No announcement yet.

Rich Geldreich: A Bad Catalyst GL Driver Is Bad For Everyone

  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by stalkerg View Post
    Please drop Catalyst driver for Linux!!! And send all workers to r300, r600, radeonsi driver! Need petitions?
    There is no need for petition (r300 and better half of r600 cards are not even supported anymore by current Catalyst drivers ), also i expect they will drop Catalyst for r600 cards complitely somewhere next year .


    • #12
      Originally posted by brad0 View Post
      NVIDIA only providing binary drivers is bad for everyone too.


      • #13
        First, this guy is a troll.
        He said openly in a previous article that old OpenGL is flawed, and that he doesn't even consider new OpenGL to be a solution because the design focus of the new API is to make it to be used less often.
        So, and old thing is flawed, but let's not look at the new thing that fixes all problems because I don't like how it has to be used now.

        Now, the test.
        The test is meant for comparing different algorithms using older/newer APIs, not for comparing different hardware.
        It is meant to hint at developers how a future game/application must be programmed in OpenGL.
        The test shows for every driver that there are some code paths that are definitely faster than others.

        The problem in the article is that it has the results in a random order, like hinting to the viewer that what's compared is the hardware.
        The test has "problems" and "solutions" and they are presented in the article all mixed up. Like, first "UntexturedObjects", then "TexturedQuadsProblem", then back to "UntexturedObjects", etc.

        Also the impact of the driver overhead is exaggerated here. This is not comparing general performance of OpenGL across different vendors. It's called micro-benchmark for a reason.
        This is only a small part of what matters for performance. It means that AMD could be very well into the "optimized enough" level for real applications when considering the big picture.

        A small observation:
        "UntexturedObjects GLDrawLoop" is ~29% faster than "UntexturedObjects GLUniform" for NVidia, but ~206% faster for AMD.
        I guess this is just AMD not caring about optimizing old OpenGL and focusing more on the newer API.
        Last edited by DeiF; 17 June 2014, 05:32 AM. Reason: small observation


        • #14
          Originally posted by [Knuckles] View Post
          Others may be "working towards" or may be "spearheading stuff" but if you want decent OpenGL in 2014, nvidia is your only option.
          Thats a reason against Nvidia then, not for opengl. and with that its a reason against opengl. If I have the choice between a "open" but also primary only implemented in proprietary drivers api for exclusivly unfree games (show me good free games that use opengl 4.4) and a highend-gaming hardware-vendor-independent api (direct3d/gl) that is os-exclusive, I will stick to the last.

          For gaming then I always just use my windows pc for ever, I dont even care about steamos that much, because I dont see a big point into switching one propriatary os for another os which is also full of propriatery software even in the kernel with unfree modules for gaming.

          And even if amd would magicly have tomorow the same speed on all opengl levels than nvidia has, opengl just sucks, its slower (no real game ever proofed that it can deliver the same huge performance improvements that mantle did generate), its api is garbage everybody asked aobut it just hates it, so lets hope amd releases the specs more or less soon of mantle and lets get rid of that opengl garbage at least for gaming.

          And then yes maybe nvidia boykots then this api, but so can amd do with opengl, yes they will not completly ignore the standard or remove code that works already, but having huge performace differences between mantle and opengl nearly forces then engine/game developers to support both apis.

          Of course intel could be the problem here, because they hate mantle because it allows gamers to not only consider 500,- Euro Core i7?s as fast enough for games. So we will see. on the other side they dont produce gpus that allow u to play high-end games. so does their cpu-market-share matter that much for what game developers choose as their target apis?


          • #15
            Originally posted by dungeon View Post
            There is no need for petition (r300 and better half of r600 cards are not even supported anymore by current Catalyst drivers ), also i expect they will drop Catalyst for r600 cards complitely somewhere next year .
            I wrote about workers.


            • #16
              What all of you little ignorant posters from the outer space don't know or don't probably think is that the next OpenGL iteration will probably be what Mantle is now. AMD is a big supporter and developer in Khronos Group. Their driver tries to closely follow the standard (although with not much success) and has promoted a lot of ARB extensions.

              OpenGL 5 "Mantle profile" and that profile currently has some engines supporting it.


              • #17
                Originally posted by stalkerg View Post
                I wrote about workers.
                They say several times they will not drop Catalyst driver for Linux because they have enterprise customers who use it, the same goes with the opensource driver, etc... so workers will have what to do: like supporting next generation hardware, Mantle maybe, etc. But OK i guess they can manage some devs time for the opensource driver if help is wanted/needed .


                • #18
                  Graham Sellers (AMD's OpenGL guy) said this in a recent tweet (about the Phoronix article with the benchmark):

                  Originally posted by Graham Sellers
                  Point completely missed, I guess. This was never about vendor vs. vendor, but about technique vs. technique.
                  My lesson learned... "Don't tell developers how to optimize their applications unless you also beat NV."
                  Full conversation:


                  • #19
                    I wish GL detractors would also mention that the problem is multiplied by the orthogonal OpenCL support-- and I do mean right angles. It's the same driver and the different companies have different non-overlapping priorities, again.


                    • #20
                      Originally posted by dungeon View Post
                      Devs just needs reliable API which fast & reliable on most vendor implementation, common thing as they can target majority of people with no problem, etc... so they want just that to be easy and easy supportible for them, with not so much vendor diversity included .
                      DirectX: Window only api slower than Mantle and New Opengl methods, Dx12 would be faster but is not out yet.

                      Mantle: It's faster than DX but only works on WINDOWS using AMD GCN hardware, is not open and even the documentation is under nda.

                      Opengl: Open, works on almost everything (Operating systems and hardware from desktop to mobile devices) is as fast as Mantle on Nvidia cards but too slow on Intel and AMD hardware.

                      Metal: Not enough information yet...

                      And sadly the more "reliable" api for AMD, Intel and Nvidia hardware is DirectX.