Originally posted by: Extelleron
Let me start with detailing the specification overhauls seen in the HD 4870 when compared to the previous model, the 3870. The only specification that at this point is not confirmed is the number of texture units. It is either 40 or 32.
actually, nothing is confirmed until NDA is lefted and we have the actual card/spec.
The HD 4870 (clocked at 750/3600) will have 2.41x the shading power, 1.94x the texture power (or 2.41x if it has 40 TMUs), 60% more memory bandwidth, and 0.97x the pixel processing power when compared to the HD 3870 (clocked at 775/2250).
this is once again an assumption, no one knows for sure unless you are under NDA, which then you were not supposed to be talking about it....
additiionally, what is important is the bottleneck and real world performance. at this point, no one knows.
thank you for the informative pm backed up by tons of assumed specs/reasoning, unfortunately, derivatives of an assumption is still an assumption, you do not have facts of its performance, yet.
I, on the other hand, support wholeheartly on AMD to do well, but not blindly so. I would be surprised if 4870 is as just fast as 8800gts, which is already a 20-30% performance gain. if t sells for $199, it would be a better deal than 9800gtx.
So with 2.41x the shader, and 1.94/2.41x the texture performance of HD 3870, we can expect the HD 4870 to be very close to 2x faster than the 3870. Even if we seriously reduce this estimate, remember the 4870 only has to perform 50% faster than the 3870 to equal the GTX 280
all these 2.41x, 1.94x are NOT confirmed.
9800gtx is about 30% faster than 3870.
GTX280 is about 50% faster than 9800gtx.
so the calculation is actually
9800gtx = 1
3870 = .7
gtx280 = 1.5
aka gtx280 is more than twice as fast as 3870.