Announcement

Collapse
No announcement yet.

AMD Open-Source S.I. Botched, Hope For The Future

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

  • #16
    Originally posted by agd5f View Post
    I don't see how SI support was "botched."
    Let me start by saying, "Thank you for your contributions, they are greatly appreciated. Keep up the great work"

    Regarding "botched". This isn't a personal criticism for any member of the team developing RadeonSI OSS drivers, it is a criticism of AMD's priorities, timelines, and release management. Any snarkiness that you are getting from entertainers like Michael, or the community over which he holds custodianship, is due to the jealousy we all feel when we see the Catalyst driver performing. People like to complain and whine. It's human nature. We like to complain about the OSS driver. Catalyst's existence makes it easy for us to do so. OSS developers are in the unfortunate position of being a liaison between consumers and the business side of AMD by virtue of their accessibility.

    I'm not certain that (the business end of) AMD understands the situation, or the potential stakes involved. Looking at their history, one would be hard pressed to remain optimistic.

    Don't get down on the snarky comments. Understand that our issues are real, and that the entity responsible for our issues is the business side of AMD.

    F

    Comment


    • #17
      I couldn't agree more. I have nothing but big respect for the developers. I think in ~5 years from now, when mesa/gallium are really mature and supporting several GPUs, people will realise how the user experience will be a far cry from any closed drivers. And anyone working on this endevour should receive a major thanks.

      For the first time I can say very confidently that Linux is the future. Right now the user experience is far from pleasant, but as more and more companies get on board, the tiping point starts to show up in the horizon.

      Many Linux based OSes are showing up and OSS drivers will be what allows any hardware vendor to seamlessly support each of these OSes with close to no extra effort at all. With Ubuntu WebApps and Office 365, it will probably be a better experience to use MS Office in Ubuntu than in Windows 8.

      This is why I whish that AMD had more vision (pun intended). Windows won't be here forever, and you cannot rely on the community to do your heavy lifting for you. Linux is no longer a toy OS and the future is getting brighter and brighter. Right now AMD only supports a single mobile OS, one that hasn't even been lauched yet: Win8. If Win8 fails in the market place, AMD cannot support android, chrome OS, firefox OS, webOS, tizen, ubuntu, meego. If Valve lauches a Linux based SteamBox, it will have to rely on the buggy Catalyst for any kind of support, nVidia and intel will be light years ahead.

      This is why I urge AMD to stop betting all your horses on a single patform, start hiring/training more great people such as these guys right now. At the cost of developing one great OSS driver, the company will be able to support countless OSes. You cannot get a better deal than this.

      Comment


      • #18
        Originally posted by bridgman View Post
        It's funny... I was expecting an news post more like "Open source SI fixes pushed, glxgears now runs, open source OpenCL work proceeding on schedule". Oh well...
        Yeah, people give you way too much shit. Just curious though, why did you guys go with GLAMOR? I have some ideas why you might have chose it, but I am quite curious as to AMD's reasoning behind it.

        Comment


        • #19
          Originally posted by LinuxID10T View Post
          Yeah, people give you way too much shit. Just curious though, why did you guys go with GLAMOR? I have some ideas why you might have chose it, but I am quite curious as to AMD's reasoning behind it.
          Please see any of the glamor threads for extended discussion. Short answer: EXA needs work to to properly support additional RENDER features. The gallium Xorg state tracker uses EXA so that would have the same limitations as a native EXA driver. Either way you start out with limitations on what can be supported going forward. Properly supporting "2D" acceleration requires most of what you'd need for 3D (shader compiler, more advanced state handling, support for multiple generations of hw, etc.) and is a huge amount of work (see SNA for example). Glamor takes advantage of the effort already required for OpenGL (the 3D driver), and allows you to implement support for more advanced functionality using OpenGL which is an API a lot of people are already familiar with.

          Comment


          • #20
            Ya, you were stupid enough to buy our hardware but you can't use it so stop whining. Is not an acceptable position for any company to take. As for MESA it will never be any where near up to par as long as the card manufacturers refuse to let people know how the hardware is designed to work. This isn't like looking at an analogue circuit. Unless you know how the chips are designed and work together the amount of work required to reverse engineer how a modern graphics card works is beyond the resources that any one could reasonably be expected to expend. Documentation is the only viable way to have what should be considered a reasonable level of performance in this day and age. But both NVidia and AMD have decided that they are never going to provide what is needed so the open source drivers for these companies will always suck.

            As for the SI cards. I bought mine after Michael had posted that AMD was dropping support for the older cards in FGLRX. I figured that if I bought any thing older than SI I wouldn't even be able to give the card away if it didn't work out. And AMD was plenty quiet about how the open source drivers were going to work on SI. I live on the internet researching things for my profession and I hadn't run across any thing until my card worked like crap. I haven't had any problems with any piece of hardware on Linux in 7 or 8 years. I assumed that had all gone away years ago. At least I was able to dig my $70 O!Play out of the closet and can watch videos on that while I decide what to do next. If I can figure out some way to take the cooling system off of my SI card and make it work on my old HD4000 card I might try that. Other wise I will just throw it in the garbage and move onto buying a whole new computer. Ask me how I feel about my customer experience AMD. The last time I felt this screwed was when NVidia put out their flawed graphics chips and HP forced a patch on me to down grade my graphics performance.

            I finally got the Fedora guys to start building and packaging the radeonsi driver but it looks like that was a complete waist of time, if the rest of the stack isn't going to work. It isn't the wasted money for the card that really ticks me off. It is being without a proper functioning computer for so long while I try to work out how to get all these new pieces working. I just can't afford the down time.

            Comment


            • #21
              Originally posted by jvillain View Post
              Ya, you were stupid enough to buy our hardware but you can't use it so stop whining. Is not an acceptable position for any company to take. As for MESA it will never be any where near up to par as long as the card manufacturers refuse to let people know how the hardware is designed to work. This isn't like looking at an analogue circuit. Unless you know how the chips are designed and work together the amount of work required to reverse engineer how a modern graphics card works is beyond the resources that any one could reasonably be expected to expend. Documentation is the only viable way to have what should be considered a reasonable level of performance in this day and age. But both NVidia and AMD have decided that they are never going to provide what is needed so the open source drivers for these companies will always suck.
              AMD official support is through the closed source driver. They support the community effort to make an open source driver but i don't think it's considered as what they recommand to use to their customer. You better get your fact right before making claim. One can totaly reverse engineer GPU without knowing deeply how things works. nouveau is a brilliant example of that. On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver. Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.

              Comment


              • #22
                Originally posted by glisse View Post
                AMD official support is through the closed source driver. They support the community effort to make an open source driver but i don't think it's considered as what they recommand to use to their customer. You better get your fact right before making claim. One can totaly reverse engineer GPU without knowing deeply how things works. nouveau is a brilliant example of that. On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver. Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.
                well make LOT of sense that features like video decoding get consireded as secondary when there are still many important bits to be finished like get parity with opengl specs or improve shaders compilers, etc but either way it seems you can't stop conspiracy junkies and nVidia fanboys for now since they seem just to bypass this kind of post and go on trolling sad but true

                Comment


                • #23
                  On the other end AMD is providing documentations that cover all the acceleration bits you need to make a driver.
                  More like several years ago AMD made some documentation available for the cards back then and have done next to nothing since. I understand people with older cards being happy. I didn't have any big beefs with how my HD4000 worked until it failed. There is all the difference in the world between reading about these problems and living through them.

                  Yes there is lacking area like power management or video decoding but it was for long time considered as secondary.
                  Every thing is lacking from the stand point of an end user because all these pieces were so late coming. Yes there is experimental code available in non main stream repositories but hunting them down building them and then trying to get them to work with out breaking other parts of the system is more than a little problematic. Especially since the only documentation on how it is supposed to fit together is what Michael is posting. If they aren't even including the required parts in Rawhide as even a proof of concept then end users aren't going to see this stuff until at least F19 or 20. It is guaranteed not to be in F18 so we are what a year away or so from support for end users in even the most bleeding edge distros? Reinstalling with Fedora 13 or what ever the last version that the blob works with isn't an option at al, for me. Not because of pig headedness but because other things I need aren't there. Ignoring these problems won't make them go away. What it will do however is fool people into thinking they can use these SI cards which will then leave them with a shit experience that will turn them off of AMD products all together like what has happened to me. Or worse they will just say fuck it and load windows or go buy a mac.

                  Comment


                  • #24
                    Originally posted by jvillain View Post
                    More like several years ago AMD made some documentation available for the cards back then and have done next to nothing since. I understand people with older cards being happy. I didn't have any big beefs with how my HD4000 worked until it failed. There is all the difference in the world between reading about these problems and living through them.



                    Every thing is lacking from the stand point of an end user because all these pieces were so late coming. Yes there is experimental code available in non main stream repositories but hunting them down building them and then trying to get them to work with out breaking other parts of the system is more than a little problematic. Especially since the only documentation on how it is supposed to fit together is what Michael is posting. If they aren't even including the required parts in Rawhide as even a proof of concept then end users aren't going to see this stuff until at least F19 or 20. It is guaranteed not to be in F18 so we are what a year away or so from support for end users in even the most bleeding edge distros? Reinstalling with Fedora 13 or what ever the last version that the blob works with isn't an option at al, for me. Not because of pig headedness but because other things I need aren't there. Ignoring these problems won't make them go away. What it will do however is fool people into thinking they can use these SI cards which will then leave them with a shit experience that will turn them off of AMD products all together like what has happened to me. Or worse they will just say fuck it and load windows or go buy a mac.
                    1. the documentation is available up to 7000 series [just missing some bits here and there like UVD] ask bridgman or glisse for a link since i don't remember it right now but you need to understand this documentation is not about how to write a driver but a pure GPU ASM and electronic registers like the CPU documentation from intel/AMD or any other vendor(take a look and you will see what pure asm register looks like) note that this is not a mistake documentation should be like that. so all opengl call end being a bunch of those ASM commands depending of the GPU so is mesa/gallium problem how it use those ASM bits and that require extensive analize and test.[trust me when i say is anything but trivial in fact extremely hard don't even gets close LOL]

                    2. OSS driver are not meant for ppl that can't properly use git for now since this drivers and most of the linux graphics is in alpha state[aka under heavy development], so you should stick to fglrx [even if it means downgrade the distro] wich is the officially supported driver[i know it sucks]. so if you are willing to be an alpha tester like many of us you should [maybe having a second partition so you don't mess your main distro] build regularly these components[i do it daily | once you get used to it is fast]

                    * kernel[stick to drm-next if possible until the next kernel rc start coming[may kill kittens so keep an stable kernel around]]
                    * libdrm with libkms git
                    * libxcb git
                    * mesa git
                    * cairo in case you wanna test weston
                    * DDX git
                    * glamor git[broken for xserver 1.13]
                    * xorg-server git[glamor you need to stick with 1.12 for now]
                    * put in a bookmark bar mesa bugzilla
                    * enjoy

                    another good tip is read daily mesa/ddx cgit logs so you can pinpoint when a nice patch lands to test it
                    another one could be install apitrace 3.0 it does wonders and even save kittens/puppies

                    i understand this seems overkill but no one ever has said that OSS graphic stack is ready for production or even worse recommended for regular user as official driver, ofc it will get there since everyday keep getting closer and remember that 2 or 3 years ago linux graphic support was near 0[like bsd or solaris] so the graphic team basically has to rethink everything from scratch to be where we are today with an competitive modern graphic stack[technically speaking since is still work in progress but technically speaking the new stack is more capable than windows/mac put togheter they just need to end putting all the bits togheter and optimize but they are like 8 ppl due to the massive level of knowledge you need to program a GPU]
                    Last edited by jrch2k8; 07-30-2012, 05:27 PM.

                    Comment


                    • #25
                      Originally posted by jvillain View Post
                      Or worse they will just say fuck it and load windows or go buy a mac.
                      Linux has a much lower desktop market share and as such a much lower level of investment across the board. This is true across the board (hw vendors, oems, desktop software stacks, etc.) Yes, I know, chicken and egg. Life is hard on the fringe, but can also be more interesting.

                      The closed source driver has full support for all currently available cards (largely because it is able to leverage code shared with windows) and supports a wide range of distro versions.

                      Comment


                      • #26
                        Originally posted by jvillain View Post
                        Ya, you were stupid enough to buy our hardware but you can't use it so stop whining. Is not an acceptable position for any company to take.
                        Neither is this an acceptable position for a user to take.
                        "Can't use it"??? Eh? HD7770 owner here. Working fine. Perfect? Nope, but tell you what, when I find a bug, my first reaction is NOT to come here waving arms around, screaming bloody murder and making doomsday claims. Instead I GATHER INFORMATION TO GET THE BUG FIXED.
                        So maybe, just maybe, being more specific instead of whining would make you more valuable?

                        Comment


                        • #27
                          jrch2k8 thank you for your well thought out post. My training was in hardware design and I have read many spec sheets for chips and devices over the years. I went and looked at what AMD had available on the developer site a couple of weeks ago and know what these are. Again thank you for your post.

                          But I have hit the limit of what I am willing to put up with. I bit the bullet and went and bought the parts to build an Intel based machine. Tomorrow I will take my HD7700 card out of my machine and enjoyed smashing it with a hammer. Yes I might be one customer but I also have considerable input in the purchasing decisions of a number of companies. I also push the companies I work with to help support open source in various ways. I'm not an arm waver as some momo just described me I have been a specialist in Linux since the early days of Slackware. If I am as frustrated as I am there is a real problem. I was also called a lousy customer. But there is an old saying in business. "If the customer is mad they have a right to be. Either because you did some thing wrong or because you didn't manage expectations." Where AMD is running into problems is in trying to have their cake and eat it to. You can't be the saviour of open source and at the same time not support open source at all. Pick a message and stick with it. But as AMD continues to shed market share I am sure they will continue to sit around and scratch their heads as to why.

                          Feel free to close my account Michael.

                          Comment


                          • #28
                            Originally posted by jvillain View Post
                            jrch2k8 thank you for your well thought out post. My training was in hardware design and I have read many spec sheets for chips and devices over the years. I went and looked at what AMD had available on the developer site a couple of weeks ago and know what these are. Again thank you for your post.
                            You mentioned "developer site" so I'm just checking... you were looking at the documentation section of RadeonFeature, right ? If you look anywhere else you're only going to see a subset...

                            http://www.x.org/wiki/RadeonFeature#Documentation

                            Originally posted by jvillain View Post
                            You can't be the saviour of open source and at the same time not support open source at all. Pick a message and stick with it.
                            With respect, I believe we are doing just that. I don't know where either of your statements came from.

                            I don't believe we have ever even hinted at being anything like the saviour of open source. We said that we would start supporting open source driver development again while continuing to work on the fglrx binary driver in parallel.

                            You say "not support open source at all", and I don't understand how you can think that. We have four full time developers working on the open source drivers and supporting other community developers, plus others working part time and behind the scenes. Is there some terribly evil thing we're doing which somehow invalidates all that for you ?
                            Last edited by bridgman; 07-30-2012, 09:29 PM.

                            Comment


                            • #29
                              What the hell... honestly it's practically amazing what the open source AMD team has accomplished with their resources.

                              I don't know how some of this stuff gets so twisted.

                              Comment


                              • #30
                                Originally posted by agd5f View Post
                                I would posit just the opposite. In that case, the user probably can't tell whether he's using vesa or a native driver. Both will put an image on the screen a perform quite well for non-3D tasks which is what the vast majority of users do. Even for basic 3D, llvmpipe should do a pretty decent job in the interim. For that, the driver already works quite well.
                                Try with a laptop, it will became warm in a second, the battery will just yell for the discharge rate. It is not only a 3D performance problem. I tried llvmpipe (by accident to be honest) on my brand new quad core ivy bridge. It is not bad, the first minute I didn't saw the difference, but then I started moving windows around.... KDE works only with xrender backend with llvmpipe and you can see the difference. Now this is with an high end CPU.... Really you can't pretend users are ok with this. Some user might be, but really you simply can't ship it as ready for the mass. I would not say it is botched, it is not. It is simply not ready. As John said the community didn't worked a lot on the driver -> you can't expect it will autocomplete. I'm not pointing the finger to someone here, nobody is doing a bad work here, I want this to be clear. But radeon still has major missing features for the masses.

                                Comment

                                Working...
                                X