Quote:
Originally Posted by
wizard69 
You can't dismiss the need for more RAM simply because the technology is different I'm mobile hardware. Nor can you dismiss the fact that a 4X increase in bit map sizes affects system bandwidth. In any event we will see how similar the new GPU is to the current ones fairly soon. I'm fairly certain that Apple will have to address GPU performance demands with the move to a retina screen.
I think Apple is pretty much on the ball here. Like I said, Quartz as a concept is nothing to sneeze at. Even on my iPad 2, 1000x1000 pixel lossless PNGs are easily animated in several layers in an app. Certainly Retina places a lot of demand on the GPU, but I think Apple would have it figured out for 2D. As I've said in the past, the iOS render engine is phenomenal. There is a 4x increase in ~final output resolution~ but we'll have to see how Quartz handles a lot of the other aspects. For example, if we have an app that can zoom in on something, that means when you zoom out, you could be looking at 4000x4000 pixel bitmaps in several layers to be rescaled and output to 2048x1536.
Certainly this will need 256MB of GPU RAM at a very, very rough estimate, but it may need no more than that amount of GPU RAM, say 300MB at most.
I think there should be 1GB of RAM in the iPad 2X, with the GPU taking up ~100MB to 300MB of that RAM at the most.
3D games will be limited to 1024x768 with specialised extremely low-performance-impact upscaling to 2045x1536 ~ given it's pure 2x upscaling they can Lanczos or equivalent optimise for that easily.
2D stuff, ballpark, 300MB at most, I don't think by design Apple would want the GPU to eat up any more than that on a 1GB RAM device.
Quote:
Originally Posted by
wizard69 
The architecture for SoC processors is changing rapidly. If Apple sees a unified address space, where both the GPU and CPU are equals on the memory bus is suspect that structures used by the GPU would become more accessible by the CPU. A bit of a moving target if you will.
In this area I have to defer to you because I'm not informed on SoC RAM addressing by CPU and GPU. I would imagine a unified address space is the best way to go since it's all one die, plus allowing a flexible use of system RAM like on a PC/Mac IGP? Dedicated memory for the GPU, or "less flexible" memory for the GPU wouldn't be a good way to go because the GPU might only need, say, less than 100MB of RAM in very basic apps, but certainly more for 2D games, let alone 3D.