Well, the way I see it, nVidia is really the one leading in terms of GPU technology right now. Why? Because looks at the differences between AMD and nVidias' previous and current options in performance. nVidia's GTX 280 is a card with 240 Shader Units, that pretty much matched the 9800 GX2, which had 256 Shader Units (128 x 2).
Now, you have to consider that the 9800 GX2 didn't scale perfectly between the two cores, but it was about equal to the percentage of scaling in the 3870 X2. The 3870 X2 had a total of 640 Stream Processors, from two cores each consisting of 64 Unified Shaders ( (64 x 5) x 2). The 4870 has a total of 800 Stream Processors from 160 Unified Shaders (160 x 5), and had close performance to the 3870 X2.
To put it simply, when you think about them both and compare, nVidia got about twice the performance with about 93% more Shader Units, and a Shader Clock that was 200 Mhz less. AMD got about twice the performace with 150% more Unified Shaders/Stream Processors, and 75 MHz less core clock.
That's not the end of it though. Now, lets compare the 8800 GT and the 9600 GT. The 9600 GT has a total of 64 Shader Units and runs on a G94 core, while the 8800 GT has a total of 112 Shader Units, which is 75% more than the 9600 GT, and runs on a G92 core. On the other hand, the 9600 GT has 50 MHz more core clock speed, and 125 MHz more shader clock speed, but they both have equal memory speed. While we already know that they 8800 GT performs better, when the two cards were put side by side, the 9600 GT was usually less than 10% slower. Could the two cores have that much of a difference? Supposidly, the G94 cores had better ROPs, but even with just that, could the performance really get that close?
One more comparison. Let's look at the 3870, the 4670, and the 4830. The 3870 and the 4670 both have 320 Stream Processors, from 64 Unified shaders. The 3870 has 25 MHz more core clock, twice the amount of ROPs, and roughly twice the bandwidth, while the 4670 has twice the TMUs, and usually comes within about 10-15% of the 3870s performance, but under a lot less power. In my opinion, that's good. However, the 4830 has 640 Stream Processors, from 128 Unified Shaders. It has twice the TMUs then the 3870, and about equal bandwidth. But, it has 200 MHz less core clock speed. In comparison, it's only about 20% faster than the 3870 on average, and about 40% faster than the 4670. Now, this may be because of the massive clock speed difference, but I'm not entirely sure it's just that.
So, what do you think about all of this?