GPU Accelerated Texture Tools

NVIDIA's existing Texture Tools collection is likely a bookmarked link and active part of most 3D developer's tool collections, and parts of it are likely familiar to a good few artists, too. Comprising a compressor (standalone and library), import tools for Max, an authoring plugin for Photoshop, a texture atlas generator, MIP chain manipulation tools and their DDS viewing tools, Texture Tools gets an update in the form of GPU acceleration for the compressor.

We presume that both the library and standalone version of nvDXT, the texture compressor, will benefit from GPU acceleration for better integration into a real- or run-time pipeline or engine, and that it'll retain support for massive textures (8Kx8K) and the texture formats it currently supports.

The core of the compressor is based on Simon Brown's Squish compression library. Simon's licensing of Squish with the MIT license also has the knock on effect that NVIDIA's accelerated implementation will also come with freely available source code so that developers can integrate it more easily into their applications and extend it with new formats and fix bugs themselves. NVIDIA say that requests to extend the compressor are among the most popular when it comes to support for that part of Texture Tools, so the opening up of the source should appeal to many.

This new version of Squish is accelerated using CUDA, and NVIDIA claim greater than an order of magnitude increase in performance compared to their existing compressor running on the CPU for certain texture formats and sizes, and that's compared to ATI's multi-threaded Compressonator and S3's texture quantisation tool. The new compressor also runs on the CPU, where NVIDIA claim a fine speedup there too.



That displayed speedup is paired with a claim that no quality is lost and that in some cases it's actually improved (in terms of RMS error rates). Measuring the performance and quality of the new NVIDIA accelerated compressor versus Simon's latest build of Squish is something we look forward to, given that recent builds have significantly improved their performance when using SSE revisions on the CPU.

With a full source release and GPU acceleration via CUDA, the new compressor should receive significant attention on its release.