Different graphics cards provide different amounts of texture memory. If you’re loading a lot of different textures, especially if they’re large, you can easily consume all of your available texture memory. In principle, this shouldn’t cause problems, as long as you don’t have all of your textures onscreen at once: both OpenGL and DirectX are supposed to automatically evict textures from graphics memory as needed.
In practice, it doesn’t always work this cleanly. In some integrated graphics cards, the “graphics memory” is actually your system memory, so the graphics driver never needs to evict textures–but if you load too many textures, there may not be enough memory left for your application. Furthermore, some graphics drivers have major bugs that manifest as you start to approach the limit of your graphics memory, causing strange rendering artifacts or even crashes.
For these reasons, it may be useful to limit the amount of graphics memory your application uses. Panda provides several tools to help with this.
Automatically reducing textures
You can reduce textures automatically as they are loaded. The easiest way to do
this is to put a line like:
texture-scale 0.5 in your Config.prc file. The
above example will scale all textures by a factor of 0.5 in each dimension (for
an overall reduction to 1/4 of the total memory requirement). If there are
certain textures that should not be scaled, for instance GUI textures, you can
exclude them with lines like:
exclude-texture-scale digits_*.png exclude-texture-scale gui*.jpg
Enabling texture compression
Another possibility is to enable and use Texture Compression, as described on the next page. If supported by your graphics card, this will reduce texture memory requirements dramatically, to 1/4 or 1/8 of the original. There is some reduction of quality, but not as much as the quality reduction you’d get from downscaling the textures by the equivalent amount. It is also possible to enable texture compression in conjunction with texture scaling.
Limiting graphics memory usage overall
Finally, it may be prudent to limit the amount of graphics memory that Panda attempts to use, with a line like:
The above example imposes a limit of 64MB (64 * 1024 * 1024) on the graphics memory that Panda will attempt to use. This can be a good idea to avoid allocating runaway textures on integrated graphics cards with no fixed texture limit, or to work around buggy graphics drivers that crash when you use too much. Panda will automatically start to unload textures when the specified limit is exceeded, even if the graphics driver would allow allocating more.
Ideally, it would be great to query the amount of useful graphics memory provided by the card, and set this as the graphics-memory-limit; unfortunately, this is impractical for several reasons, including the reasons given above. Typically, if you wish your application to work on a variety of hardware, you will need to come up with a handful of default settings and allow the user to select between them, according to his own knowledge of his hardware capabilities.
Monitoring memory usage
You can see how much graphics memory you are actually consuming with the PStats tool. Select the “Graphics memory” option. This graph will show the amount of memory required for active (onscreen), and inactive (offscreen) textures. It also includes memory required for vertex and index buffers, though these are typically much smaller than your texture memory requirements.