Originally posted by: coldpower27
Originally posted by: Acanthus
To the ones arguing with me about the 6800GT upgrade.
I am aware that the GT was higher tier part, but, it has been TWO generations, memory technology and process technology alone have advanced enough to make the same friggin chip on a smaller process with more pipes that would stomp it into the ground for less money. I know I know, DX10 support...
Theres still no excuse for VERY little real gains on the mid range since the 6000 series.
And DirectX 10 functionality is very expensive, so the initial generation has to spend the transistor budget on increase functionality rather then increasing performance.
I would say overall the 8600 GTS if your average the benchmarks out is competitive to the 7900 GS/GT or X1950 GT/Pro. There are some cases of 7800 GT level performance, and some cases of 7950 GT level performance.
Yes it has been quite a long time since the 6800 GT, but keep in mind, that SKU has a die of 288mm2, with the majority of the die dedicated to pipelines rather then functionality, while the 8600 GTS is on the 80nm node which makes it 1/3 the size assuming it had the same transistor count which it doesn't, it is only 160mm2 and most of it needs to be dedicated to functionality rather then processing units.
Process generation technology is not moving quick enough to keep up with the gains Nvidia and ATI are making on the GPU front, if you haven't noticed each generation has had a larger die then the last. As long as your adding alot of fucntionality and not just straight shrinking.
130nm NV30 was 200mm2 give or take
130nm NV40 was 288mm2 give or take
110nm G70 was 333mm2 give or take
90nm G71 was 196mm2 give or take and was an oddball, transistors were removed! and it was a cost reducing generation with no functionality added.
90nm G80 480mm2 give or take
At this rate, I wouldn't be surprised if we reached Dual Core Itanium 2's Montencito's level of 596mm2 within the next 2 generation if the continuing trend is upwards
So since process technology hasn't really been keeping pace and the die "MUST" remain between 120-175mm2 give or take this severely limits how much you can add in terms of transistor functionality and concessions have to be made.
130 nm to 90nm only allows 2x the amount of transistors in a given area more or less and transistors count has increased over 3x from NV40 to G80. So process technology is falling behind.
The 6600 GT was Nvidia spoiling people to regain their brand image for the most part, remember this card came out after the generation of the FX which Nvidia had tough times with, they need a hard winner to regain thier brand image which was weakened a lot that generation.
And if you haven't noticed compared to the 6600 GT, the 8600 GTS is vastly more powerful and feature rich. Not bad considering the die size between these 2 SKU are very similar. And comparing their process generations 110nm to 80nm you see that you can fit 2x the amount of transistors. So you would expect it to be about 2x as fast not factoring that you also need to factor in transistor budget for new features.
If you want a performance increase wait for a generation where very limited feature functionality is added like NV43 to G73. I would think something like Mid Range G9x on the 65nm process if Nvidia adds no more feature functionality will allow Nvidia to concentrate on increasing performance.