Hark back to the introduction and you'll see us blather on about our imagined G84 and how it compares to what eventually arrived. Revisiting that, we expected what was essentially half a G80 in the shading core, with four clusters and their associated shadering and sampling abilities and rates, and two of G80's ROP partitions to satisfy the needs for a cheap 128-bit memory bus while maintaining decent performance versus G73 and friends.
With half what we expected in the shading core, but with a small boost in sampler addressing power, we feel slightly robbed. Harsh? Maybe. Would such a configuration be doable in a decent sized mid-range GPU at 80nm, given they aren't using NVIO and have added video processing logic? Probably not, and using large chips in the mid-range isn't NVIDIA's cup of tea, so it seems we misjudged it a little.
What's arrived is a GPU with performance to comfortably see off GeForce 7600 GT on paper, but one which we feel will struggle against other current $199 parts such as 7900 GS and ATI's Radeon X1950 PRO. We're sure that'll be shown in game performance measurements, both here when we get round to it and elsewhere today, as NDA expires. Obviously, the architecture also has some advantages, so it will win certain benchmarks, but given the specs, it seems completely impossible for it to be a clear-cut victory against the parts it is really competing against today.
Del med dine venner