Putting me right how?
Like when he utterly dismissed my suggestion that Cayman would have around the same performance-per-Watt and performance-per-transistor as Cypress, saying that it would be "massively MASSIVELY better"? (how did that turn out?)
Or perhaps you mean when he claims that Fermi is only a tiny update to GT200, and that Cypress to Barts is a much bigger update (flying in the face of every technical analysis made, not to mention common sense)?
OMG, here we go again.
Lets see who makes more sense, you insist the 6970 is still the same basic architecture of the 2900xt in 90% of your posts, despite literally nothing being comparable. Then you maybe backtracked a little on your Cayman idea but the 2900xt is still the same architecture as everything up to the 6870.
The biggest changes from the 2900xt to 5870 are simply in the arrangement, modularisation and dx11 features. Of course, these are the SAME changes Fermi has you insist make it an ENTIRE NEW ARCHITECTURE.
So lets see, I say arrangement and modularisation aren't "architecture" per say, its the fundamental ability and speed of the said individual pieces. You say any arrangement change means an entire new generation, but when AMD makes those same types of change with every new core after teh 2900xt, for them its the same architecture.
Take a football team, if Arsenal on monday lined up with the keeper up top and Song in goal, the players were the same, just standing in different places.
The fundamental nature of gpu's and their fundamental speed and ability to process data into something meaningfull is their shaders. This is why EVERY REVIEW has agreed up to the 6870 relies on the same underlying architecture as the 2900xt.
Now again, go and look at architectural diagrams of the 2900xt and the 6870 and everything inbetween, they show MORE change with more large differences than the 8800 to Fermi, yet you are happy to call them the same design, so are ALL the sites who you insist say Fermi is a fundamentally different architecture. So whose right, the guy who insists the same changes on both sides amount to the same architecture for AMD, and a completely and utterly new generation for Nvidia, or, fundamentally, these are not architectural changes.
P4 Northwood and Prescott has pretty massive changes, but their fundamental ability and speed was dictated by the double pumped Alu and dual issue core, this did not change throughout many steppings and several "completely new versions". Performance was still reliant on the underlying actual design of the main component.
Core 2 duo through one of the last couple cores, they are all reliant on a basic same issue core and performance can not pass the level they impose.
In general every new generation of the architecture seeks to improve efficiency in feeding the data to that fundamentally same underlying architecture.
For GPU's this is whats being done, more shaders, efficiency/modularisation improvements every gen to improve feeding the higher number of shaders. The shaders, their layout, their 1 instruction per clock, this is what has fundamentally not changed from 8800 to Fermi, as the 5 way issue is what made the 2900xt the same fundamental architecture as the 6870, while EVERYTHING else in the core is HEAVILY modified.
Again you and everyone else insist, quite correctly they are heavily linked and the same base architecture, there is without question MORE changes to the 2900xt up to the 6870 than 8800 to Fermi.
Cayman IS an ENTIRE new architecture, every last thing in the core has changed, nothing is recycled from the 2900xt at all, there is not one feature that has anything to do with the 2900xt.
As for claiming it will be a massive improvement, it still is, unfortunately many of the changes were "wasted" on GPGPU, with many things CYpress is incapable of, not just faster at(like Fermi) but completely different base capability.
Drivers are also ALL OVER the place. In one review its losing in F1 to the 570gtx, in another, its FASTER than the 580gtx.
Go back to Cayman speculation threads, go on, see how wrong I was.
MONTHS ago I guess 2.6-2.7billion transistors, sub 400mm2(but not by much) and 35-40% faster. Lately the rumours were it had Barts uncore(heavily modified) and expectations for everyone went up, and I've been skeptical, every time I've said how it could be more than 35-40% faster, I've ALSO said IF it includes the same optimisations.
At EVERY STAGE I've quantified that these were made at the same time, just because an optimisation is in Barts doesn't mean its compatible or will work, or will be in Cayman.
So before you go making up crap, you'll find, firstly, you're wrong because your own argument changes if you're talking about AMD or Nvidia. Secondly, my expectations for Cayman were LOWER THAN ANYONES and quantified higher expectations proved to be spot on.
Drivers WILL increase performance dramatically, lots of reviews showed massively different results already. As said [H] puts the 6950 ahead or on par with the 570gtx, and the 6970 basically on par with the 580gtx, ahead in a couple of instances.
Other reviews don't, some are being power tune limited, some aren't, its a complete shambles of a launch, I've been massively dissappointed in both, the launch drivers, the massive array of completely differing results, the lack of any gpu-z monitoring to check clock speeds throughout tests to see if powertune is smegging anything around.
But do try and remember, for no apparent reason(its honestly beyond me why they did this) the 255mm2 6870 is rated at 150W, the 389mm2 6950 is rated at 140-150W depending on which site you believe, and it IS 20-30% faster at a similar power draw.
In tweaktowns not altogether nice review of it, its using 118W less than a 580gtx.
THe 6950 should be a 200W card and the 6970 220-225W, is powertune limiting benchmark performance, why on earth are the power ratings so low for such massive cores, why are reviews so vastly different.
Time will tell what the real story is, today, we don't have it.