Announcement

Collapse
No announcement yet.

Radeon Linux Benchmarks: Catalyst 15.3 Beta vs. Linux 4.0 + Mesa 10.6-devel

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

  • #11
    So where is the 15.3 driver that is supposed to be released this month?

    Comment


    • #12
      Hey Michael, have you tried running:

      Code:
      find ~/.steam/root/ \( -name "libgcc_s.so*" -o -name "libstdc++.so*" -o -name "libxcb.so*" \) -print -delete
      This should allow you to run Steam game benchmarks using open source drivers. It gets rid of libgcc_s.so, libstdc++.so, and libxcb.so files from the Steam runtime libraries.

      Comment


      • #13
        Originally posted by GraysonPeddie View Post
        Hey Michael, have you tried running:

        Code:
        find ~/.steam/root/ \( -name "libgcc_s.so*" -o -name "libstdc++.so*" -o -name "libxcb.so*" \) -print -delete
        This should allow you to run Steam game benchmarks using open source drivers. It gets rid of libgcc_s.so, libstdc++.so, and libxcb.so files from the Steam runtime libraries.
        For gentoo help too.

        Comment


        • #14
          well he wrote

          not the common library issue
          so I guess he already tried it.

          Comment


          • #15
            He should write "probably not the common library issue", because when "X errors were being generated and blocked the tests" it is probably again library issue

            Comment


            • #16
              Originally posted by smitty3268 View Post
              Neither are the Mesa and kernel versions he is testing.

              I just think it's kind of silly to use git versions of those but not LLVM - either go all-git, or all-released. A frankenstein mix of the two is weird, and claiming in the article that it's the latest code is just plain misleading.
              Not really seeing as AMD isn't using LLVM 3.7 for the Catalyst driver, until LLVM 3.7 is out.

              Comment


              • #17
                In case it makes a difference we don't use LLVM in the Catalyst graphics stack -- only in the OpenCL compiler.
                Test signature

                Comment


                • #18
                  Catalyst vs Mesa

                  I don't know much about how gfx drivers work, but I wonder why, after so many years, there is still such a huge difference in performances?

                  Comment


                  • #19
                    Originally posted by jacob View Post
                    I don't know much about how gfx drivers work, but I wonder why, after so many years, there is still such a huge difference in performances?
                    1. The most obvious difference is the number of developers working on each of the drivers. Catalyst shares a lot of code across multiple OSes, particularly the 3D bits, so performance tuning done for other OSes usually helps Linux as well. The open source stack is not quite Linux-specific (it's used by BSD and a couple of embedded OSes as well) but pretty close.

                    Catching up with Catalyst is not as obvious an outcome as you might expect because performance work on Catalyst is still going on as well, so while the open source driver keeps getting faster the Catalyst stack keeps getting faster too. Main focus for the open source developers right now is adding GL features and new HW support, not performance tuning.

                    2. The open source shader compiler for GCN (3 of the 4 cards) needs to do some complicated things -- GCN compute units have scalar & vector processors running off the same instruction stream, so essentially the compiler is being asked to auto-generate the kind of code that is still written as hand-tweaked assembler on most x86es -- and some of the more advanced functions are just starting to come up now. That's why you're seeing the flak about LLVM 3.6 vs 3.7 -- it's not the generic LLVM code folks are talking about but the GCN-specific code generator that is maintained in the LLVM stack.

                    3. A few folks have mentioned that there seems to be some kind of regression for the 6870, which is the only card with a driver stack that actually *has* been around for anything like "so many years".

                    4. Optimizing the driver for frame rates much over 100 Hz is mostly useful for benchmark competitions and the open source folks don't have to care about them. Catalyst is a different story because companies still make buying decisions based on one card running 275 FPS and another running 297 FPS, even though they're both being used with a 60 Hz display.
                    Last edited by bridgman; 31 March 2015, 08:18 PM.
                    Test signature

                    Comment


                    • #20
                      Originally posted by bridgman View Post
                      1. The most obvious difference is the number of developers working on each of the drivers. Catalyst shares a lot of code across multiple OSes, particularly the 3D bits, so performance tuning done for other OSes usually helps Linux as well. The open source stack is not quite Linux-specific (it's used by BSD and a couple of embedded OSes as well) but pretty close.

                      Catching up with Catalyst is not as obvious an outcome as you might expect because performance work on Catalyst is still going on as well, so while the open source driver keeps getting faster the Catalyst stack keeps getting faster too. Main focus for the open source developers right now is adding GL features and new HW support, not performance tuning.

                      2. The open source shader compiler for GCN (3 of the 4 cards) needs to do some complicated things -- GCN compute units have scalar & vector processors running off the same instruction stream, so essentially the compiler is being asked to auto-generate the kind of code that is still written as hand-tweaked assembler on most x86es -- and some of the more advanced functions are just starting to come up now. That's why you're seeing the flak about LLVM 3.6 vs 3.7 -- it's not the generic LLVM code folks are talking about but the GCN-specific code generator that is maintained in the LLVM stack.

                      3. A few folks have mentioned that there seems to be some kind of regression for the 6870, which is the only card with a driver stack that actually *has* been around for anything like "so many years".

                      4. Optimizing the driver for frame rates much over 100 Hz is mostly useful for benchmark competitions and the open source folks don't have to care about them. Catalyst is a different story because companies still make buying decisions based on one card running 275 FPS and another running 297 FPS, even though they're both being used with a 60 Hz display.

                      well bridgman and all the people in this thread, yes fglrx show higher FPS and i did install it in my Arch system just now. in resume awesome for benchmarks awful as always for everything else. In specific:

                      1.) All my current wine games stopped working properly (skyrim, deadpool, starcraft 2 HOTS) and FPS went downhill pretty fast where gallium nine runned them in my 7770 just perfectly fine
                      2.) Most my steam games are actually slower, specially the witcher 2(totally fucking unplayable) and Civ5 + lots of mods(playable but lots slower than radeonsi <-- WoW just Wow)
                      3.) Video accel went to hell, xvba still hangs the whole freaking kernel all the way to panic (tested from 3.15 to 3.19) <-- this one may be me, i know there are some wizards out there that spend weeks on it and kinda work for them but i just wont waste time when radeonsi give it to me out of the box through any API i favor(VA, Vdpau, OpenMAX)
                      4.) OpenCL just dont work properly(as always) where mesa opencl runs some demos like luxmark pretty nicely
                      5.) Of course 0 wayland support <-- this was obvious, just saying
                      6.) weird random X crashes with 1.16 and 1.17 not so much with 1.15, where radeonsi is mother freaking rock solid for me
                      7.) no gallium HUD <-- another obvious one but i feel i miss a leg without it
                      8.) btw gnome-3.16 is not amused about fglrx either, plasma don't show much love for it either
                      9.) blender for some reason went hostal too but when it works is a bit faster than radeonsi
                      10.) 2D accel beyond benchmarks is nothing to be proud of, specially on Qt5(i suspect fglrx team just supercharged the routines needed for benchs to look good and just fuck off the rest)
                      11.) i feel firefox and chromium render slower but is very eye ish, don't trust me on this one

                      At the end of the day, i really appreciate radeonsi exist otherwise i would have through my graphic card in the trashcan and run to the closest store to buy an nvidia card. As my thanks thanks for your support for radeonsi, i just bought an radeon 280X OC yesterday(not for fglrx i really feel like making it clear, all fglrx do is make me regret choosing AMD very viscerally with lots of bad words).

                      System:
                      AMD FX 6350 OC 4.4G
                      AMD 970FX MSI MB(fucking hate MSI bioses btw, avoid like the plague)
                      AMD Radeon 280X sapphire dual X OC(Next week), AMD 7770 Ghz edition currently
                      16GB DDR3 1866
                      2 x 256 GB Sata3 SSD Vector 3 in Raid 1

                      In resume, keep at it with radeonsi because beyond few trolls(is phoronix after all, the trolls nest, what do you expect?) we really love it and is the pretty much the only usable driver for linux for your products, beyond the few lucky guys that use the few things fglrx do better and nothing else(is a mystery to me how they can tolerate it more than few mins, my respect to those)

                      Comment

                      Working...
                      X