Announcement

Collapse
No announcement yet.

How Old ATI GPUs Can Be Faster On Open Drivers

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

  • How Old ATI GPUs Can Be Faster On Open Drivers

    Phoronix: How Old ATI GPUs Can Be Faster On Open Drivers

    A few days ago when publishing the results of benchmarking a lot of graphics cards on their Gallium3D drivers (about a dozen graphics cards) this left a number of people surprised. A number of these results from the open-source Gallium3D drivers illustrated the older graphics processors as being much faster than the newer hardware, even though the newer hardware is far superior to the vintage products. This shouldn't have been a surprise if you stay up-to-date with the Linux graphics news on Phoronix, but it comes down to features found in the older Gallium3D drivers not yet implemented in the newer open-source drivers...

    http://www.phoronix.com/vr.php?view=OTA5OQ

  • #2
    Why are the open source drivers unable to implement S3TC? As long as its use is restricted to hardware where the manufacturer paid the licensing fees, there should not be a problem.

    Comment


    • #3
      Hmmmmmm, S3TC....

      It's the only thing I wait for to switch on r600g open drivers (for EVE-Online on Wine)

      Comment


      • #4
        Additionally to the question of Shining Arcade:

        Will AMD ever bind selling of new cards to opensource driver development?
        This seems as extremely critical to me now, as kernel devs claim themself that hardware gets more complex with each release.
        The opensource driver development vector should at least keep the pace with new hardware development vector.
        Otherwise all opensource community ends in situation of perfect opensource driver for the card that is not sold, or even not compatible with modern hardware anymore.
        There is no point in such software and the whole effort can be considered as wasted then.

        Comment


        • #5
          Originally posted by crazycheese View Post
          Additionally to the question of Shining Arcade:

          Will AMD ever bind selling of new cards to opensource driver development? This seems as extremely critical to me now, as kernel devs claim themself that hardware gets more complex with each release. The opensource driver development vector should at least keep the pace with new hardware development vector. Otherwise all opensource community ends in situation of perfect opensource driver for the card that is not sold, or even not compatible with modern hardware anymore. There is no point in such software and the whole effort can be considered as wasted then.
          Crazycheese, same answer as the last few times you asked...

          If there was a reliable and reasonably accurate way to track opensource driver usage (one which was reasonably immune to being gamed or scripted) then I'm sure that information would get considered in our development plans at some point. I haven't heard of or seen any approaches which meet those criteria yet. Self-selecting mechanisms such as response forms aren't even close. Let's say we get 23,654 responses all saying "I only use open source drivers" - 100% of users, right ? Now, let's look at that in the context of say 100M units sold per year - now we're down to 0.025% market share, which means we have too many developers working on Linux already.

          That said, our commitment is to provide documentation and/or sample code and developer support to let the driver dev community build good open source drivers without having to reverse-engineer the hardware, not to develop those drivers ourselves, and we have been doing that for a few years now, so I'm not sure why you feel it is so important for us to provide additional developers.

          It seems to me that the first thing you are asking for is a wholesale change in strategy, maybe the "give up on the workstation market and focus our efforts on the consumer Linux client business so we can justify diverting more resources to open source drivers" approach ?

          Comment


          • #6
            Originally posted by Shining Arcanine View Post
            Why are the open source drivers unable to implement S3TC? As long as its use is restricted to hardware where the manufacturer paid the licensing fees, there should not be a problem.
            I vaguely remember one of the devs saying something about that. I think it was something like this: the hardware S3TC support doesn't necessarily work for all operations allowed by the spec (e.g. all combinations of texture formats and textures/pbuffers located in system vs. GPU memory), so S3TC support would require a separate software implementation as a fallback, and then Mesa is basically shipping its own implementation rather than just supporting the hardware implementation, which complicates things.

            Comment


            • #7
              GPU openhardware or at least openspec

              We need a GPU openhardware to evolve as fast and well as opensoftware does.
              At least Open specifications.
              The first company doing it will earn a lot of money, even if they also have propietary products.

              Comment


              • #8
                Typo:
                Mark ends with
                Hello there Mark

                Comment


                • #9
                  Originally posted by mitcoes View Post
                  We need a GPU openhardware to evolve as fast and well as opensoftware does.
                  At least Open specifications.
                  The first company doing it will earn a lot of money, even if they also have propietary products.
                  openhardware doesn't matter that much (at least for the end user)

                  What would seem like a "feasible" thing to do is, for a manufacturer, to move its driver development to G3D on all operating systems (it is portable afaik). But i don't thing none of the big 3 (nv, ati, intel) are going to do it (or can ) .

                  Comment


                  • #10
                    Just woke up, so forgive thinkos, please.

                    VMWare already uses Gallium for their drivers on all platforms. At least one incarnation of the Poulsbo driver uses Gallium on Windows and Linux. If the appropriate kernel piece and winsys were written, all nouveau drivers, r300g, and r600g, could be ported to Windows. Nobody cares about Windows enough to make it happen.

                    Stop talking about S3TC. We already have done as much as we can.

                    Also, I'm going to sum up the article as: Nobody pays the community developers to write code that makes GPUs more efficient. I'm sure this'll start a flame war, but whatever.

                    ~ C.

                    Comment


                    • #11
                      Originally posted by bridgman View Post
                      Crazycheese, same answer as the last few times you asked...
                      Yes, I understand you put a lot of resources into workstation segment. But it is not related to desktop at all - workstation cards are too expensive, the software using this cards is barely good or affordable for any consumer. What I mean is desktop market.

                      Right now we have:
                      Intel with more than working HTPC/Office market, opensource.
                      Nvidia with more than working Gaming/GPGPU market, HTPC to some extent via ION, closed source.
                      Additionally there is noveau hack, backing up some very basic server/office usage.

                      And AMD, that provides closed source drivers that are inferior to nvidia in Gaming/GPGPU/HTPC:
                      - no video accel
                      - much much shorter hardware support window
                      - longer reaction to new hardware
                      - much less kernel/xorg version support
                      - more issues on desktop
                      - GPGPU is prefered on Nvidia (asked at overclockers.com)
                      The only frame for AMD in this segment is multi-monitor office configurations or 3D software solutions that must be developed together with AMD.

                      Opensource drivers, that are inferior to Intel in HTPC/Office:
                      - Intel CPU includes GPU and is also opensource
                      - AMD performance cards run inefficient with opensource drivers
                      - Intel CPUs/GPUs run more energy efficient than AMD currently
                      The only frame for AMD in this segment are laptops with basic graphics(and there aren't many due to less efficient energy design; which should hopefully corrected with nextgen AMD APUs) or desktops with IGP(no factors affect AMD or Intel selection here).

                      Correct me, if Im wrong please, but not being able to use 4 year old card and its capabilities - the reason it was purchased, is anything but acceptable.

                      Looking at overall situation, you are not interested in improving anything related to non-workstation segment, but you do support hardware specifications, from which only corporate entities are able to develop solutions for your cards that are acceptable to THEM. Redhat and Novell want basic support (for servers I guess) and their crew does it finely(no more, no less).

                      So there is zero point to any entity not affiliated with Redhat or Novell to purchase your hardware to use on Linux.

                      As you mentioned, lets take 24k sold cards in simulated calculation, assume that they were registered via manual or kernel-based SN reporting and this provided mapping for:
                      - card type
                      - estimated purchase time graph
                      - overall variability (20k highend cards/4k IGP - bring more calculated profit than 20k IGP/4k highend) factor
                      Everything else is not needed and does provide anonymity.

                      The missing components are:
                      - real-world feedback in form of quateryear financial report & user reactions on internet. And
                      - task prioritization - the customer should be able to choose the GPU software/driver infrastructure part he wishes/prioritizes AMD to work on next. This feedback can be done at time of manual registration.

                      Another point is to provide support window for legacy hardware - users of such hardware should be able to fix it themselves and upload patches to AMD; or to pay AMD additionally to fix the issues. Both mechanisms should run as transparent and understandable to the users as possible.

                      The very rough calculated amounts may be 12k cards for 300$ resulting in 40$ for driver team each card and 12k cards for 50$ with 8$ for driver team. The segment includes only linux or opensource segment.12k*40 + 12k*8= ~600,000$ for projected period.

                      This does not include dynamics which MAY happen depending on company steps/intensity. The hype of having opensource for-ever-supported hardware may shift the desktop market segment, claiming market shares from competition. Intel is not providing performance hardware, Nvidia only provides closed source drivers. Both of them do not provide the described mechanism. This may boost the 0,6 Mln a bit further.

                      The weak point of the calculation is maybe absence of gaming (to some degree compensated due to Wine, Steam via Wine, planned Desura client, OpenCL, BOINC), but there is good tendency. DRM market is not affected as most people using linux are using it for reason opposing DRM itself and being treated as conscious people.

                      Patented technologies are not problem, unless they are closed source(look at Google and VP8 - copyleft patent is realistic). These way they could be implemented as optional attachable modules and may be replaced over time. Protecting IP maybe possible this way as well - impossiblity to close down the implementation for what AMD holds a patent. If all legal stuff is cleared that way, the only thing left will be work and money for that work from consumers.

                      Technologies requiring whole path shutdown are conflicting with opensource in their nature. If AMD supports viruses, people will decide if to follow. Currently Nvidia supports viruses, except this (and Optimus case) it performs fairly well, which may be seen as good attack vector for AMD. And if AMD follows Nvidia strategy, as I mentioned, there will be no sense for people to prefer AMD except fanatism. 0,024% will join Nvidia.

                      And if AMD does not change the current approach, usable opensource drivers will actually disappear, (rather small) userbase shrink even further and people still won't be able to repay AMD for effort it does on opensource. AMD will look just as Ubuntu then - 99,9% talks, 0.1% input.

                      Comment


                      • #12
                        Originally posted by MostAwesomeDude View Post
                        VMWare already uses Gallium for their drivers on all platforms. At least one incarnation of the Poulsbo driver uses Gallium on Windows and Linux. If the appropriate kernel piece and winsys were written, all nouveau drivers, r300g, and r600g, could be ported to Windows. Nobody cares about Windows enough to make it happen.
                        the foos dev community couldn't care less about win but is there something that could stop a manufacturer from moving all driver development (and resources) on gallium on all 3 major desktop environments??? (ie licensing, cost)

                        Comment


                        • #13
                          bridgman: As above, the problem with ATI is not really its open-source strategy, but its colossal failure in the proprietary driver. I've had nothing but problems with it and it's always way behind X; people have to grab the patched Ubuntu-exclusive version on new Ubuntu releases to get support for current X servers basically every time a new X server comes out. It's just silly.

                          If I were a systems OEM, I would go nvidia without a second thought. I appreciate AMD's open-source strategy, but without additional support from AMD like actually devoting some developer resources or code, it's not an adequate replacement for a good proprietary driver. As such, until I am confident that AMD can produce a functional, reliable, and updated driver for *nix that is well-maintained, or until AMD can boost the OSS devs so that things can proceed reasonably, AMD is out of the picture, no matter how good the cards are.

                          nvidia even has things like VDPAU, made especially for Linux users. AMD half implements several broken standards and still doesn't really have good video acceleration.

                          The open-source drivers are rad, but they are just too far behind to be usable for any 3D purposes. If you don't want to worry about supporting Linux, give the open-source guys a boost by hiring 3-4 of them full time for 1-2 years to really get your support up to par there, and then maybe AMD will be worth considering. For now, the progress is just too slow for AMD to ever be a good choice; it takes 5-8 years for a GPU to be well-supported end-to-end in the open drivers.

                          Comment


                          • #14
                            Also, the argument about insufficient marketshare is wrong. AMD has to look at this as an investment, as the only Linux users that use AMD are those that don't know any better or don't have any other option. nvidia has been king of Linux forever and no one buys ATI cards for general use on a Linux workstation because they are so far behind. The effort nvidia pours into Linux should show you that there's some potential in this space if you can only establish yourselves as a reliable Linux vendor. In the meantime, I've been buying nvidia for years and telling everyone else to do so, too, as are almost all other *nix users. The marketshare to justify it doesn't exist yet and it won't exist until AMD invests in making the platform competitive and viable.

                            Comment


                            • #15
                              Originally posted by bridgman View Post
                              It seems to me that the first thing you are asking for is a wholesale change in strategy, maybe the "give up on the workstation market and focus our efforts on the consumer Linux client business so we can justify diverting more resources to open source drivers" approach ?
                              AMD miss one important point:

                              If the workstation customers only care about closed source drivers for me as a AMD card user is no single Point of argument why anyone should buy an amd card?

                              i read all over the forum there is no amd/ati card user with a point like "amd cloused source drivers beats nvidia one in all my stuff i do"

                              i only can read stuff like this: "i prefer amd because the catalyst works for me right now and the weakness of the catalyst is less important than having an NDA Free spec and a second fall-back opensource driver"

                              AMD think like a brain-death Zombie:
                              "99% catalyst users means we need to hire 99% devs for the catalyst"
                              because they don't get the point of the part of the catalyst users care about the not used opensource driver.

                              call it PR Public Relations but i don't believe you found any catalyst-Linux customer that don't care about the radeon.

                              in general "Public Relations" is very important and in the past AMD lose on that part of business.

                              Comment

                              Working...
                              X