basis_universal VS crunch

The author of the OP also develops Basis, which is designed to take advantage of GPU texture compression formats. GPU formats have a fixed bitrate which makes them efficient to sample but not the most efficient to transmit, so Basis layers on a second level of compression that unpacks into a GPU texture.

https://github.com/BinomialLLC/basis_universal

You might find https://github.com/BinomialLLC/basis_universal interesting :)

Erm, aren't both WebP and PNG rather useless for games? How do you convert those formats into one of the hardware-compressed texture formats consumed by the GPU (like BCx, ETC or ASTC)? Without those you're wasting a ton of GPU memory bandwidth when sampling textures.

(there are some alternatives, like https://github.com/BinomialLLC/basis_universal, or http://www.radgametools.com/oodletexture.htm)

In addition to this, texture compression is optimized for random access and has to use a fixed amount of bytes per each block, but there is a possibility to further compress this during the delivery. This is called the supercompression and Basis Universal is a good example of this. JPEG XL is also based on tiles for parallel decoding, but those tiles are too large (128x128 minimum, compare with 12x12 maximum of ASTC) so it is not yet suitable for GPU as it stands.

Im interested to see what image format specific compressors go gpu. JpegXL, AVIF, WebP... who wants to show up & throw down? Or even just fastpng?

Meanwhile we dont really hear or regard many of the gpu-oriented compression techs. TIL Basis/KTX2 is itself zstd compressed (formerly LZ apparently?). https://github.com/BinomialLLC/basis_universal

You could check out Binomial's basis_universal. After Google bought them, they open sourced it and allowed use for free.

Another alternative you should consider is https://github.com/BinomialLLC/basis_universal

Simple, open-source, standardized and supports every GPU-oriented format around. And, built-in library support for https://github.com/BinomialLLC/basis_universal

Yep. That lib is great. And, similarly https://github.com/nfrechette/acl and https://github.com/BinomialLLC/basis_universal

As an aside, crunch compression is a bit of a red herring here. While it does reduce the size on disk, it needs to be expanded back to regular DXT when it's first displayed, so it won't solve our memory issues by itself.

Have you considered compressed texture formats? Fixed compression ratio (usually 1 byte per pixel), arbitrary random access, decompress the pixels you need on the fly. Can be further compressed for efficient storage closer to JPEG ratios with https://github.com/BinomialLLC/crunch.

After spending some time trying several libraries, I settled with FasTC for now, but it seems to only be able to manage 32 bits RGBA image formats... I am looking into Crunch2 which is basically Crunch with modern CMake.