Announcement

Collapse
No announcement yet.

Nouveau vs. NVIDIA Linux Comparison Shows Shortcomings

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

  • #11
    ..Radeon driver holds the leads with power management...
    - Huh, its that bad.

    Comment


    • #12
      Originally posted by GreatEmerald View Post
      Then again, having docs may very well raise interest in the project. It wouldn't look like doing things blindly any more.
      Docs can also be crappy and near useless (except for public opinion). If you tried to write a driver for r600 having just the released docs, you'd probably fail miserably. Unless you count existing code as docs. Hey then we have docs, too
      We actually have a nice register database and a shader [dis]assembler. That's just as useful as the stuff at http://www.x.org/docs/AMD/.
      Last edited by calim; 04-23-2013, 04:03 PM.

      Comment


      • #13
        I don't think nouveau is a wasted effort. Its a good research project. Ofc, they will barely catch-up against corporation.
        .. then, if you buy nvidia card, you accept the consequences. You buy a card that should work with nvidia-own drivers on nvidia-supported system, and from windows-like user perspective, its ok. Ie, you buy a piece of hardware with specific performance and specific life-time.

        Its totally not ok, if you believe in free software however. Currently we have Intel doing best in free drivers, but their hardware is very limited.

        I recently purchased laptop with mobile radeon x1900 and HD5850.
        Just gave it a shot, I can sell it off anyway if it doesn't work out.

        You can clock the card. You can get temperatures. You can check card load. You can run games. You can use desktop, its a bit slower than Intel though.
        You have developer attention. Opensource radeon driver is developed, ofc way slower than Intel, but much faster than nouveau.

        Recently you have video acceleration by means of R600+ driver and OpenCL is in the works.

        My experience on opensource drivers is following:
        [x] want automatic power managment
        [x] want mesa to catch-up in terms of still lacking 3D performance

        nothing more. Its way better than I expected, actually.
        Last edited by brosis; 04-23-2013, 05:57 PM.

        Comment


        • #14
          I must admit I'm a little surprised not only to see regular people calling Nouveau a total waste, but to think that an actual _dev_ comes here to confirm it. I mean that's just.. amazing.

          And Nouveau is nowhere near perfect. It never will be. Something will always be broken, so including phrases like "shortcomings" in the title is just pure ignorance.

          Was the goal of this thing ever to even catch up to the proprietary driver anyway or to just mainly add things like KMS support for Nvidia and then graphics acceleration as the situations come along?

          I don't see the reason to, if it's not _just_ as fast as the binary version, call it a "waste". A waste is something that turned out to be pointless work. I think it's plain r*tarded to call an open source display driver _pointless_ on the grounds of its performance not equalling the original one.
          Last edited by ArchLinux; 04-23-2013, 07:12 PM.

          Comment


          • #15
            I tested Nv GTX 295 and Nv GT630 OEM (Kepler) with TF2 and Mesa 9.1 final some weeks ago. Interestingly the Kepler based card had much fewer rendering errors than the much older one. I did not check the clocks while testing and just used the default settings. Some would guess that nouveau is better for older cards but that does not seem to be the case, maybe with the exception of some very slow onboard solutions like ION. In general you get what you want with the official Nv drivers if speed is what really matters, nouveau is usally fine enough to see something and use composite effects for kde/compiz/gnome3. My GTX 295 however gets really hot using nouveau and i would not use it for long time - if you have got a better card than a very lowend one. Same applies more or less to radeon as well - with binary drivers you can still save energy/money. Basically Intel is still the way to go when it comes to oss only drivers, sometimes you need to fix applications too to get it stable however (for Intel+Mesa 9.1 i had to patch the kde-workspace package - basically only kwin). Until powermanagement is fully working with every oss driver i would always prefer binary drivers - but radeon is interesting now too because of the new uvd/vdpau support that does not exist in the binary one.

            http://kanotix.com/files/fix/tmp/tf2/tf2-benchmarks.txt

            Comment


            • #16
              Originally posted by calim View Post
              That's not a fix, it's a workaround. If you had a fix, we might just care.
              It's certainly a workaround if acceleration is important. But consider Live CDs and the like. These simply won't boot properly on these GPUs right now, so you can't even install a modern Linux-based OS! No acceleration is still preferable to crashing. I just don't see the point in not applying the workaround, after this has been broken for such a long time with no proper fixes in sight.

              Comment


              • #17
                Wow, I didn't realise it was a developer! Oh my gosh...
                Michael should really post an article about this...


                Subliminal message: "We don't care if this driver works at all"
                Resolution: Workaround allows to work around the problem, so people can still work, waiting for a real fix. Ignoring workarounds equals to not caring about anyone who uses the software. Thats a kick in userbase butt.
                Originally posted by calim View Post
                That's not a fix, it's a workaround. If you had a fix, we might just care.

                Subliminal message: "We don't care about quality of this driver at all"
                Resolution: This is as "hobby" as it gets.
                Originally posted by calim View Post
                Myself I really don't care (well, I do, but not to the point where I'd do something about it), my ~2000 hours of wasted free time on this project would be even more wasted (on the personal level anyway)

                Subliminal message: "We don't care about Linux user base. They can go to hell."
                Resolution: If someone refuses even to accept free cards in order to make Linux more attractive to new users, he definately gives a brick about Linux userbase. Expect nothing even close to serious from this crowd. The code he pushes this way is "mess" in all forms.
                Originally posted by calim View Post
                Myself I really don't care (well, I do, but not to the point where I'd do something about it), my ~2000 hours of wasted free time on this project would be even more wasted (on the personal level anyway) if I'd spent them with the annoying bugs of users with cards I don't have (which is the factor that makes fixing things really hard, but don't suggest sending cards to me I don't have the time or nerve), instead of focusing on the fun stuff.

                Subliminal message: "Our driver will always be constantly broken"
                Resolution: -
                Originally posted by calim View Post
                It's either progress or stability, and we seem to focus very much on progress over stability overall.

                Subliminal message: "We don't code, we don't improve, we don't want to organize anything. We just hang around, breaking stuff for fun. And we like to make fun of AMD developers. Because, they, unlike us, actually write STABLE CODE. We - either write mess, or don't write anything. Hell, we better off watch TV drinkin beers. Linux sucks anyways --.. and not because of us! Rly."
                Resolution: -
                Originally posted by calim View Post
                Of course, "no docs" is just an excuse, even with docs things wouldn't be moving along much faster at this point because there's no one there to do the time-consuming part, i.e. actually implementing things. That's their biggest advantage, AMD hires devs (and what ever happened to "the community will write the driver" ?)


                Suggestion:
                Linus - Please REMOVE NOUVEAU out of kernel. Reiserfs is much more worthy of attention than this.
                Distributions - under NO WAY build or include nouveau. Use either VESA or nvidia. This is sure way to immediately kill your userbase with nvidia cards.
                Developers - unless you want to join Nvidia, its much better to join forces who actually CARE about code they write. That means on x86 - Intel or AMD.
                Nvidia users - under NO CIRCUMSTANCE donate your hardware to nouveau. Please, SELL it. At least, you will not waste your money.
                Last edited by brosis; 04-23-2013, 08:41 PM.

                Comment


                • #18
                  Originally posted by Kano View Post
                  Until powermanagement is fully working with every oss driver i would always prefer binary drivers - but radeon is interesting now too because of the new uvd/vdpau support that does not exist in the binary one.

                  http://kanotix.com/files/fix/tmp/tf2/tf2-benchmarks.txt
                  Kano, powermanagement works with radeon. Its automatical powerstate switching that is not implemented or even better load-based powerswitching. But user can manually switch the powerstates.
                  Perhaps you mean "powermanagement is complete" instead of "fully working", because its fully working, but is not complete featurewise ...

                  Comment


                  • #19
                    lol

                    let the trolling begin.

                    Comment


                    • #20
                      Originally posted by brosis View Post
                      Kano, powermanagement works with radeon. Its automatical powerstate switching that is not implemented or even better load-based powerswitching. But user can manually switch the powerstates.
                      Perhaps you mean "powermanagement is complete" instead of "fully working", because its fully working, but is not complete featurewise ...
                      I don't think that's true on the newer APU parts - they aren't allowed to run at full speed at all, because of PM issues.

                      Comment

                      Working...
                      X