Announcement

Collapse
No announcement yet.

AMD Catalyst 11.8 Linux Driver Released

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

  • #41
    I used opensource driver on 4770 and turning compiz off resulted in 20% performance increase
    My fault I should have mentioned, the performance increase is with fglrx only.
    Any time I play Doom 3 with Nvidia proprietary drivers with compiz on the performance falls on my laptop.


    Yeah, AMD does not support open driver too. Don?t get me wrong, they released documentation and have at least 3 people working on drivers full day, yet official driver is closed source catalyst.
    This doesn't make sense to me. They DO support open-source drivers by releasing docs and having 3 devs working on them, does nvidia support their open source drivers in ANY way? No. And open-sourcing a small config software that works only for their closed source driver seems to me like saying: "ok, can't bother making a decent config interface for our binary drivers, let's make it open source so that some else does it".

    Yes. Pretty much correct. But small details:
    - nvidia does not hinder noveau, amd controls(less or more, bad or good) radeon
    - noveau is firmware-free as it can generate the firmware itself, unlike radeon(xf86-video-ati and corresponding kernel driver)
    - noveau, in features and performance is same to nvidia, as radeon to fglrx. Well, due to AMD not turning away, radeon is 150% of noveau, yet not even close to nvidia blob or fglrx. In terms of power management noveau is better..
    - noveau is clean-room reverse engineered driver, but radeon is not link
    - amd is completely satisfied with how, and how fast radeon driver is developed and financed; and how it lives together with fglrx. Well, I?m not. But everyone is free to choose,no?
    - if you buy discrete card for 3d, neither of the drivers is usable, both opensource are interesting mostly for legacy unsupported cards(much less impacts nvidia)
    Seems to me like saying: radeon was better when it was being reverse engineered by people that had nothing to do with Ati, when Ati had no control over it and did not hinder it, and it was exactly what nouveau is now. So, it's like saying that all the open-sourced docs and the devs working on it and the whole ati open-source strategy actually worsened the state of the driver. Sorry, but I have no complaints over Ati open-source strategy if I compare to Nvidia's non-existent one.

    Comment


    • #42
      Originally posted by vervelover View Post
      My fault I should have mentioned, the performance increase is with fglrx only.
      Any time I play Doom 3 with Nvidia proprietary drivers with compiz on the performance falls on my laptop.
      Mate, just use the driver and hardware you see fit. Right now nvidia is ok for me, unlike amd. As I sayd, I have no performance hits with or without compositing and no screen tearing in any videos (vdpau, smplayer). Contrary to your statement, I had composite slow down 3d performance on opensource drivers with hd4770(thats equivalent to ~9800gt in raw opengl).


      Originally posted by vervelover View Post
      This doesn't make sense to me. They DO support open-source drivers by releasing docs and having 3 devs working on them, does nvidia support their open source drivers in ANY way? No. And open-sourcing a small config software that works only for their closed source driver seems to me like saying: "ok, can't bother making a decent config interface for our binary drivers, let's make it open source so that some else does it".
      Nvidia has no official opensource drivers, at all. They have contributed suggestions and patches for opensource stack and they provide many things in opensource.
      Amd does not have official opensource drivers (their only official driver is catalyst), but has opensource strategy - where they release the specs and fund some developers to collaboratively improve opensource drivers.
      To help you understand me, I ask you to look at both amd and nvidia hardware - which is through and through proprietary. This is the reason opensource radeon driver cannot work without microcode blobs and there is Mr. Bridgeman to prevent something not acceptable for AMD to come into opensource driver(yet it is used in their proprietary driver). Nvidia decided just to close the driver completely and work on it in-house instead.
      Which is better depends on your demands and amd opensource driver is inferior to amd official closed source driver when you ask for performance and features. So whats your problem? Comparing two different evils?

      Originally posted by vervelover View Post
      Seems to me like saying: radeon was better when it was being reverse engineered by people that had nothing to do with Ati, when Ati had no control over it and did not hinder it, and it was exactly what nouveau is now. So, it's like saying that all the open-sourced docs and the devs working on it and the whole ati open-source strategy actually worsened the state of the driver.
      Its like saying, it limited the driver possibilities and put it on permanent position number 2. Yes, it improved driver development speed and being officially supported is nice, but, at same time, it cut the headroom and independence. This "headroom", of course, has no use any way if nobody can reach it due to closed specs harming development rate and possible cease-and-desist due to patent infringements.

      Originally posted by vervelover View Post
      Sorry, but I have no complaints over Ati open-source strategy if I compare to Nvidia's non-existent one.
      Yes, I would agree with you if I had a built-in IGP graphics and be OK with OpenGL2, unimportant power management and fps.
      But right now, no opensource driver is usable for me. Sure, when radeon driver improves past current state, it will change, but I?m talking about "now".

      Comment


      • #43
        Contrary to your statement, I had composite slow down 3d performance on opensource drivers with hd4770
        As I said, I'm talking about fglrx having a perf boost with compiz enabled, not the opensource drivers.

        So whats your problem? Comparing two different evils?
        No problem at all, I just think Ati is doing a better job than Nvidia with their opensource stack, also because I don't think Ati can ever

        cut the headroom and independence
        of their radeon driver. Since it's open-source, if anyone has a problem with the way it's being developed, it can still be forked, while Noveau will never see the benefits of released specs, and can never have an alternative to reverse engeneering. You said everyone is free to choose, I'd say everyone except the Nouveau devs.

        Comment


        • #44
          Originally posted by vervelover View Post
          As I said, I'm talking about fglrx having a perf boost with compiz enabled, not the opensource drivers.
          I?m definitely not fglrx fan, since version 8.6(if I remember correctly) crashed the system when switching terminal windows... And opensource radeon is currently not my choice.

          Originally posted by vervelover View Post
          No problem at all, I just think Ati is doing a better job than Nvidia with their opensource stack, also because I don't think Ati can ever
          Sure.

          Originally posted by vervelover View Post
          of their radeon driver. Since it's open-source, if anyone has a problem with the way it's being developed, it can still be forked, while Noveau will never see the benefits of released specs, and can never have an alternative to reverse engeneering. You said everyone is free to choose, I'd say everyone except the Nouveau devs.
          You missed the point that opensource amd drivers need firmware blobs in order to operate, so noveau is more clean opensource than radeon.
          Also, I do not think AMD will allow features in radeon that will compromise the need for fglrx. Like efficient power management and performance improvements that would make radeon outperform fglrx in 3D and make it useless (but its wild guess).
          Every time I rised an idea to make radeon opensource driver officially supported part of amd and to put its development on monetary basis (cause the main income for driver development comes from card selling, yes I know amd produces only chips and PCB design - they still need them sold). Every time amd expressed its ok with current state, current development rate and current performance of radeon compared to fglrx.
          Well, if fglrx is their best bet, then I?m better with nvidia. But this is only due to my usage scenario. If you find opensource radeon sufficient for you, than I congratulate you.

          Comment


          • #45
            Cool, so many things I disagree with in a single post

            Originally posted by crazycheese View Post
            You missed the point that opensource amd drivers need firmware blobs in order to operate, so noveau is more clean opensource than radeon.
            That's not a radeon vs nouveau thing as much as a hardware thing -- the fact that AMD GPUs keep pretty much all of their microcode in RAM while other vendors build most of it into the chip. If our microcode was burned into the chip where you couldn't see it would we somehow be more "open source friendly" ?

            I see the same thing with CPUs -- processors which don't have upgradable microcode are somehow felt to be "more open source friendly" than processors where you can update the microcode after purchase. This whole "if we have to see the microcode it's evil but if it's out of sight it's OK" perspective baffles me -- it's like the belief that building a wall around a resort so you can't see the poor people outside somehow makes the problem go away.

            Originally posted by crazycheese View Post
            Also, I do not think AMD will allow features in radeon that will compromise the need for fglrx. Like efficient power management and performance improvements that would make radeon outperform fglrx in 3D and make it useless (but its wild guess).
            Why would you even think this ? The only thing we have said w.r.t. performance is that BASED ON THE COMPLEXITY OF THE HARDWARE, THE SIZE OF THE COMMUNITY, AND COMMUNITY DEVELOPER STATEMENTS ABOUT THE TYPES OF OPTIMIZATIONS THEY WOULD AND WOULD NOT DO we expected the open drivers to run with approximately 60-70% of the fglrx driver's performance. I think it's fair to say that we would be ecstatic if the open source driver eliminated the need for fglrx in markets like workstation, but it seems unlikely given the magnitude of the task. When we don't release info it's because of IP or DRM reasons, certainly not a desire to hold back the open source driver.

            Originally posted by crazycheese View Post
            Every time I rised an idea to make radeon opensource driver officially supported part of amd and to put its development on monetary basis (cause the main income for driver development comes from card selling, yes I know amd produces only chips and PCB design - they still need them sold). Every time amd expressed its ok with current state, current development rate and current performance of radeon compared to fglrx.
            This is totally untrue. I said a couple of things, none of them even remotely close to your statement :

            1. if we put a mechanism in place to allocate development funds between proprietary and open source drivers based on the money we made from the two markets there was a serious chance that the outcome would be *less* funding for open source, not *more*

            2. even if the numbers worked out in favour of shifting relatively more funding to the open drivers at some point in the future I don't think we could afford to substantially reduce fglrx funding relative to the requirements of the markets it serves, so this really boils down to a "give me more money" request

            If we were happy with current state why would we have been working for years to bring open source driver support earlier in time and closer to hardware launch ? Why would we be working to release programming info in more areas ?

            If we were happy with current development rate why would we be hiring more developers ?

            How can you possibly twist a performance estimate based on the size of the development team relative to the domain complexity into a claim that "we are happy with driver performance" ?

            Can I ask a favour ? When you make claims like this please quote the statements where you think we said those things. I expect that will help both of us
            Last edited by bridgman; 08 September 2011, 09:01 AM.
            Test signature

            Comment


            • #46
              Originally posted by bridgman View Post
              That's not a radeon vs nouveau thing as much as a hardware thing -- the fact that AMD GPUs keep pretty much all of their microcode in RAM while other vendors build most of it into the chip. If our microcode was burned into the chip where you couldn't see it would we somehow be more "open source friendly" ?
              Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

              Nouveau has less to mess with microcode and blob loading. All three major graphics providers utilize proprietary hardware design, so .. less is more. Next step is uncloak microcode source, which is not to happen for both, so nouveau(not nvidia) wins as it does not have to inject blobs into memory. Nouveau?s walls are a bit more distanced.

              Originally posted by bridgman View Post
              I see the same thing with CPUs -- processors which don't have upgradable microcode are somehow felt to be "more open source friendly" than processors where you can update the microcode after purchase. This whole "if we have to see the microcode it's evil but if it's out of sight it's OK" perspective baffles me -- it's like the belief that building a wall around a resort so you can't see the poor people outside somehow makes the problem go away.
              First step to freedom is open application, second are open libraries, kernel and drivers, third - open hardware. Hardware?s bits and bytes are not referred as application due to it performing mostly technical low-level stuff normally not exceeding the level of the hardware part where it seats. This is why microcode in flashrom is referred much more to hardware functionality that doing same in driver. Sure, you load microcode on each boot, were others have it in ROM. Were there any microcode based keyboard keyloggers/spyware/viruses written so far? A driver that loads something closed source in memory on each boot..

              Originally posted by bridgman View Post
              Why would you even think this ? The only thing we have said w.r.t. performance is that BASED ON THE COMPLEXITY OF THE HARDWARE, THE SIZE OF THE COMMUNITY, AND COMMUNITY DEVELOPER STATEMENTS ABOUT THE TYPES OF OPTIMIZATIONS THEY WOULD AND WOULD NOT DO we expected the open drivers to run with approximately 60-70% of the fglrx driver's performance. I think it's fair to say that we would be ecstatic if the open source driver eliminated the need for fglrx in markets like workstation, but it seems unlikely given the magnitude of the task.
              Drivers are rather complex and they must adapt quickly to newer models/libraries which means it requires huge driving force. Driving force that may be achieved by:
              a) every single AMD card buyer hacking your card for better drivers
              b) a split of payment you get from each AMD card buyer when he purchases the card
              So, if you wanted opensource, you would make it happen. But instead we watch the clouds passing by.
              Was microsoft also watching clouds when they formed monopoly? Regarding the size of the market... I?m sure they were planning its future size, not current. Cause currently there were just clouds

              Originally posted by bridgman View Post
              When we don't release info it's because of IP or DRM reasons, certainly not a desire to hold back the open source driver.
              Seriously, I hold it for not probable for AMD to create its windows driver and not tracking IP they use; solely refering to fglrx obfuscator to blend everything just no one notes. I mean, when I buy the card and install the driver, I?m not instantly sued by AMD?s 3rd party companies for using that driver. So obfuscation works?
              This whole IP thing, recognized only in US and banned in New Zealand, works and spins if major players accept it. So, unless AMD does something about it (like Google), the opensource driver would be a beef tenderloin of code remnants. Something what surely won?t support the card possibilities in the matter of assuring the paying customer to pick your card from the shelves of hardware store.
              The impression I got from our smalltalk was that AMD is completely ok with current situation, which I?m not, so...


              Originally posted by bridgman View Post
              This is totally untrue. I said a couple of things, none of them even remotely close to your statement :
              1. if we put a mechanism in place to allocate development funds between proprietary and open source drivers based on the money we made from the two markets there was a serious chance that the outcome would be *less* funding for open source, not *more*
              2. even if the numbers worked out in favour of shifting relatively more funding to the open drivers at some point in the future I don't think we could afford to substantially reduce fglrx funding relative to the requirements of the markets it serves, so this really boils down to a "give me more money" request
              1. I?m very sure that the most expensive car(card, space rocket, coffee bean, diamond,software) is the most polished one - in terms that >justify its price<. Not the one that is of blackbox or glass-clean design. Actually glass-clean design adds to the point that people know its nanometals(silicon particles, hull plating, structure, carats, bits) ain?t glued from bio-waste. So, there is no sense to sell open driver, unless you want it yourself and probably no one is going to push it forward except yourself. Of course community support of improves this. So you are perfectly ok with current situation, which forces me to use closed driver. Im glad you cannot force me to use windows, thanks at least for that
              2. Your customers will appreciate open drivers, thats very sure. If they are on paar functionality with closed ones. I don?t know any case, where customer will prefer closed driver to equal quality open one.

              Originally posted by bridgman View Post
              If we were happy with current state why would we have been working for years to bring open source driver support earlier in time and closer to hardware launch ? Why would we be working to release programming info in more areas ?
              "We" are barely enough to handle the complexity, which requires some financial model; otherwise "We" are perfect.

              Originally posted by bridgman View Post
              If we were happy with current development rate why would we be hiring more developers ?
              So, you sure you will be able to provide sameday support like catalyst for hd8k? So, that people could go out to shops, pick card up, insert it and enjoy?

              Originally posted by bridgman View Post
              How can you possibly twist a performance estimate based on the size of the development team relative to the domain complexity into a claim that "we are happy with driver performance" ?
              The devil is in details. The devil means completeness, power, efficiency.
              Fglrx has much more details and I don?t think whole 2k crew is drinking beer (maybe wrong beer? ) so fglrx still has bugs with each release. I think, 2k people actually only help correct specific workstation bugs from firegl customers, but, although I do not use AMD cards now, I see fglrx gets more and more in form... To the point when there will be no difference between proprietary nvidia and proprietary amd.

              You claim you can reach 70% speed of highly detailed supercar(sorry! :P) assembled by 2k people with car which, assembled by 10 people with help of neighbourhood from household objects. While I never argumented with this "less is a bit less from more" strategy, I seriously question you can win championships in f1 made form household parts. I wonder where will my money go when I purchase AMD card again.. best possibility is fglrx. The car you are racing against. Or together against nvidia.

              So, my opinion is: fglrx tries to match nvidia driver; opensource radeon tries to match... erm,.. nouveau driver. Everyone is happy, except me :/
              Oh well, maybe I should put a test drive of proprietary blobs, ie gods at some much later time. I don?t think fglrx will loose to opensource radeon; and I know the reason, which you probably know too, but call it miracle twist. Oh well


              Originally posted by bridgman View Post
              Can I ask a favour ? When you make claims like this please quote the statements where you think we said those things. I expect that will help both of us
              Sure, but statements not always transfer the exact meaning, especially over internet. I think even when speaking directly, there is place for missunderstanding. But sure.
              Last edited by crazycheese; 08 September 2011, 12:55 PM.

              Comment


              • #47
                Originally posted by crazycheese View Post
                So, if you wanted opensource, you would make it happen. But instead we watch the clouds passing by.
                We did make it happen.

                Now you are asking if we can take the open source drivers back from the community, make them "AMD official", and spend a big pile of new money funding open source driver development with a goal of matching the performance and functionality of the proprietary drivers, right ?

                Whatever happened to people asking for things like "world peace" ?
                Last edited by bridgman; 08 September 2011, 01:48 PM.
                Test signature

                Comment


                • #48
                  Originally posted by bridgman View Post
                  We did make it happen.

                  Now you are asking if we can take the open source drivers back from the community, make them "AMD official", and spend a big pile of new money funding open source driver development with a goal of matching the performance and functionality of the proprietary drivers, right ?

                  Whatever happened to people asking for things like "world peace" ?
                  People thought they should set their goals higher.

                  Comment


                  • #49
                    Can't argue with that, I guess...
                    Test signature

                    Comment


                    • #50
                      Originally posted by bridgman View Post
                      We did make it happen.

                      Now you are asking if we can take the open source drivers back from the community, make them "AMD official", and spend a big pile of new money funding open source driver development with a goal of matching the performance and functionality of the proprietary drivers, right ?

                      Whatever happened to people asking for things like "world peace" ?
                      I?m asking if you can make them "AMD official", and allow people to spend their money funding open source driver development (using transparent system) with a goal of matching the performance and functionality of the proprietary drivers.

                      I don?t know what happened to them. But should radeon match nvidia blob in performance/features even close and should it be backed up seriously by your company, same as catalyst now; hardly anyone will buy nvidia for linux market. Small ever increasing linux gaming market.

                      Comment

                      Working...
                      X