Texture size

With ease of distribution solved by KTX, we still need to address another major concern: file size. With the need for a unique file per texture per platform, texture storage, distribution, and loading cost requirements can balloon. Whenever you’re sending textures (or any files) from A to B, you’re going to be filling some percentage of that devices bandwidth, whether it be a network device, a CPU/GPU bus, or when transferring between memory and storage devices. When this bandwidth is at capacity, it often becomes the bottleneck that will throttle software performance. This is why reducing file size is vital, as the smaller our data, the more we can push through our various hardware pipelines in a given timeframe.

This is of particular concern to networked applications and live services, such as online games or web-based 3D model renderers. Such applications typically employ data steaming for things such as custom user textures, or changes to textures in real-time. We’re also seeing an industry rise in collaborative features in digital content creation platforms that require tight synchronisation between clients, so this problem is more prevalent than ever.

To alleviate the problem somewhat, rather than managing and distributing textures for each target platform, we can encode our textures into a hardware-agnostic supercompressed format such as Basis Universal. This relatively new format was developed by Binomial (an image/texture compression company) such that it can be transcoded directly on the CPU into the appropriate format. This technology doesn’t require full decompression, making the transcoding process as efficient as possible and retaining the ability to use the format for real-time graphics operations. This simplifies our number of texture formats from the number of GPU targets down to one – a great improvement.