No announcement yet.

Initial impressions of ATI's drivers...

  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by pfunkman View Post
    Have you tried playing wow in openGL? I know for at least wheni played wow with an ATI card in wine the performance was much much better through openGL.
    Yup. I did try directx, but I ran into a fatal flaw -- the 3D parts of the screen were drawn upside down. (Yeah, I know, that's a pretty weird bug.) It's specific to turning on the "Full Screen Glow" effect, which I think just enables bloom.

    I turned off "Catalyst A.I.", and things have improved some (not so many crashes), but in big open areas with large effects, I still get a lot of random brightly colored polygons in front of other stuff sometimes.


    • #17
      Originally posted by monraaf View Post
      aticonfig something. Can't remember the exact command, but just running aticonfig without any arguments dumps a list of valid arguments.
      Thanks! Looks like --odgt is the right magic. Stable around 35 degrees C while just poking at firefoix. I'll leave it running every few seconds for a bit, log in, and see what it does.


      • #18
        Okay, followup. I've had some video glitches, so I ran aticonfig in the background every 2 seconds for a while. I was getting those glitches when the card reported a temperature of about 43 degrees. Haven't seen anything much higher. That makes sense to me; this card is lower power than the GeForce 9800 I replaced with it, but has a bigger heat sink.

        That said, I'm having one issue that's really a problem for me. With the GeForce drivers, I had a fair number of crashes, but they always resulted in a window being popped up saying that the application was being killed, and then it was killed.

        The typical failure mode for the ATI drivers is that the display locks up, and nothing short of ssh'ing in and rebooting will recover it. I can kill X, and everything stops running, but I still can't cause the video driver to reset.

        Or is there a command for resetting the card, too?

        At least now I know it's not overheating.


        • #19
          Originally posted by seebs View Post
          Or is there a command for resetting the card, too?
          In theory, removing and reloading the kernel module should reinitialize the hardware. In practice, it doesn't work and I ended up rebooting anyway.
          I'd guess that fglrx leaves some kernel structures broken when crashing, thus reinitializing the hardware isn't enough. Reinitializing the kernel means rebooting.


          • #20
            Messing around, I've noticed that in most circumstances, the "GPU load" value is always 0%, even when the card's rendering full-screen 3D and running at its full clock speed. I'm not sure what that means, but I'm guessing that number isn't very accurate. ("aticonfig --od-getclocks"). That said, it does look as though the card has PLENTY of spare time -- the issues I'm having with performance look like they're CPU-bound.


            • #21
              I should clarify: *almost* always. Every so often it hops up to 17% or 38% or so briefly. But it'll stay 0 for several consecutive runs of aticonfig while the screen is full of 3D rendering.


              • #22
                You can try Unigine tropics benchmark to stress test your card. It's quite a nice demo that actually works with fglrx and it should get the load and temps of your GPU up.


                • #23
                  You can run Furmark with WINE. It actually works and you can use it to stress the GPU. Use the ATI Overclocking Utility to monitor GPU temps. and load.


                  • #24
                    Oh, this is interesting.

                    When the display freezes... The GPU is at "99%" utilization, and just stays there, with low CPU usage from WoW. Nothing I can find at this point recovers it. "aticonfig --help" says something about a set-powerstate and list-powerstates, but they don't seem to exist anymore.

                    I sure wouldn't mind something to the effect of "aticonfig --youre-wedged-reset-now".


                    • #25
                      ... Oh, that's just too interesting.

                      It can come back from screenblank when wedged, but it doesn't unwedge. Very odd.


                      • #26
                        Originally posted by Melcar View Post
                        You can run Furmark with WINE. It actually works and you can use it to stress the GPU. Use the ATI Overclocking Utility to monitor GPU temps. and load.
                        Yeah, I was actually running this again the other day but in Windows...

                        My GPU load according to ATI Tray Tools overlay was at 99.8% and temps maxed out right around 79-80C (once the nb fan kicked into high gear).

                        Under linux I run the Unigine demos though and unfortunately never bothered to write a script to log the GPU load & temps, but it should be easy to do an infinite loop and call aticonfig for the values and output to a temp file...)

                        (Was having those annoying display driver has stopped responding but has recovered, and accidentally seemd to have discovered that by changing Realtek audio from 48kHz 24b to 48kHz 16b that the problem suddenly went away in one game at least...)

                        fglrx: 2010 has gotten off to a crap start is all that I can say, but I'm hanging in with 10.2 since 10.3 should be out soon and here's to hoping that it's actually a good one this time around.

                        GW char selection screen is still garbled for me, and it was so bad this time that I just give up until 10.3. i.e. I didn't even bother to test ingame and this was (pure as the driven snow)wine 1.1.40 + 10.2 + x86-64 Ubuntu 9.10 + 4850.


                        • #27
                          In WoW, my GPU load pegs at about 30-40% unless the card wedges, in which case it wedges at 99% and nothing ever happens until I reboot the machine.

                          Which suggests that my biggest problem is that the game's currently CPU-bound. Bleh.

                          On the other hand... Total crashes per day, definitely lower than with nVidia. The only issues I have are horrible graphic corruption with distant lighting effects (or something similar to that, anyway), and the periodic hard hangs that require a machine reset to unstick the GPU.

                          The temperature never gets bad at all; I've yet to see it above 45 or so, maybe 46. (This is a passive-cooling card with a ginormous heatsink, and the case fan happens to be blowing air over it...)


                          • #28
                            Mine is actual a notebook, MSI GT725 w/a Radeon Mobility 4850 HD 512MB GDDR3.

                            The heatsink is linked with the CPU heatsink which has the actual fan, generally I idle around 47-52C dependent upon time of year, etc. and as far as I have seen Furmark is one of the most GPU stressful apps that I've run and is around the temps that I've seen in many reviews for high temps w/this card(well the desktop variant anyways). So, it seems to be OK for a notebook given the heatsink setup.

                            When I was testing a mod for Oblivious, I wasn't even hitting 70C and IIRC the max GPU load that I ever saw was around 75% but generally it was like your WoW report closer to 40% or so. (The only reason I started tracking all of this was because of that atikmdag has stopped working annoyance trying to see if the GPU was overheating, powerplay doing stupid things, etc. but in the end it seems that the integrated sound just doesn't work well at 24b with Catalyst apparently, something which I didn't have a problem with on my desktop with a similar integrated sound chip plus nVidia GPU, but with an AMD CPU(4800+). Anyways this is going way offtopic since this isn't a windows forum and the same problems aren't likely to apply.


                            • #29
                              I think I'm gonna give up.

                              I've done a bunch more testing, and I'm having two major categories of bugs:

                              1. Hard hangs, hard enough that I have to hit the physical reset button. Couple-few times a night.
                              2. Random polygons shearing across the screen, which comes and goes.

                              No, the card's not overheated; it's never made it above 46C. I dunno if the problems are in WINE or in Catalyst, but this is simply not stable enough. With the nVidia drivers, I could frequently play all day without crashing. Sometimes it'd get into a state where it crashed a lot, but at least it would be predictable -- if it was doing that, it'd go poof within 10 minutes. This stuff, I can't tell whether I'm gonna be able to play for ten minutes or two hours, and the constant flickering of stray polygons is pretty unlivable.

                              Any notion of when the next drivers will be out, and whether they're likely to make a difference? I would love to see improved stability here, the card is otherwise pretty nice.


                              • #30
                                9.11 was the last stable driver for me. 9.12 would lock up intermittently, and 10.1/10.2 would lock up often.

                                10.3 will probably be out next week or so, worth giving a try.