[3dcenter.org] AMD 8000 series code name (Sea Islands ) specs. - Page 6

Originally Posted by BizzareRide 

No, GK110 is expected to be at least 50% faster than GK104 in gaming. Assuming the 8000-series is 40% faster than a 7970, that means the GK110 will be 20% faster still.
Rumors of the GK110 have been floating around the internet for well over a year now.

Notice it says "GK104 x2." Due to inefficiencies/scaling, it won't really be twice as faster, but with double the processors, 25%+ more ROPs, I expect it to be 50-75% faster than GK104.
Of course some of the data is inaccurate, but GK110 is close to being twice the size of GK104 in compute, processors and die size.
Since when does math == statistics? Statistics are useless if you do not frame the discussion.

Originally Posted by veyron1001 

The problem with Nvidia is that the G110 has been nothing but rumors for over a year. We should expect pics of the ati card around sept. or oct. It does not matter how fast the nvidia card is when its going to be released around 8months later. They havent even released the 660 yet.

Originally Posted by superericla 

And AMD hasn't even released the 7990 yet. Your point is?

Originally Posted by raghu78 

AMD has released no details. so its difficult to say what the real specs and performance is going to be. Nvidia have disclosed few details about GK110. 2880 CUDA cores, 5 GPC, 15 SMX, 48 ROP, 384 bit memory. AMD will focus more on efficiency and price perf. I don't think competing on performance with a 7 billion transistor chip like GK110 is possible. AMD traditionally has not gone for huge dies above 500 sq mm. So you can expect them to go down that road again. Maybe 420 - 440 sq mm at max. Having said that AMD Tahiti does not scale well compared to Pitcairn. Tahiti has 60% more shaders, but cannot outperform Pitcairn even by 40% consistently. Only in bandwidth hungry games like Metro 2033 and Crysis warhead the scaling is close to 50%. In most other games its close to 20 - 30%. Tahiti is definitely limited by front end processing. I think they were conservative with 2 ACE, 2 rasterizers and 2 tesselators given that its their first 28nm chip and a completely new architecture. So if they double a Pitcairn literally with 2560 sp,4 ACE, 4 rasterizers , 4 tesselators and go for 48 ROP they can improve efficiency and scaling. Effectively you are looking at 1.8x Pitcairn perf. Given Tahiti is close to 1.3x, thats a 40% perf improvement. Not bad at all if they can get the chip to run at 950 Mhz with 1 Ghz boost speeds. I am hoping AMD comes out with a more efficient chip than Tahiti. Also going from 2048 to 2560 SP is not a big die cost. shaders are the most densely packed in a GPU. The real die cost lies in doubling the front end and going for 48 ROP. Normally AMD is able to improve on transistor density/packing on a mature process (Cayman had better transistor density than Cypress) . So if AMD can get a double Pitcairn for 420 - 440 sq mm they would have done good.
Another factor is GK110 is a very large chip with 7 billion transistors so it remains to be seen what clock speeds they too can hit. Hitting 850 Mhz is not easy for a chip thats twice the transistor count of GK104. Also GK110 is going to allocate considerable amount of chip level resources for compute which might not make as much of an impact on gaming performance. most of the features added for Tesla/Quadro aren't going to help consumer gaming. So it really remains to be seen what is the actual speeds and performance of the chips. If AMD can be around 15% slower than GTX 780 and provide better perf/watt and perf/sq mm that should be enough to remain competitive with GTX 770 while helping AMD make good margins. A healthy AMD is important for the consumer gaming market because monopolies are never in the best interests for the consumer.

Originally Posted by Clocknut 

Even with the means of having 50% more resources over Pitcairn is already much faster than 7900 and easily beating 680.

Pitcairn is = 1280SP, 80TMU, 32 ROP @ 210 mm2 die size @ 175w TDP
1.5x of pitcairn = 1920SP, 120TMU, 48ROP @ 315mm2 die size @ 260w TDP <-- down clocking it by 50-100MHz could do below 250w TDP easily, which is about 40% increase in performance over 7870, or 30% if u assume it doesnt scale well. (on 7870 it does scale well vs 7770 wheres 7870 is effectively twice of 7770)

IMO 7900 should have been this kind of spec instead of the current imbalance garbage Tahiti GPU. The 7770-7870 architecture are just a perfect match.

Originally Posted by superericla 

Why don't we keep this about what the actual topic is guys.

Originally Posted by bernardblack 

See, this is the thing I wish Graphics card manufactures would get into their skulls. Sit back and R&D existing technology and include those enhancement in the next Direct X revision of cards. There's no need to constantly come out with 3-4 different current generation of DX cards. Look at how much awesome use t he PS3 and 360 see out of their largely dated GPU's. Sure they may not be running insane AA and AF and ungodly resolutions but if those consoles weren't limited in RAM it's be amazing to see what these old GPU's can do.