Thanks for the link. Maybe I'll actually get unlazy and try to fix it. :) Just thought I'd check since I see crashes from very oddly-associated applications whenever my box starts doing its wild timer ride. Plus, I half hoped that a solution was commonly known and someone would provide a link. :) 15 minutes of googling 4 months ago didn't turn up anything promising and it's easily worked around in this situation. <shrug>
Thanks again. -Paul (Funny that Thunderbird's spell check still considers "googling" a spelling error. ;)) Wojciech Lewandowski wrote: > Hi Paul, > > There is well known timing issue with AMD CPUs. Have you seen this document: > http://developer.amd.com/assets/TSC_Dual-Core_Utility.pdf > > Btw I have these AMD timing patches installed. But also tried to run without > them. They don't affect the issues I descried earlier. Frankly, I do not > really care about timing precision. My concern is the fact that application > crashes or freezes and I cannot use FBO with multi monitor and multi > threaded modes. All the CPUs these days are multicore so this is a big > issue. Sure I can force SingleThreading but it would mean loosing power of > aditional cores. > > Cheers, > Wojtek > > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] Behalf Of Paul Speed > Sent: Wednesday, April 02, 2008 6:54 PM > To: OpenSceneGraph Users > Subject: Re: [osg-users] Windows / NVidia / multi monitor / muti threading > (/FBO) problems > > > One thing... have you tried with a fresh reboot or had the machine been > running for a while? > > The reason I ask is because I see odd timing behavior on my Athlon 64 X2 > (also running XP). The clearest example of the oddity is querying the > high resolution timers. Occasionally the value jumps around... but only > after the machine has been running a while. This really messes with > some of the timing/profiling stuff in one of our applications and > sometimes causes some odd system stability when running timing sensitive > applications (music recorders, etc.) > > The working theory (and I have not been able to prove it) is that each > core has its own time but that they get out of sync. So if a thread > switches cores the high res timers go crazy. > > It's not just my apps that show this either. CPU monitors, etc. all > exhibit this problem. The monitoring application that came with my > motherboard even shows the CPU speed jumping around on occasion. > > But only after running for a while. Note: I haven't tried bios > refreshes or anything like that. I have no idea if there is a fix for > this issue but thought it might be something to check before condemning > nVidia, etc.. > > I like AMD a lot but my dual core intel does not have this problem. > -Paul > > Wojciech Lewandowski wrote: >> Hi All, >> >> Few threads touched the same subject recently so I decided to share >> my observations and maybe provoke some discussion on the problem. Maybe >> together we can get it fixed. I said "got it fixed" because I am afraid >> that it might be difficult without driver changes. >> >> Problems are related to OSG on Windows XP & Vista with fairly recent >> NVidia boards (6x00-9x00) / dual view / multi threading modes. I don't >> want to discuss single monitor (or horizontal span) issues in this post. >> >> In my testing I went through following GeForce driver versions: 94.24, >> 97.94, 162.65, 169.21, 163.75, 169.28, 169.44, 169.61, 174.74. >> Tests done in DUAL VIEW / MULTITHREADING modes. Athlon62 X2 (DUAL >> CORE). OSG latest SVN from last 2 weeks (2.3.x). >> >> OBSERVED ISSUES: >> >> All pre 169.xx drivers: >> - Statistics does not show GPU stats. Draw timings are growing till the >> moment when framerate hz start to drop. This issue is present with >> number of examples. Use osgviewer dumptruck.osg to reproduce it. >> >> 169.xx and 174.74 drivers: >> - GPU stats seem ok. But osgviewer dumptruck.osg shows complete garbage. >> Things get better when osgviewer is started with --SingleThreaded >> option and Threading modes are later changed with Threading handler. >> >> All driver versions: >> - FBO use problems. Number of examples using RTT cameras (prerender, >> shadow, simulation) show erratic behaviour. Usually one screen is >> incorrect, sometimes apllication freezes. Output console prints FBO >> error messages. FBO apply fails on one screen. Whats interesting both >> screens are correctly drawn for the first frame, error appears in >> consecutive frames. >> >> - FBO problems are also present in SINGLE monitor mutithreaded >> configurations. Examples usually start correctly but freeze or FBO >> textures stop to update when Threading handler changes mode. One >> exception: I noticed that oldest (94.24) drivers tested here are working >> correctly in single monitor mode (no matter what threading mode is >> selected). They are so old that they may be single threaded >> internally. Unfortunately they do show the same erratic behaviour as the >> other versions in multi monitor mode. >> >> All above problems dissapear when examples are run with --SingleThreaded >> option. Most of the problems also dissapear when run with One monitor. >> There are exceptions but they seem to vary with driver versions. >> >> SOLUTIONS (or rather lack of them): >> >> Driver Release Notes, number of posts on the WEB, posts on the OSG forum >> suggest that many OpenGL problems might be related to >> multithreaded usage. There is a number of CPU fixes available on the >> net. Some from microsoft, some from AMD or Intel. NVidia recommends >> registry tweaks or turning off the Threaded optimization. I tried all >> them without success. >> >> Only one method did change the situation. When I disabled second core on >> the CPU, allmost all above problems vanished. In other words I could use >> MULTIMONITOR and MULTITHREADED OSG on one core CPU. In my opinion this >> is the strongest argument for the claim that there are NVidia driver >> isuses with mutithreaded OpenGL. The only case where FBO was still >> freezing on single core machine was when Threading mode was changed to >> CullThreadPerCameraDrawThreadPerContext. >> >> Unfortunatelly, disabling all the cores except one is no solution at >> all. I guess its better to run --SingleThreaded. I am curious what are >> the others' observations ? Maybe someone found some real solutions ? >> Maybe some workarounds could be made in OSG ? >> >> DISCLAIMERS: >> >> - I made all my testing on NVidia boards and have no idea how ATI boards >> would behave in similar scenarios. >> >> - Most of the tests were done on Windows XP SP2 / Athlon 64 X2 / NVidia >> GeForce 7800 GTX. In a number of cases I changed graphics board to 8800 >> GTS or tested on my collegues Vista 64 bit Athlon Quad core or Intel >> Core 2 Duo with 8800 GTS. In none of this cases I have noticed that >> switch to other archtiecture would change observed issues that I had >> with my primary configuration. Thats why I claim that problems are wider >> and repeatable but your results may still vary. >> >> - Attempts to change "Threaded optimization" or "Multi-display/mixed-GPU >> acceleration" in NVidia Control Panel either didn't help or made it >> worse (instant freeze or no display). So I generally keept them at >> defaults: ie Multiple display performance mode and Threaded >> optimization set to Auto. >> >> >> Thanks for the attention, >> Do you have similar or different observations ? Hopefully solutions ? >> >> Wojtek Lewandowski >> >> >> ------------------------------------------------------------------------ >> >> _______________________________________________ >> osg-users mailing list >> osg-users@lists.openscenegraph.org >> http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > _______________________________________________ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org > > _______________________________________________ > osg-users mailing list > osg-users@lists.openscenegraph.org > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
