Announcement

Collapse
No announcement yet.

AMD "Hawaii" Open-Source GPU Acceleration Still Not Working Right

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

  • #41
    Originally posted by _SXX_ View Post
    Unfortunately this theory doesn't work for FGLRX on Linux. Most of games that have problems with FGLRX does work well with both Nvidia proprietary and Intel / AMD / Nvidia open source drivers.

    Some developers just recommend to use open source drivers instead of Catalyst.
    Unfortunately it works on fglrx as well. Mesa works because it has to mimic nvidia (read my post carefully with an example of Mesa's commit). Also I saw a couple of bug reports in fglrx's bugzilla where the problem was not in fglrx but in the software not properly following specs. But, of course, there are some troubles in fglrx but some of them are not the fault of AMD.

    Comment


    • #42
      "Release the docs and there will be drivers" worked much better when the hardware was much simpler. Nowadays, it helps a lot to have a driver team that knows not only how to read specs, but how to work together and what they can expect of each other.

      Drivers don't write themselves, teams of people write them. The bigger, smarter, better equipped, and better coordinated the team, the faster and better the results can be.

      Comment


      • #43
        Originally posted by justmy2cents View Post
        since i don't have ultimate preference, i can say bs on that.

        in my case,
        - if it is server, Intel. no question here.
        - if it is normal desktop, Amd (or maybe Intel when i want to cut as much hw and price as i can) no question here. OSS drivers are so nice on desktop
        - if i need working 3D, be it gaming or coding, NVidia. Amd will come in question when OSS hits 4.4 which i need, catalyst... never again after the troubles i had with it until i changed to NVidia

        for each purpose, there is a vendor and not one vendor is best in 2 departments. at least until Mesa hits 4.4, where i would consider spending more money for same performance and no blob. even going card range up to diminish Mesa performance lack would be worth it
        I'm talking GPU performance per dollar between Intel and AMD. Intel's iGPUs at the same price point as AMD's are markedly slower, to match or beat the AMD iGPUs you have to buy a much more expensive chip from Intel.

        Does Intel have the better raw single core CPU performance? Yes. Are their iGPUs any good at the same price as AMDs? Not at all. If you want a decent performing Intel iGPU you have to buy one with eDram, which jacks the price up significantly.

        We build custom comps at the shop I work at, we got with AMD APUs for iGPU based builds, because for the vast majority of home users the CPU has been "fast enough" for a long time now, but if that user tries to do anything that is GPU heavy they will be worse served with the Intel kit then the AMD kit.

        Comment


        • #44
          Originally posted by A Laggy Grunt View Post
          "Release the docs and there will be drivers" worked much better when the hardware was much simpler. Nowadays, it helps a lot to have a driver team that knows not only how to read specs, but how to work together and what they can expect of each other.

          Drivers don't write themselves, teams of people write them. The bigger, smarter, better equipped, and better coordinated the team, the faster and better the results can be.
          Theres a bunch of community devs, most of them got hired on.

          The Nouveau team has zero support at all from Nvidia and it shows because all of what they do has only come out of reverse engineering and as such can never be as good as the OSS Radeon drivers no matter how many dev hours they throw at it.

          Comment


          • #45
            a fully functional OpenCL compute stack
            It's going to be better to wait until they fixed the compiler for large kernels before calling this FULLY FUNCTIONAL.

            7950 here, with catalyst i have random crashes here and there but i give up on 14.4 drivers so i'm not sure if it's better now, foss drivers works great but lacks performance and features maybe next year they'll perform better than the proprietary driver cause they're improving faster than cataclysm.

            Also i hope amd managed to push the Opencl fix this month cause we've waited long enough to get a fully functional OpenCL compute stack.

            Comment


            • #46
              Originally posted by Kivada View Post
              Theres a bunch of community devs, most of them got hired on.

              The Nouveau team has zero support at all from Nvidia and it shows because all of what they do has only come out of reverse engineering and as such can never be as good as the OSS Radeon drivers no matter how many dev hours they throw at it.
              Seems like reverse engineering knowledge is much needed in this case , i mean to make this HAWAII chip respond . Really, when devs knows how to program it but it refuse to work and they have all information available then... how to figured out that chip other then trying small reing of fglrx .

              Comment


              • #47
                Sea Islands: High-end no accelleration support, mid-range almost perfect support.

                My R7 260X works perfect with both Catalyst and open source stack (oibaf PPA, Linux 3.15).
                Almost perfect. Runs cool, fast and stable. There isn't a single Steam game it didn't play with decent performance using OSS drivers.

                Your GPU isn't from Hawaii series.
                Hawaii is R9 290X and R9 290 only.
                R9 290X and R9 290 are high-end Sea Islands chips, R7 260 and R7260X are mid range... which are very related.

                Comment


                • #48
                  Originally posted by Rakot View Post
                  Unfortunately it works on fglrx as well. Mesa works because it has to mimic nvidia (read my post carefully with an example of Mesa's commit). Also I saw a couple of bug reports in fglrx's bugzilla where the problem was not in fglrx but in the software not properly following specs. But, of course, there are some troubles in fglrx but some of them are not the fault of AMD.
                  Unigine it's just one example which is pretty small and surprise: it's works with FGLRX which "follows specs more closely" and in same time don't work on Intel (but work on R600g). I seen many game releases on Steam and Catalyst have a more compatibility problems compared to Mesa/Nvidia. Not so long ago "good" fglrx just crashed my X server every time I tried to run Chrome and I pretty sure it's not because Chrome didn't follow some spec.

                  It's just funny to see AMD fanboys in this topic: drivers will work in future, more features, Nvidia kernel crashes, fglrx not that bad, never have a problems. In same time this topic isn't about Nvidia, but about the fact: in few months it's will be year after Hawaii GPUs released and GPUs support it's still broken in open source drivers.

                  I personally stuck with my HD6950 1GB (and 1GB it's really really low in 2014) and I want new faster GPU with more VRAM and open source drivers. I don't want to waste money on "slightly better" because only alternative AMD provide it's rebranded 7XXX. Seriously I can deal with current R600g/RadeonSI limitations, but I can't buy broken GPU and I'm not going to use Catalyst because it's will never work well for most recent kernels (I need them for KVM/QEMU).

                  Comment


                  • #49
                    Originally posted by Modu View Post
                    R9 290X and R9 290 are high-end Sea Islands chips, R7 260 and R7260X are mid range... which are very related.
                    No they are not.
                    All R series GPU except R9 290(X) it's basically rebranded HD7XXX with slightly different clocks/voltages/etc.

                    Comment


                    • #50
                      Originally posted by dungeon View Post
                      Seems like reverse engineering knowledge is much needed in this case , i mean to make this HAWAII chip respond . Really, when devs knows how to program it but it refuse to work and they have all information available then... how to figured out that chip other then trying small reing of fglrx .
                      No need, the docs are already there, there aren't that many Linux users using high end GPUs to put it on the high priority list. Give it another year, Star Citizen, Metro Complete Pack, Civilization: Beyond Earth, Witcher 3 and whatever else gets announced will provide a need for a focus on the high end GPUs sold in the hundreds of thousands as well as the low and mid range that have sold in the millions.

                      Comment

                      Working...
                      X