Complex OpenGL 3D issue, varies with driver

mathog · 07-30-10, 04:32 PM

The Swiss PDB Viewer application may be used to view solid renderings of molecules, and to spin them around in real time. All this worked swimmingly back on XP, about 4 years ago, the last time I set up a bunch of systems. However, the solid 3D rendering of this program on W7
on newer hardware was horrible

http://www.nvnews.net/vbulletin/show...86#post2246686

After much experimenting on XP SP3, I now think it is mostly a driver issue, and not an OS issue. Perhaps some of you can make more sense of the results than I can.

Here is how to reproduce these tests exactly:

1. On an XP SP3 system install Swiss PDB Viewer 4.0.1 from: http://spdbv.vital-it.ch/
2. Download and save the PDB molecule 2r9r from here:
http://www.pdb.org/pdb/download/down...ructureId=2R9R
3. Configure the Graphics window to be 1280 x 1024 and the highest supported frequency
(on the Quadro FX 1400 it was 85 Hz., and for the GeForce it was 75 Hz, with the monitor I was using).
4. Start SPDBV, move the toolbar to the upper left of the screen.
5. Open the 2r9r file downloaded at (2) and expand it so that fills the rest of the display down to the top of the windows taskbar.
6. The molecule looks a little like a dragon, align it so that the dragon's mouth is to the right,
it's eyes up, and it's body horizontal.
7. Prefs -> Rock and Roll
Angle steps: 5.0, frame delay 25ms, uncheck the reverse box.
(The application will remember this setting)
8. Display -> Render in Solid 3D
9. Press the ">" key on the keyboard, which will start the molecule spinning.
10. Time how long it takes for 10 rotations, starting/ending when the "dragon's mouth" is facing out of the screen.

Here are my results from two different Athlon 3700+ single core machine and Quadro FX 1400 graphics cards (except for OS, machines are identical, except also for one test using a different graphics card which is rated about 3X faster). The tests were performed in the order shown, and that turns out to be significant, at least on XP:

Code:

OS: XP SP3, Machine 1
Test  drv Ver  drv Date     Avg time/rot
0.     8.1.6.7 11/04/2005   4.2s
1      258.96   7/19/2010   deathly slow, long periods of no application response
2.     191.87   2/09/2010   deathly slow, long periods of no application response
(device manager, uninstall card, let XP pull a driver in from Microsoft's site)
3.     178.13   9/17/2008   4.9s
4.     182.65   5/01/2009   5.1s
5.     190.38   7/23/2009   6.0s
6.     191.00   9/22/2009   5.6s
7.     191.78  12/23/2009   5.9s
8.     191.87   2/09/2010   5.9s  (Note test 2!)
9.     197.03   3/19/2010   5.9s
10.    258.96   7/19/2010   9.0s (Note test 1!  Visible pauses in rotation, jerky movement)
11.    197.03   2/09/2010   deathly slow, long periods of no application response
(device manager, uninstall card, select 197.03 driver from disk)
12.    197.03   2/09/2010   deathly slow, long periods of no application response
(device manager, uninstall card, let XP pull a driver in from Microsoft's site)
13.    178.13   9/17/2008   4.9s (Note test 3, at least THIS is repeatable)
14.    197.03   2/09/2010   deathly slow, long periods of no application response

Os W7, Machine 2
15.    258.96   7/19/2010   5.3s (visible pauses, GeForce 9800 GT card)
16.    258.96   7/19/2010  13.6s (often jerking, long periods of no application response)
17.    197.03   3/19/2010   9.0s (with visible pauses)

Os XP SP3, Machine 2 (clones from Machine 1)
18.    8.1.6.7  11/04/2005  3.1s

If Task Manager is run while the molecule is spinning it shows the CPU pegged for the SPDV application and about 13M of memory in use. If the molecule is observed to jerk or stop spinning the memory use starts going up. If it stalls for a long time it can climb to 100M or above.

Subjectively tests 3->9 were all about the same, even though the driver seems to have slowed up a bit somewhere near 6/2009. However, it is really disturbing how iffy the installs are. Also, and this bit me repeatedly, after driver installs the driver likes to default to the largest possible resolution, which in this case was much too high for the monitor, so on many reboots it would go from the windows start up screen to a no video signal, and I had to resort to F8 and a 640x480 start resolution to recover.

Thoughts?

Edit. Cloned the XP SP3 partition from machine 1 to machine 2. After some difficulty getting the proper motherboard drivers installed, and a dozen or so reboots, obtained Result 18, the fastest time yet.

nvgl · 09-16-10, 03:18 PM

Hi mathog,

I'm an OpenGL driver engineer @ NVIDIA, thanks a lot for the detailed report! We have repro'd the bug and we will work on a fix for the stutter problem. In the meantime activating the "3D app - visual simulation profile" should give you more consistent times.

In the future I recommend you to cross-post any OpenGL problems you may have at http://www.opengl.org or any driver problem at the NVIDIA forums (http://forums.nvidia.com), we normally watch those forums more closely (no offense to nvnews!).

07-30-10, 04:32 PM	#1
mathog Registered User Join Date: May 2006 Posts: 90	Complex OpenGL 3D issue, varies with driver The Swiss PDB Viewer application may be used to view solid renderings of molecules, and to spin them around in real time. All this worked swimmingly back on XP, about 4 years ago, the last time I set up a bunch of systems. However, the solid 3D rendering of this program on W7 on newer hardware was horrible http://www.nvnews.net/vbulletin/show...86#post2246686 After much experimenting on XP SP3, I now think it is mostly a driver issue, and not an OS issue. Perhaps some of you can make more sense of the results than I can. Here is how to reproduce these tests exactly: 1. On an XP SP3 system install Swiss PDB Viewer 4.0.1 from: http://spdbv.vital-it.ch/ 2. Download and save the PDB molecule 2r9r from here: http://www.pdb.org/pdb/download/down...ructureId=2R9R 3. Configure the Graphics window to be 1280 x 1024 and the highest supported frequency (on the Quadro FX 1400 it was 85 Hz., and for the GeForce it was 75 Hz, with the monitor I was using). 4. Start SPDBV, move the toolbar to the upper left of the screen. 5. Open the 2r9r file downloaded at (2) and expand it so that fills the rest of the display down to the top of the windows taskbar. 6. The molecule looks a little like a dragon, align it so that the dragon's mouth is to the right, it's eyes up, and it's body horizontal. 7. Prefs -> Rock and Roll Angle steps: 5.0, frame delay 25ms, uncheck the reverse box. (The application will remember this setting) 8. Display -> Render in Solid 3D 9. Press the ">" key on the keyboard, which will start the molecule spinning. 10. Time how long it takes for 10 rotations, starting/ending when the "dragon's mouth" is facing out of the screen. Here are my results from two different Athlon 3700+ single core machine and Quadro FX 1400 graphics cards (except for OS, machines are identical, except also for one test using a different graphics card which is rated about 3X faster). The tests were performed in the order shown, and that turns out to be significant, at least on XP: Code: OS: XP SP3, Machine 1 Test drv Ver drv Date Avg time/rot 0. 8.1.6.7 11/04/2005 4.2s 1 258.96 7/19/2010 deathly slow, long periods of no application response 2. 191.87 2/09/2010 deathly slow, long periods of no application response (device manager, uninstall card, let XP pull a driver in from Microsoft's site) 3. 178.13 9/17/2008 4.9s 4. 182.65 5/01/2009 5.1s 5. 190.38 7/23/2009 6.0s 6. 191.00 9/22/2009 5.6s 7. 191.78 12/23/2009 5.9s 8. 191.87 2/09/2010 5.9s (Note test 2!) 9. 197.03 3/19/2010 5.9s 10. 258.96 7/19/2010 9.0s (Note test 1! Visible pauses in rotation, jerky movement) 11. 197.03 2/09/2010 deathly slow, long periods of no application response (device manager, uninstall card, select 197.03 driver from disk) 12. 197.03 2/09/2010 deathly slow, long periods of no application response (device manager, uninstall card, let XP pull a driver in from Microsoft's site) 13. 178.13 9/17/2008 4.9s (Note test 3, at least THIS is repeatable) 14. 197.03 2/09/2010 deathly slow, long periods of no application response Os W7, Machine 2 15. 258.96 7/19/2010 5.3s (visible pauses, GeForce 9800 GT card) 16. 258.96 7/19/2010 13.6s (often jerking, long periods of no application response) 17. 197.03 3/19/2010 9.0s (with visible pauses) Os XP SP3, Machine 2 (clones from Machine 1) 18. 8.1.6.7 11/04/2005 3.1s If Task Manager is run while the molecule is spinning it shows the CPU pegged for the SPDV application and about 13M of memory in use. If the molecule is observed to jerk or stop spinning the memory use starts going up. If it stalls for a long time it can climb to 100M or above. Subjectively tests 3->9 were all about the same, even though the driver seems to have slowed up a bit somewhere near 6/2009. However, it is really disturbing how iffy the installs are. Also, and this bit me repeatedly, after driver installs the driver likes to default to the largest possible resolution, which in this case was much too high for the monitor, so on many reboots it would go from the windows start up screen to a no video signal, and I had to resort to F8 and a 640x480 start resolution to recover. Thoughts? Edit. Cloned the XP SP3 partition from machine 1 to machine 2. After some difficulty getting the proper motherboard drivers installed, and a dozen or so reboots, obtained Result 18, the fastest time yet. Last edited by mathog; 08-20-10 at 06:10 PM.

09-16-10, 03:18 PM	#2
nvgl Registered User Join Date: Sep 2010 Posts: 1	Re: Complex OpenGL 3D issue, varies with driver Hi mathog, I'm an OpenGL driver engineer @ NVIDIA, thanks a lot for the detailed report! We have repro'd the bug and we will work on a fix for the stutter problem. In the meantime activating the "3D app - visual simulation profile" should give you more consistent times. In the future I recommend you to cross-post any OpenGL problems you may have at http://www.opengl.org or any driver problem at the NVIDIA forums (http://forums.nvidia.com), we normally watch those forums more closely (no offense to nvnews!).