As I've said before, it puzzles me that people would think that AMD would do 512-bit for just a checkbox. But it *really* puzzles me why people would think that AMD would do 512-bit *and* GDDR4 for just a checkbox.
I don't think anyone is suggesting
that. But just because the designers had a good reason for including a 512-bit bus in the design that doesn't necessarily mean that it will be useful in real-life. Designers make mistakes sometimes.
By way of a counter-example, think about Intel's "Netburst" CPU architecture. It has a 31-stage pipeline. Now, before it launched, everyone could have been sitting around speculating that the chips were going to overclock all the way to 10GHz, and saying "if the chips can't clock that high, why would Intel have made the pipeline so long? They would hardly have done it just as a checkbox feature, would they? There must be a
reason why it's 31 stages. And apart from insane clock speeds, what else could it be?"
There was, in fact, a very good reason why the pipeline was that long: Intel genuinely intended that the architecture (after a couple of process shrinks)
would ultimately go all the way up to 10GHz. However, once they actually had the chips in production and started to ramp the speeds up, they hit a problem: the chip was less energy-efficient than they'd anticipated, and, in particular, process shrinks had far less of an impact on heat-production than they'd expected. So, while Netburst chips can, in fact, hit 8GHz or higher with off-the-scale custom cooling solutions in the laboratory, in normal use they were never going to get anywhere near that, because they ran too hot.
This meant that the official clock-speeds never got anywhere near what Intel had originally intended, and the 31-stage pipeline was therefore (with hindsight)
totally useless. However, by the time everyone had realised this, Intel was already committed to the Netburst design and it was too late to do anything about it.
Given that R600 is at least six months late (or will be by the time it launches) ATI has clearly encountered significant problems of some kind, which may or may not have been entirely resolved. Either way, it is perfectly possible that, if R600 had worked as ATI originally envisaged, the 512-bit bus
would have been immensely useful, but that,
in real life, the bandwidth been rendered pointless because of practical limitations that the ATI engineers didn't anticipate (in the same way as the Netburst pipeline depth was rendered pointless by problems the Intel engineers didn't anticipate).
I'm not saying this
is what has happened; but it's a
possibility. It's quite right to say that ATI would not have included a 512-bit bus just for the hell of it; they must have had what seemed like a very good reason at the time. But it is wrong to assume that this decision
must turn out to be correct, even with the benefit of hindsight. Engineers make mistakes sometimes.
