That's 100% guaranteed for 2 reasons:

1) Existing high-end GPUs such as GTX680/7970GE use 190-230W of power. The next round is still made on 28nm, which means power consumption is likely to go up to 240-250W. 

2) Next gen's tier of high-end might increase in price from $500 to $900 due to the cost of 28nm wafers + lower yields when fabbing 550mm2 die chips, starting with Nvidia's Titan card:

http://www.sweclockers.com/nyhet/16402-nvidia-gor-geforce-titan-med-kepler-gk110

Sony/MS will not release a console with GPU alone using 200-250W of power on its own and certainly will not put a $500-900 GPU inside a console. 

=================================

Found some benchmarks for 3D Ray-Tracing as I seen some people mention that MS/Sony might target this for games next generation:

"The five second clip with 3D Ray-Tracing that we were rendering took about an hour to run" (http://www.legitreviews.com/article/2127/1/)

That we can squash the Raytracing games rumor automatically.

When someone lists out full specs, they list the total # of ROPs and TMUs a GPU has. Notice how they listed 18 CUs for the specs but then listed 18 TMUs. That's confusing and is not how people list GPU specs. Think about it, using the same logic you described, if someone listed HD7970 as a 32 CUs, 32 ROP, 128 TMU card ==> HD7970 would suddenly become a 128 ROP, 512 TMU part...

Sorry fella,GTX680/7970GE isn't high end in my eyes,it's super high end

I think you are confusing high range with high end. By saying end that technically means the top area. I think we are looking at mid-high range for GPUs right now.

ROPs are not in CU but each CU have 4 TMU in GCN arch... HD 7970 have 32 CUs, 128 TMUs (32x4) and 32 ROPs (not related to CU).

So Orbis.

18 CUs
72 TMUs (or 32 TMU backends... each backends have 4 TMUs)
32 ROPs (or 8 ROPs backends... echa backends have 4 ROPs)

That's how the GCN arch works... there is no other way to get the numbers.

You mean 18 TMU backends. And didn't VGleaks admit they screwed up the labeling on listing the TMU count, so that's why there is so much confusion.

So what are you expecting then? The cards mentioned are reasonably good as it is.

Did you see the GPU update??? There is no confusion because any tech guy undestood what VGLeaks said.

18 Texture units
- 56 billion bilinear texture reads/s
- Can utilize full memory bandwith

56 billion bilienar texture is accounted for 72 TMUs... the GCN archtecture have 1 TMU module per CU but each module have 4 Texture units... that's not hard to understand if you know the GCN arch design.

See? 4 units per GU... so 18 (clusters) x 4 = 72 TMUs.

Yeah I know, but most people didn't and I guess staff had to clarify for people who were jumping to the wrong conclusion. So if anyone is still wondering, even VGLeaks clarified. Plenty of non tech people looking at this info.

I think the VGLeaks just used AMD specification... the AMD refers to ROPs like clusters and not the units.

Yes,yes,like this