Its late and making us 40nm guineapigs. Winndows7 is around corner i.e. DX11. these DX10.1 40nm are for 4-6 months. Plus there is no mentioning of Mobility FireGL. it wasnt with 55nm and even not in 40nm. Is M-FireGL dead till DX11. I guess we should avoid it to force them for DX10.1 based FireGL at least.

from 256bit to 128bit, why step back in time ATI?

I want 1024bit and 2048bit, now!!! ;-)

More likely, NVidia mocked up the design, while ATI designed theirs right - the result: lower power.

With all this success at 40nm, I am very eager to see what they do with the next desktop product line. Hopefully no more stupid power requirements.

Hell, they could fit two of the HD 4870 X2 cards onto one desktop card!! today. the HD 4870x4 yah baby, give it to me now!

Not even a month ago INQ reported that the TSMC 40nm process was supposed to be very leaky. As a result no performance increase or power consumption reduction was to be expected.

Now we are seeing 40nm parts with the same speed as their 55nm counterparts, but with halved power consumption.

Does it mean that the TSMC 40nm process is all fixed and ready to go?

Does it also mean that we can expect a 100% increase of desktop parts performance?

They chopped the bandwidth to 128bit and still got a speed bump? I guess we'll see how well they stand up once the benchmarkers get at them.

Thingamajic, the 4860 is not able to do more calculations per second because of an increased RAM speed; that has nothing to do with it. The increased rate of calculation, despite the reduction in stream processors, is due to the higher clock speed, and whatever improvements may have been made to the GPU.

You'll also notice that the number of shaders has decreased from 800 to 640. The speed increase is possible though due to the use of GDDR5 (4860) vs. GDDR3 (4850) memory.

This may be an interesting process, if the 8% claim and the chart provided are true, that 8% performance gain comes with 130 million LESS trannies.