Announcement

Collapse
No announcement yet.

Benchmarks Of AMD's Newest Gallium3D Driver

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

  • Anyway, I just hope that the rate of improvements will not slow down.

    Btw, for the last few days there are barely any commits to mesa master. Is there a specific reason for that?

    Comment


    • Originally posted by HokTar View Post
      Anyway, I just hope that the rate of improvements will not slow down.

      Btw, for the last few days there are barely any commits to mesa master. Is there a specific reason for that?
      Probably the US holiday.

      Comment


      • FWIW, r300g and M66 is very impressive

        Originally posted by pingufunkybeat View Post
        I believe that r300g passed the 50% mark.
        I have a T60p laptop with built-in M66 (RV535?) graphics, and I have just expunged Fedora 10/fglrx from it in favour of Fedora 14 and Mesa-git. And my laptop is suddenly fun to use again... If there's a 50% loss in performance somewhere by using Mesa, I haven't noticed it yet. Maybe the pleasure of simply being able to use Xv without crashing X has made me blind(*)?

        Conversely, since that alleged 50% performance was bought at the cost of having to use fglrx, (an experience I would liken to having AMD metaphorically beat the soles of my feet with a spiked mace) then as far as I'm concerned, AMD can keep it.

        And Fedora 10's version of Wine is now too old and buggy to run the newest WoW client anyway. An upgrade was simply unavoidable.

        [* Did that come out right? ]

        Comment


        • I never asked but why in the world users are running away from AMD closed driver to open source, when AMD DOES support the closed driver for linux on their latest cards?

          Why there are tons of issues and so many bugs that people are relieved to have the open source? How many AMD engineers are assigned for the AMD closed driver?

          Comment


          • Because running open source > running closed source.

            At least that's my reason.

            If there is an open source program which does what I need it to do, I will use it. I'd only use closed source if I really have no choice. Most of the time, I don't have to.

            Comment


            • Originally posted by Beiruty View Post
              I never asked but why in the world users are running away from AMD closed driver to open source, when AMD DOES support the closed driver for linux on their latest cards?

              Why there are tons of issues and so many bugs that people are relieved to have the open source? How many AMD engineers are assigned for the AMD closed driver?
              Last time i check AMD doesn't support average Linux user but only workstation which is a completely different world. No compiz, kwin, not unlikely to never see a web browser or other apps, completely turned toward supporting heavy CAD program, don't care about video or games.

              That's, at least the original target of the linux closed source driver. AMD believed it is good enough to also fit more wider audience, and i believe fglrx works pretty well for most of its user.

              Then of course there is the open source motivation, and i prefer running open source software that i can fix myself. I think open source is always a superior choice because of the freedom it gives. You can adapt the software to your need (at least if you have either the skillset or the money to pay for it).


              On previous topic, i meaned more like somethings accross the stack (but you don't rewrite that much in the end, you copy a lot and change things and rename because new name make it looks cool, especially if you use a french word )), but i can't really say for sure yet if it's what we need or not. We are very new at writing fast driver, we are also bit too naive and often forget the whole picture (driver is not alone there is a world outhere), at least i am feeling like that. I think we did uneducated design decision and we are slowly educating ourself. All i can say is that my personal aim is to be as close to fglrx on performance without going down to application specific optimizations.

              Comment


              • I'm glad to hear that. I hope r600g will reach a reasonable performance (say 70+% of fglrx) level soon.

                Keep up the good work!

                Comment


                • Originally posted by Beiruty View Post
                  I never asked but why in the world users are running away from AMD closed driver to open source, when AMD DOES support the closed driver for linux on their latest cards?

                  Why there are tons of issues and so many bugs that people are relieved to have the open source? How many AMD engineers are assigned for the AMD closed driver?
                  i just give you 3 exampels now


                  "deanjo: Absolutely a bug "should" be fixed in the apps source. Unfortunately that is not always the case. Video drivers are crammed full of app specific fixes. Look at ANY driver release from ANY vendor and you will see thousands of these examples where the bug was addressed at the driver level instead of at the app level."

                  Deanjo is a cloused source fanboy he likes nvidia so much ...

                  "movieman: LOL... I can tell you've never written hardware drivers. We often had to work around braindead programming in applications by detecting that said application was running and doing something other than what they asked for. For example, in one case we implemented a feature in hardware that the competition did in software, and an application was reconfiguring it hundreds of times a second which required barely any effort in software but caused a major stall in hardware; we had little choice but to detect that application and push it into a software implementation of the feature because what it was doing was simply insane.
                  Actually, it's worse than that in that we actually had to _put bugs into our drivers_ because some applications relied on bugs in the drivers they were testing with in order to function, so if our drivers didn't have the same bugs then their application would fall over.
                  Ah, the joys of proprietary closed-source software where no-one wants to take responsibility for fixing their bugs! "

                  "bridgman: A lot of the "driver bug fixes" are really workarounds for app bugs - apps tend not to get updated once they have released so the only practical solution for customers is to work around the problem in the driver. That's one of the reasons you need to refactor/rewrite driver components periodically -- they end up full of so many app workarounds that the code becomes very difficult to maintain. "

                  the 3 statemends orginaly postet here: http://www.phoronix.com/forums/showthread.php?t=27409

                  means opensource users just don't wana have crap into there drivers...


                  and no crap means a stable and secure system

                  Comment


                  • Originally posted by airlied View Post
                    Its about 100 developers working full time.

                    Dave.

                    So.... Where can I donate money that will staff developers to work on this?

                    Comment


                    • Originally posted by castlefox View Post
                      So.... Where can I donate money that will staff developers to work on this?
                      Bounty doesn't work because people need to be sure that they will still earn money in 6 month from now. I doubt any consultant would be interested to work on this. Maybe if some org like linux foundation hired someone on behalf of the community this could work but i doubt LF would hire anyone unless it sees significant donation (read well above 100k$) so i doubt anyone have this kind of money to donate for free driver and i doubt even as a all community would not reach such sum (just go look at what other project which are much more popular like blender succeeded to secure and divide by at least 10). Even if money showed up i don't think LF would do it, i think LF is more about getting company to support and hire engineer for linux/kernel than to build hire directly hacker.

                      Comment


                      • Originally posted by glisse View Post
                        Bounty doesn't work because people need to be sure that they will still earn money in 6 month from now. I doubt any consultant would be interested to work on this. Maybe if some org like linux foundation hired someone on behalf of the community this could work but i doubt LF would hire anyone unless it sees significant donation (read well above 100k$) so i doubt anyone have this kind of money to donate for free driver and i doubt even as a all community would not reach such sum (just go look at what other project which are much more popular like blender succeeded to secure and divide by at least 10). Even if money showed up i don't think LF would do it, i think LF is more about getting company to support and hire engineer for linux/kernel than to build hire directly hacker.
                        maybe a campaign from the linux foundation or a call from someone influential (read torvalds) might attract people

                        something like an open graphics foundation that will support driver development (and all graphics stack related projects) on all open OSes would be cool but i doubt its viable

                        Comment


                        • Originally posted by 89c51 View Post
                          maybe a campaign from the linux foundation or a call from someone influential (read torvalds) might attract people

                          something like an open graphics foundation that will support driver development (and all graphics stack related projects) on all open OSes would be cool but i doubt its viable
                          If we can gather several millions of dollars i proposing myself as CEO ;-)

                          Comment


                          • I guess, if all dirstros charged $1 per month or $10 per year and all the money went to dedicated low level linux infrastructure, there would be a lot of money to go around for critical developers.

                            Comment


                            • For several millions of dollars I'm sure glisse would finish the driver stack himself

                              Comment


                              • For that kind of money...

                                Originally posted by bridgman View Post
                                For several millions of dollars I'm sure glisse would finish the driver stack himself
                                I'd expect stringent delivery criteria. With eating/sleeping probably considered optional...

                                Comment

                                Working...
                                X