What reallity?

In fact the HD 6970 consume more power than HD 5870 when in full load... the HD 6970 only consume less than the GPU is in idle due the better power management (there are more power states and the clock is downclocked more when not in use).

And the difference is 10-20% in power consumption.

No. You are right... his example show that the HD 6970 consumes more than HD 5870 using the same 40nm process.

The only thing that can make the Volcanic uses less power is a new process... 20nm or 22nm.

 Volcanic Islands seens to use a revision of GCN 1 (GCN 1.1?) in 28nm... GCN 2.0 only when then 20/22nm get ready for production.

Thanks, for both things.

As I said, I'll wait until the newer architectures to upgrade my card. i'll be the best.

Regardless of the manufacturing process 28nm or 20nm, I can't wait for my HD 9850 :D

My point went completely over your head, but nice attempt anyway.

http://www.anandtech.com/bench/Product/293?vs=294

Performance+die size per watt is better on the 6900 series, note: That includes both the 6950 and 6970.

Because I can use every ounce of Performance I can get. :)
I'll throw my 3x 7970 cards into my AMD machine and sell off the 6950's in that rig which will make up for some of the price.

Yeap.... the permance/watt of HD 6900 is better than HD 5800... there are improviments here but there is no miracle the HD 6900 consumes more than HD 5800.

From your link (Load consumption)...

HD 6970 (880/5500Mhz): 340w
HD 5870 (850/4800Mhz): 310w
HD 6950 (800/5000Mhz): 292w
HD 5850 (725/4000Mhz): 277w

The Volcanic Islands will have better performance but the consumption will be higher too.

The White elephant though is the 6950, both faster (More so after a couple years of driver updates!) and uses less power than the 5870, with a larger die-size on top of it.

Yeah... less clock and disabled units (not all die-size is used) too... a HD 6950 @ 850Mhz could consume close to HD 5870.

The performance/watt advantage was created by VLIW5 ISA vs VLIW4 ISA... VLIW4 have better performance (well at really better efficience that translate in performance).

Big improviments in power consumption just happen when there is a shrink to lower process (eg. 40nm to 28nm).

The majority of the die-size explosion was mostly due to the rather large boost in the Rops and Geometry Engines (Which is where the 6950/6970 really pulls ahead of the 5800 series in games that utilise the extra Tessellation capability.)
The shaders actually use up less die-space on the 6900 series than the 5800 series (Due to the dropping of the 5th ALU and making the 4 other ALU's more flexible leading to higher utilisation of the shaders.)

Plus, if you unlock a Radeon 6950 into a 6970 like I have in my other machine, you will actually see hardly any increases in power consumption, the 6970 used higher core and memory voltages due to the higher clockspeeds, which is where the main drain on power consumption seems to lay.

Granted, if you set Powertune to 20%, Overclock and unlock on a 6950 you will see power consumption skyrocket.

I'm not saying power consumption improvements won't happen with die-shrinks, I'm just saying that there are other ways of making power consumption savings even with larger die-sizes on the same fabrication process as history has told us.