I'm testing different compressions for my spritesheets for a game on iOS. In a surprising way, I get a more important memory (RAM) use with PVR 2 bits with alpha instead of a PNG 32 (RGBA 4444). I would say the consumption is 25% higher with PVR 2 bits instead of PNG 32 once the spritesheets are loaded inside memory. I'm using Instruments with xCode to verify the memory use on the physical device (iPad Air 2)
I'm using TexturePacker to generate my spritesheets.
I've read evrywhere PVR 2 or 4 is much less memory consumer than PNG 32. How is it possible ?
Edit:
This is strange because according my observations, PVRTC 4 bits RGBA uses a lot more memory (RAM) than PNG 32, neraly 3 times more according Instruments from XCode. PVRTC 2 bits RGBA is 25% higher than PNG 32 RGBA 4444. I'm talking about live RAM consomption, not disque size which has nothing to do and is not a problem. It seems iOS manages PVR differently than it's supposed to do, especially when loading them into RAM.
Edit2:
My textures are 2048x2048, there are POT and have the square format. Evrything work fine, except the RAM consomption is much higher that it should be. I make all my tests with a physical iPad Air 2 device connected to my Mac with a USB cable. I use Instruments inside xcode to verify and follow the RAM consumption in live. I've solved the RAM consomption problem by switching to a PNG 8 bits (indexed) format with a texture divided by 2 (1024x1024). I make a scale x2 in the code to recover a normal size texture. The RAM consumption droped to 240 MB (PNG 8 bits indexed) instead of 950 MB with (PVR 2 bits RGBA). My game is a video puzzles (with 8 seconds video loops at 15 fps) and uses a lot of sprites. (43 spritesheets in each puzzle generated by TexturePacker, around 130 sprites in each spritesheet)