 4 at a time is a hell of a lot faster than 1 at a time.

        You have 4 pipes, might as well make use of all of them at the same time for best performance. Not sure if Radeon does this or not as well.

        Comment


        • #5
          Galvin,

          It does that already. It processes four pixels at a time.

          -[Ch]ams

          Comment


          • #6
            Where does it say that?
            Still you have 4 pipes with 4 texturing units so technically you should be able to pass 4x4 textures per pass 16 textures per pass. So if you take say like 2 textures split it across 16 that's a big performance jump, sorta like raid when you take a 1gig file and write it to 4 HD at the same time you get awesome bandwith and speed.

            Comment


            • #7
              Like I said before, you're trying to lay two textures on eight pixels per pass and it can't be done. That would require an eight pixel pipeline.

              Yeah, it's a waste in the case of apps that only apply two textures and no texture filtering but there's only so much room on the die.

              -[Ch]ams

              Comment


              • #8
                I have the feeling and I am not sure if I am right or wrong.

                This is not truely realistic but may be better to understand what I am getting at.

                What I am getting at is that i think a lot of GPU power is not being used on the matrox card. I still haven't cancled my preorder.

                If you have 4 holes in a tub to drain the water, but you plug 2 of the 4, the water will take 2x longer to drain.

                Am I making sense, these games that the matrox was tested on, just how many texture units do they need to render one frame per second, if it's 8, then why aren't the other 8 being used?

                Comment


                • #9
                  Sure the Parhelia has an advantage in more texture units per pipe, strange that the Serious Sam benchmarks didn't show a better result as it can take advantage of them.
                  "That's right fool! Now I'm a flying talking donkey!"

                  P4 2.66, 512 mb PC2700, ATI Radeon 9000, Seagate Barracude IV 80 gb, Acer Al 732 17" TFT

                  Comment


                  • #10
                    What I am getting at is that i think a lot of GPU power is not being used on the matrox card.
                    Bingo! And my guess is by the time it's full potential is used, it will no longer be competitive.

                    -[Ch]ams

                    Comment

                    Working...
                    X