Announcement

Collapse
No announcement yet.

Not all Unhappy campers on ati

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

  • #51
    Originally posted by Kano View Post
    Install Kanotix, then do a dist-upgrade (otherwise mesa will be replaced as soon as you do) and follow those steps:



    In theory you dont need to upgrade libdrm for radeon only, just when you want to compile mesa completely (like my script does as i build it not only for one driver as i like to replace cards) then you need 2.4.24 or newer. When you use 32 bit and you are able to install libdrm in the correct version and the other build-deps you could try the scripts even with another distro, but it is only tested with Kanotix (Debian squeeze+ should work too, maybe change kdm to gdm at the end of the scripts). I am sure that you will have got full mesa speed only when you compile it on your own compared to a repo.
    Thanks for this.
    Slackware uses .tgz or .txz packages, which essentially are a tar.gz from / with no dependency tracking. I have the readme you linked me to, and I can raid slackware-current as needed, so I'll get going from there. Kernel is now 2.6.38.4 - I should have mentioned that. I'll work carefully through your scripts and apply the sense of them in slackware. The laptop is 64 bit, and I don't know the current state of compat-32 packages. But I can compile 32 bit & 64 bit, and install this way
    make DESTDIR=/foo install (32 bit)
    make DESTDIR=/foo install (64 bit)
    so the 64 bit overwrites 32 bit anywhere it matters :-D. Then
    cd /foo
    makepkg ../full_64_32.tgz
    which gives me an (un)installable package which can be added.

    and install that. I had not thought of updating mplayer, but of course it makes sense as soon as you say it. I will compile mesa and xf86-video-ati in any case.

    Comment


    • #52
      Originally posted by Kano View Post
      Since when does nouveau NOT support fermi? Maybe you live in a parallel world Try git for kernel, ddx + mesa gallium.
      Fermi in general, but does nouveau support GTX 560?

      The binary drivers just got official support for it a few days ago.

      Comment


      • #53
        I don't think there is a huge diff compared to 460/470, why don't try it?

        Comment


        • #54
          Originally posted by pingufunkybeat View Post
          Today I tried to install the nvidia drivers for a new GTX 560 board at work.

          I don't know why people put up with that crap. Nothing worked out of the box, but that's understandable for such a new card. Installed the drivers form Debian repositories, no graphics. Had to debug from the command line, to find out that the drivers don't support it. Pulled the latest stable drivers from "unstable" (adding unstable tree to sources.list, then removing it immediately after allez!), same thing. Then I find out that I need the 270 beta drivers, which are not even in Debian experimental (at least not in the nvidia-recommended version).

          Oh, and the Debian nvidia installation guide has about 300 pages on all the tricks and pitfalls. The nouveau section is a one-liner. It's hilarious, really: http://wiki.debian.org/NvidiaGraphicsDrivers

          If anyone starts another rant about how user-friendly the nvidia blobs are, I'll shit on him. I spent half a friggin day, and I still only have a framebuffer. I'll either have to wait for debian unstable to package this stuff (which will take weeks) -- this is the way Linux is supposed to work -- or I download the blob from the nvidia webpage (which has problems with both firefox and webkit) and run unknown binaries as root, thus potentially compromising the system. I'd take installing from git by hand any day.

          If nouveau were an option with this board, I'd ditch the blob to hell. All because of some old CUDA code we have, which I /might/ have to touch at some point.
          Time to convert that CUDA code to OpenCL code then.

          Comment


          • #55
            Originally posted by plonoma View Post
            Time to convert that CUDA code to OpenCL code then.
            And that helps him how? The only openCL support is with the blobs any ways (unless you want to run CPU only negating any GPGPU advantage).

            Comment


            • #56
              Originally posted by deanjo View Post
              And that helps him how? The only openCL support is with the blobs any ways (unless you want to run CPU only negating any GPGPU advantage).
              speaking of OpenCL how much work does Zack Rusin's clover needs to be usable???

              Comment


              • #57
                Originally posted by 89c51 View Post
                speaking of OpenCL how much work does Zack Rusin's clover needs to be usable???
                A fair amount I would guess seeing that there hasn't even been a commit since november.

                Comment


                • #58
                  If you have got nforce2/3 chipset then you can forget agp + ati because nv has got no vista/7 agp driver - at max xp possible. For sis/via you can get an updated agp driver however... nv cards work in agp slot but with pci speed only without driver - ati cards do not initialize without.

                  Comment


                  • #59
                    Originally posted by Kano View Post
                    If you have got nforce2/3 chipset then you can forget agp + ati because nv has got no vista/7 agp driver - at max xp possible.
                    Gart drivers from XP work in win vista and Win 7 Kano. You have to disable driver signing however.

                    Comment


                    • #60
                      Then show me an ati agp card working with it.

                      Comment

                      Working...
                      X