Tady jsem jeste nasel krasny odpovedi na stackexchange prave o tom, proc se na GPU pracuje s raw daty (diky cemuz je mozne na ne aplikovat implementation-specific lossless kompresi):
"JPG and PNG files will almost always be smaller on-disk than in memory; they need to be decompressed on-the-fly to acquire raw RGB data"
https://gamedev.stackexchange.com/questions/5171/how-much-memory-does-a-texture-take-up-on-the-gpu
Ty zlepseni v kompresi dat na GPU nejsou v tom, ze by vymysleli nejake super-efektivni algoritmy (typu JPEG). Ty jsou v tom, jak ulozit komprimovana data, aby se zaroven dala rychle cist pri samplovani.