Uhhh...I don't mean to disagree with you but try to keep your facts straight. Fermi was in no way anything like Bulldozer. Sure, it ran hot and loud, but it also grossly outperformed the previous architecture. What, you think a GTX 470 BARELY beats out a GTX 275? lolwut

You might remember that Fermi wasn't actually designed for desktop use. Nvidia's desktop part (GT 212) was cancelled for a variety of reasons. The Fermi architecture was originally conceived solely for GPGPU use in the enterprise/HPC sector as Quadro/Tesla SKU's - as evident by the amount of die space set aside for compute functions. The GF 100/110 were also destined for TSMC's aborted 32nm process, and whereas AMD/ATi had a smaller die/smaller power envelope, a better understanding of foundry process, and a great deal of experience with the 40nm node with which to mitigate a backport of a 32nm design to 40nm, Nvidia could not do likewise, due to market strategy and executive intransigence/lack of foresight.

Nvidia have also been working with a double edged sword - a large monolithic die strategy to fulfil needs of the pro graphics markets that can pull double duty as a gaming GPU. Inevitably there will be design compromises...but if that direction was so wrong, why is it that AMD's new GCN architecture now embodies elements of compute functionality that Nvidia have made virtually standard?

Now, as for Nvidia's "Bulldozer"....remember that like BD, Fermi is aimed primarily at the server/ws/hpc markets...

...You see Bulldozer claiming anything close to that number from Xeon ? (Hint: AMD's present server marketshare is 5.5%. That isn't tipped to rise greatly-if at all- in the face of competition from Intel's Xeon E3, E5 and E7 range of processors

...except for

...I assume

I'd like to see the possibility of unlocking to a 7970! That'd be interesting.

If I am reading this thread right, there is an appalling amount of trolling/getting trolled.

Would certainly make the 7950 THE value proposition, although I'm guessing that AMD might have learned from the last series, where 6950 sales cannibalized the 6970 market. Judging by the 7970's pricing, it seems as though AMD are looking a little more towards the bottom line this time around.

Colour me amazed as well. This kind of thing never happens in a graphics launch thread.

Just a little heads up for NVidia fan boys. AMD has had the southern islands GPU family ready to launch for nearly 10 month, and have been waiting for TSMC to finish their 28nm nodes. So by the time NVidia have launched their 600 series AMD will have had over a year to design their 8000 series. NVidia is going to be a whole generation behind.

"Every generation AMD have increased performance 40-60% if not more from one flagship to another Nvidia can’t say the same."

That's totally untrue.

GeForce 3 Ti 500 --> GeForce 4 4600 (at least 40-60% faster with AA)
GeForce 4 4600 --> GeForce 5900 U (at least 40-60% faster)
GeForce 5950 Ultra --> GeForce 6800 Ultra (75-100% faster, and 150%+ faster in DX9 games).
GeForce 7900GTX ---> 8800GTX (easily 100% faster)
8800GTX --> GTX280 (at least 40-60% more)
GTX280 --> GTX480 (at least 40-60% more).

I think you are WAY off. Both NV and AMD have been improving the performance of their cards very well. But your logic fails anyway since how much more 1 card improves its performance over itself is meaningless. Since HD6970 sucked in DX11 (tessellation), of course HD7970 improves its performance over it a lot. But vs. the GTX580, it's only 20-25% faster. Good, but not that amazing for a 28nm architecture.

"Just a little heads up for NVidia fan boys. AMD has had the southern islands GPU family ready to launch for nearly 10 month, and have been waiting for TSMC to finish their 28nm nodes. So by the time NVidia have launched their 600 series AMD will have had over a year to design their 8000 series. NVidia is going to be a whole generation behind."

News for AMD fanboy:

1) AMD is going to take almost 1 full quarter to launch the entire lineup of HD7000 series cards.

2) NV still has competitive cards in every price level on the desktop outside of the fastest single GPU. They can still lower prices on their current NV lineup and be competitive.

3) HD7970 is only 20-25% faster than GTX580. AMD might actually need HD8000 series to beat a full-blown Kepler....

4) AMD fanboys don't consider the possibility of an interim stop-gap 28nm Fermi shrink (i.e, 640-768 SPs card).

5) Having a theoretical design completed 10 months ahead of production is not "had the card ready". These new architectures take 3-4 years to develop. Fermi was also ready as a theoretical design probably a year before manufacturing. The only measurement that has any relevance is the date a card is available for sale, not when it was "ready".

5) Launching first is a great strategy to capitalize on higher prices (if you have better performance, to impress investors and meet the needs of enthusiasts who love to upgrade from one fastest card to the next. For everyone else, even when NV launched full 6 months later after HD5800 series, it was easily able to reocver with GTX460/560/560Ti/570/580 series and currently has almost a 60% market share on the desktop.

AMD fanboys love to make up their opinion as facts that launching first somehow means winning when market share numbers on the discrete desktop size tell a completely different story.

We might see a lower percentage of stable unlockable cards because of the new manufacturing process. They do have the dual bios so I'd really hope to see unlockable 7950s, if not now, maybe later.

Probable.
Difference in shader count in Cayman 8.3% ( 1536 for HD 6970/CaymanXT, 1408 for HD 6950/Cayman Pro)
Difference in shader count in Tahiti 12.5% (2048 for HD 7970/Tahiti XT, 1792 for HD 7950/Tahiti Pro).

Bear in mind that a percentage of the unlockable cards have proven to be long-term unstable due to running the HD 6950's slower 5Gb/sec vRAM out of spec over an extended time. The HD 6970 natively uses higher specced vRAM. The HD 7970 uses Hynix 1.5GB/sec (6GB effective) RAM @ 5500MHz...the 7950 is specced at 5000MHz effective, so it wouldn't be surprising if AMD went with cheaper vRAM to keep costs down as they did with the previous series.

Conventional core/memory overclocking yeilds much the same performance boost without the attendant risks of locking the memory IC's into the ragged edge of their performance envelope (i.e. how many unlocked HD 6950's can further push their vRAM past 1350MHz "stock" HD 6970 speed?)...

HD 7970's (and likely all other 7000 series) have dual BIOS. It seems more like a general feature than an "unlocking" specific one.

I would like to get the 28nm, but can't as the only option is AMD and we don't know what Nvidia will bring up. I will go with teh HD 6950 2GB and wait for the GTX 800 series or the HD 8000 series.