Odd thing is that Intel - the other whipping boy of graphics acceleration - also does tile-based rendering. In many respects it's quite superior to the more traditional modes; a tile-based renderer with some real muscle behind it would be an nice thing to see.
As far as i know, Intel licensed PowerVR tech for their integrated graphics.
Makes sense as bandwidth is even more crucial when cpu and gpu share the same memory path as PowerVR chips don't just use tiling but also deferred texturing and eliminate the need for z-buffering if used natively with running infinite planes instead of z-buffered polygons.
I still reactivate my Dreamcast (PowerVR 2) from time to time and it's still a thrill to see what this tiny piece of hardware with only 8 megs of video memory and just 1.4 gigaflops can do.
The japanese town featured in AM2's Shen Mue (1999) is heavily loaded with tons of different textures, architecture and npcs.
Considering what pc games were capable of at that time when running with similar video memory installed, this is very impressive and definitely groundbraking.
Maybe nobody ever tried to build a tile based renderer with brute force power as this type of hardware could satisfy the market for a very long time reducing the need for buying new hardware.
It's better to deliver small steps in power gain than large ones if you want to achieve maximum profit...
What I find interesting about that link is how well developed the concept of megatexture was even at such an early stage - not even half a year after the release of Quake 3. What's even more interesting is that the software renderer in Quake 1 used a surface caching system that essentially is megatexture. All of this happened well before the current console generation too, and totally blows away the whole "megatexture is designed for consoles" slur.
Carmack was and still is a rendering tech nerd (in a positive sense

) so i'm not really surprised.
Absolutely right that Rage's technology has got nothing to do with current consoles.
It's just about time to use different methods for rendering to push the limit if rasterization a little bit further.
The time when Quake 1 was made (1996!) software rendering was common and hardwired graphics accelerators like the 3Dfx' Voodoo were new.
Innovation slowed down when graphics acceleration was standardized to z-buffered polygons.
Before that, we had sprites, voxels, planes, polygons....
This leads me back to what i meant by fast but dumb gpus limiting programming flexibility.
In software mode, when using the (fully flexible) cpu, there's certain methods you can use for rendering you just can't do when going for maximum performance with accelerated graphics in (less flexible)gpus that are limited to a certain technology.
When looking at Ati/Amd's problems with megatexture, it's absolutely clear that speed optimizations can lead to incompatibilities.
Btw.:Megatexture is memory hungry. Rage is the largest game installation i've seen to date.
If Rage was made for consoles, which have always been limited in memory and storage space compared to the PC, it would have turned out significantly smaller in size.
21 Gigabytes to install and using these etremely large texture sizes shouldn't be a problem for modern PC systems but definitely for the heavily limited consoles.
So it can't be made for consoles. Otherwise, Carmack would be a total idiot...