Announcement

Collapse
No announcement yet.

Windows 10 vs. Linux Radeon Software Performance, Including AMDGPU-PRO & RadeonSI

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

  • #71
    Patches are in the heads of everybody, those just waiting to materilse from there

    It is enough to look at a code, hard part might be trying to understand it and then you can play with it.

    Comment


    • #72
      BTW, there is no what to hate in proprietary driver at least i don't see it... it is only matter how you (are allowed or not to) distribute that code and nothing else.

      At the end of the day, people wrote code there, people wrote code here - no difference

      Comment


      • #73
        Originally posted by CapsAdmin View Post

        I didn't know this so thanks for pointing it out.

        I've looked at dirc and it seems like they are used to workaround illegal OpenGL code rather than optimize it. Most of it seems to mainly be GLSL fixes as the compiler on the AMD and NVidia drivers are a bit relaxed and don't follow the specification properly. Without these the shaders wouldn't compile and in turn the applications would crash. The workarounds are probably fixed by the developers by now but because of how closed source software is shipped and how games rely on it this can sometimes be difficult or take a very long time.

        In the case of Unigine there are a couple of workarounds related to illegal use of GLSL making the shaders not compile on mesa. I think these have been fixed in unigine by now.

        But this is far from replacing shaders and buffer formats though which can sometimes even degrade quality in order to improve performance. When this is done in benchmarks this not only becomes hacky but also very wrong.

        I tried to post this earlier with some more content and urls for citation but my post seemed to have disappeared for some reason
        I somehow missed that one

        That is wrong way of thinking about those replacements When they do such of thing it is mostly because of scalability, so that every product down from the slowest APU up to the Crossfire setups scale well.

        Second reason after *scalability* is *usability* you know, most important thing out there you replace to fix something so that user does not got a shit

        In opensource, i often have these comments... like who cares about openarena, lets break it 3 times who cares Or Trine series games all misrender, as someone somehow conclude that they break specs while blobs do it fine with no fixes or nothing, Unigine benchmarks too... so with all that "properness" usability approaching to Zero

        How many times Mesa breaked specs, i can't even count anymore ... but wait no one talking about that, that might be only parallel reality or something

        Comment


        • #74
          He, he and also specs are not only current specs As specs also has bugs too, so proper specs implementation is with all bugs in them supported

          Yup, you read it right with all bugs supported... as some software might depend on these bugs, but is not anymore maintained or so

          Comment


          • #75
            So I guess it's safe to assume you don't have the patches

            Comment


            • #76
              He, he i had patches always, when i am up to speed with all git

              But currently i am resting from that and look at things from distance

              Comment


              • #77
                Originally posted by dungeon View Post

                I somehow missed that one

                That is wrong way of thinking about those replacements When they do such of thing it is mostly because of scalability, so that every product down from the slowest APU up to the Crossfire setups scale well.

                Second reason after *scalability* is *usability* you know, most important thing out there you replace to fix something so that user does not got a shit

                In opensource, i often have these comments... like who cares about openarena, lets break it 3 times who cares Or Trine series games all misrender, as someone somehow conclude that they break specs while blobs do it fine with no fixes or nothing, Unigine benchmarks too... so with all that "properness" usability approaching to Zero

                How many times Mesa breaked specs, i can't even count anymore ... but wait no one talking about that, that might be only parallel reality or something
                Sorry but I'm not really sure what your response is here in relation to performance, which was my main point here.

                Tripple A game developers tend to work more closely with Nvidia and AMD to ensure things don't break and perform good. I'm speculating here but if Nvidia released a new driver that would break OpenArena I'm sure it's OpenArena that would have to change rather than Nvidia's driver. If you replaced OpenArena with Call Of Duty I'd imagine the situation would be different.

                Comment


                • #78
                  Triple-A game devs target gaming consoles, where perf targets are 30 or 60 fps - end of story about performance . Games are there sort of interactive movies, play, enjoy and that is about it.

                  But PCs are diversive so impossible to target properly, second but even more proper name for PC is actually Personal Customisation, where thanks to the more or less handy drivers you can use it like- console, setup and experiment as much as you wish or even ingore whole thing

                  My bug about openarena is not just openarena, it is example on breaking perf for probably about half games on steam

                  Comment


                  • #79
                    And really Unigine Valley picture quality looks better with blobs, anyone non ignorant and with good eyes condition can see it

                    Mesa for sure misses something there with all that "correctness"

                    Comment


                    • #80
                      Originally posted by duby229 View Post
                      I see you had to look all the way back to a pentuim classic to find one
                      so, your metric only work since ppro? good for you. btw, if you had any brain, you could see that shared cache of N per two cores is more than separate N/2 per core, because 1) cores could share some memory and 2) the one needing more cache will get more.

                      Comment

                      Working...
                      X