Out of VRAM?

1
Dear All,
 
I get an out of VRAM error when using Screen OpenOffscreenWindow. Strangely, I don't get this error on a setup with less VRAM (and the same version of windows and same screen resolution (1900x1200), different version of matlab though).
I actually have plenty of VRAM: 512 MB. This error occurs on MATLAB 2008b, Windows XP3. The other system is Windows XP3 with MATLAB 2007b.
 
Error and info messages with all relevant details in there:
 
------
PTB-INFO: This is the OpenGL-Psychtoolbox for Microsoft Windows, version 3.0.8. (Build date: Jan  9 2009)
PTB-INFO: OpenGL-Renderer is ATI Technologies Inc. :: ATI Radeon HD 4800 Series :: 2.1.8395 Release
PTB-INFO: Using OpenGL GL_TEXTURE_RECTANGLE_EXT extension for efficient high-performance texture mapping...
PTB-ERROR: Texture creation failed or malfunctioned for a texture of requested size w x h = 800 x 800 texels
PTB-ERROR: and at least 4 bytes VRAM memory consumption per texel.
PTB-ERROR: Currently there are already 49 textures, offscreen windows, movies or proxies open.
PTB-ERROR: All these objects consume system memory and could lead to ressource shortage.
PTB-ERROR: My current (rough and probably way too low) estimate is that at least 114.898682 MB of memory are
PTB-ERROR: consumed for textures, offscreen windows and similar objects.
PTB-ERROR: At least 100 MB memory consumed for textures, probably much more. Could it be that you forgot to dispose no longer
PTB-ERROR: needed textures from previous experiment trials (missing Screen('Close' [, texturePtr]))??PTB-ERROR: Another cause of failure could be that your graphics hardware doesn't have sufficient amounts of
PTB-ERROR: free VRAM memory. Try to reduce the precision and/or size of your texture image to the lowest
PTB-ERROR: acceptable setting for your purpose.
PTB-ERROR: Read the online help for Screen MakeTexture? or Screen OpenOffscreenWindow? for information
PTB-ERROR: about how to reduce the number of color channels and/or precision of the texture.
PTB-ERROR: It may also help to reduce general VRAM memory consumption:
PTB-ERROR: Reduce the display resolution to the smallest acceptable resolution or disable the 2nd
PTB-ERROR: display of a dual-display setup if it isn't strictly needed for your study.
PTB-ERROR: Other than that you could upgrade to more powerful graphics hardware (more VRAM) or try to split
PTB-ERROR: your oversized image into multiple separate subimages and draw them sequentially to the screen.
PTB-ERROR: That way you'll be able to draw larger images, albeit at drastically lowered performance and more
PTB-ERROR: coding hazzle.
------
 
Note again that I have 512 MB of VRAM, so I shouldn't run out yet. On an older Nvidia Quadro with less VRAM, i can create plenty more offscreen windows (which I need, each trial will consist of 60--120 frames; no problem on the other setup).
Could this be a driver configuration problem? I have had the same issues with Catalyst 8.7 up till 9.1 (just out).
 
Note that I am not using Screen OpenOffscreenWindow for PTB2 compatibility, but because I need to prerender stuff using Screen DrawDots. To my understanding, Screen OpenOffscreenWindow basically creates a texture that you can draw to; exactly what I need.
 
-> Mario, note a small formatting bug for the error message, a return is missing. This is as it appears in MATLAB, not an error in copying it into the email.
 
Any tips very welcome, I would like to use this machine (only one at home) for development too.
Thanks!
Diederick
2
Hi Mario,

Thanks again for the long reply, I've learned quite a bit from that
about MATLAB's memory use and VRAM stuff.

> One approach to this would be to precompute the
> 'xy' dot arrays before a trial, then use 'DrawDots' in
> a loop to quickly draw precomputed dot sets. For a
> moderate sizes of your dot set, this will probably take
> up much less memory and be faster as well.
>
> You could store about 150000 2D dots for the amount
> of memory a single 800 x 800 color offscreen window uses,
> so that is roughly your break even point at least in
> terms of memory usage, not neccessarily in speed,
> but a couple of thousand dots should be fine.

HA! Good idea. Got stuck on my old style I suppose. I've implemented
it and it runs great. It actually has some advantages for me, I can
pregenerate much longer sequences now.

> There is also the wonderful world of GLSL vertex shaders
> in OpenGL for real power-users for on-the-fly
> computation of supercomplex stuff if you use the
> native OpenGL support instead of 'DrawDots', although
> it involves a steep learning curve...

That would be cool too, but not necessary for now.

Thanks!
Diederick