Or it may not have to do either of those things. Do recall, though, that the old iPad had enough filtrate and performance to do AAx2 at 1024 x 768. If it could do that much, a 2x performance increase means it should effortlessly be able to render the same games at 2048 x 1536, since AAx2 is almost the equivalent of quadrupling resolution on an offscreen buffer and then scaling down, except that if the resolution is already 2048 x 1536, then scaling is not necessary.
The only problem is memory bandwidth for pushing around 4x the frame buffer size, but if Apple did include 1GB of RAM, that'll allow for more room to stretch and less swapping, thus less moving.
Do note, though, that those are GPUs of old running older games there. So they weren't doing any post-processing effect, shading, etc...
Not to mention texture resolution might in fact be lower.
With current modern shading techniques, parts of a frame (or the whole frame) need to be preserved in memory to be shaded. If I completely ignore texture size and VRAM usage (assuming system RAM on the Vita can be used for that purpose), then to maintain 60fps at 960 x 540 means that it needs:
960 * 540 * 32 (32-bit color depth) * 60 (60fps) / 8 (there's 8 bits in a byte) / 1024 (kilobyte) / 1024 (megabyte) = 118.65MB
So that means you get only about 10MB of VRAM left for other things like storing shader programs and such.
In reality, that means a developer might be more inclined to use just a fraction that amount of frame buffer, or maintaining just a minimum of 30fps rather than 60fps because resources are so constrained.
This is mostly a problem with modern games doing shading. If there is no shader to be applied, they only need one frame buffer, in which case, it'll look like this for those older cards:
1600 * 1200 * 32 / 8 / 1024 / 1024 = ~7MB
That leaves 25MB or so left for textures, so that means they can store about 25 textures at 512 * 512 size, or 50 textures at 256 * 256.
All things considered... that's actually not a hell of a lot of space, but it's better than on modern games where if they were to render at measly half that res, the VRAM usage shoots up exponentially.
Again, if shaders are involved, then that whole frame buffer thing needs to be preserved in memory so that shader effects can be applied to those separate frames. And you can't just shade a frame and then move on to the next because that'll create huge latency. It's faster to let the GPU render as many frames as it can to VRAM and then let the shaders do their job before pushing the frame to output, otherwise, you're just stalling needlessly.
This is why you keep seeing games on the PS3 and XBox360 running at lower resolution, but in general, the 360 has games that run at slightly higher resolution with more effects than the PS3 because the 360's GPU has some internal buffer RAM that runs at "warp speed" (~200GB/s if I recall correctly, but I guess that means it can be accessed 10x faster than built-in VRAM) that can be used to quickly apply shader effects and push things out, effectively skipping the step that requires those frame buffers to be stored in VRAM. If I recall correctly, it's about 10MB... and 10MB is actually more than enough to store a single 1920 * 1080 frame. In my opinions, with that kind of architecture, the 360 is more limited by shader performance rather than memory, even though it has about the same amount of memory as the PS3.
The PS3, on the other hand, is obviously limited by memory... because there is no where for it to swap things into at faster access speed like on the 360.
What does that tell you about the Vita?
Last edited: