Announcement

Collapse
No announcement yet.

AMD Kaveri's Open Radeon Performance Now Multiple Times Faster

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

  • #31
    Originally posted by zanny View Post
    You don't want to make an xorg.conf if there isn't one - modern mesa does an autogen using the contents of xorg.conf.d, just go in there and make a file called 99-radeon.conf with these contents:

    Code:
    Section "Device"
        Driver "radeon"
        Option "ColorTiling" "on"
        Option "ColorTiling2D" "on" 
    EndSection
    How does this work for people with multiple GPUs though, or does this just apply to all devices on the system that uses the radeon driver?

    Comment


    • #32
      Originally posted by benmoran View Post
      Because I have a mini-itx rig for my main desktop, and don't plan to buy a discreet GPU any time soon. It's mostly a space issue. I've got my eye on a new fm2+ motherboard anyway that has some nice features, but I'm waiting to get an APU+MB combo at the same time. I'll do that once the performance is better than what I have.
      If you already have a rig, though, why not just build a new pc and use the old one for something else or sell it? Better than having junk parts. My current rig is a mini itx Elite 130 with an i7 and I now have a 7870 in the mail.

      Comment


      • #33
        Originally posted by Spittie View Post
        Let's compare Catalyst:

        ...

        That's what I call impressive
        Good job AMD!
        Thanks. I don't see why Michael didn't run it against that bench in first place.

        Comment


        • #34
          Originally posted by Espionage724 View Post
          How does this work for people with multiple GPUs though, or does this just apply to all devices on the system that uses the radeon driver?
          You can specify the busid if you want to apply to a specific instance.

          Comment


          • #35
            Originally posted by Espionage724 View Post
            If I recall right, the Xorg.log (/var/log/Xorg.0.log on Ubuntu) should specify whether Color Tiling is on or off (just search for "tiling"). If you have multiple GPUs, it should mention tiling for each GPU.
            Looks like color tiling is already enabled for Radeon by default:

            [code]grayson@grayson-htpc:~$ cat /var/log/Xorg.0.log | grep Tiling
            [ 21.594] (II) RADEON(0): KMS Color Tiling: enabled
            [ 21.594] (II) RADEON(0): KMS Color Tiling 2D: enabled[/quote]

            Thanks.

            Comment


            • #36
              Tested on my desktop with a discrete 7770 and open source drivers:

              http://openbenchmarking.org/result/1...SO-1401267PL84

              I did try with fglrx, but my system locked up in the middle.

              EDIT:
              The tests I did see with fglrx were faster... but it was obviously less stable.

              Comment


              • #37
                Can anyone explain what ColorTiling actually is?

                Comment


                • #38
                  Originally posted by sarmad View Post
                  Can anyone explain what ColorTiling actually is?
                  Tiling is way of storing data in memory to better utilize caches and memory channels. Generally GPUs want data with 2 dimensional data locality rather than 1 dimensional (e.g., you usually only are about a small portion of the screen at any given time so it's nice to have all that data nearby in memory). Rather than storing data linearly in a big array, you might store data for 8x8 regions of the image in subsequent addresses. It improves cache locality and maximizes memory bandwidth.
                  Last edited by agd5f; 01-28-2014, 02:09 PM.

                  Comment


                  • #39
                    Originally posted by agd5f View Post
                    Tiling is way of storing data in memory to better utilize caches and memory channels. Generally GPUs want data with 2 dimensional data locality rather than 1 dimensional (e.g., you usually only are about a small portion of the screen at any given time so it's nice to have all that data nearby in memory). Rather than storing data linearly in a big array, you might store data for 8x8 regions of the image in subsequent addresses. It improves cache locality and maximizes memory bandwidth.
                    Thanks a lot for the explanation, and for the hard work on the ATI driver

                    Comment


                    • #40
                      Originally posted by schmidtbag View Post
                      I don't know exactly what XMP is, but I'm assuming that it overrides your other RAM options. It wouldn't surprise me if it prevents you from changing your FSB too.
                      If your mobo supports the ability to read the ram's XMP profile(something Nvidia cooked up around the time they launched their C2D SLI chipsets) the mobo will automagically set up the ram to what the sticker says the speed, CAS and voltages are supposed to be instead of defaulting to the JEDEC standard that a normal mobo or a stick or ram without a XMP profile would do.

                      Basically it's overclocking for dummies and you can over ride it with your own settings if the mobo supports it.

                      Comment

                      Working...
                      X