Texture mapping has traditionally played a key role in real-time 3D computer graphics, where it is used as the primary means for adding realism to the scene. Programmable Graphics Processor Units (GPUs) allow techniques which utilize textures as a generic look-up resource, thereby allowing textures to represent non-visual information about the database area, such as spectral data at multiple wave bands, thermal data, normal displacement maps for improved terrain shading, digital elevation maps for the terrain shape, or material-encoded maps for parametric approaches to providing dense organic scene content.

For large area visual simulations, the total amount of raster information for a database typically far exceeds available computer and video memory. Therefore, the image generator subsystem faces a substantial data management problem where it has to provide and combine heterogeneous resources (textures containing various data formats) to achieve the desired image quality and real-time performance characteristics. The data management (streaming) aspect of large-area coverage is equally important. The approach should be inherently efficient to address the challenges of modern combat training, where sometimes only a limited bandwidth is available for on-demand database streaming.

The paper presents a novel approach, called Virtual Texture, that addresses the challenges of utilizing huge amounts of raster data on the programmable graphics pipeline. The Virtual Texture formulation satisfies the key requirements of managing the data at deterministic real-time frame rates, while at the same time behaving as if it were a regular texture available on any texturing unit. The latter aspect of the formulation makes it possible to utilize multiple Virtual Textures in the context of any GPU-based technique or a specific vendor extension (such as SGI's detail texture), thereby significantly expanding its application domain. The paper includes examples of techniques where Virtual Texture has been used successfully to address simulation demands.