ATI's RD600 shows signs of tottering on pegs

DURING THE last Computex in hot Old Taipei, I had a good chat with the ATI RD600 guys who also demoed a supposedly working RD600 system with a dual-core Intel CPU.

Since, by that time it already had the similarly named Xpress 3200, (codename RD580) AMD chipset with two PCI-E 16 lanes for Crossfire, I was naturally curious what would the new offering bring in terms of PCI-E lane number.

With Intel systems, at least before CSI kicks in, the North Bridge problem is that you have to squeeze in both the dual-channel memory controller and FSB link to the CPU, plus all the PCI-E lanes required for parallel graphics and the I/O links. That could lead to some nasty pin-count ASICs, unless you go the Nvidia NF 680i way and use another bridge over HT to add the extra PCI-E lanes. A man from ATI, which was still not DAAMIT then, had an answer at Computex which was clear. Two PCI-E 16 lanes in those two graphics slots, plus one PCI-E x 4 slot (x 16 connector) for the physics accelerator - that was the time when ATI started pushing the idea of on-GPU physics and demoed the X1900 chips doing it all over the place (but no big news since then on that front).

All that on top of extreme overclocking possibilities - they already talked about successful boots at 2.1 GHz FSB then! - plenty of interfaces, fast memory controller and few more PCI-E lanes for other peripherals.

Then the DAAMIT happened, and so did many other things, including uncertainty over the fate of this chipset, coupled with substantial success and performance crown taken by the Nforce 680i. After all, Intel doesn't have much interest in its chief competitor influencing systems, nor did AMD have much interest in creating a 'too good' chipset for its, well, chief competitor.

So, when, after all the delays and other shenanigans, DAAMIT RD600 finally appeared, I was hoping to finally see their long-promised pole-position competitor to the 680i, and... what a disappointment - the number of PCI-E lanes, at 2 x 8 plus 1 x 4, was the same as on the 2 year old Intel D975XBX 'Bad Axe' mainboard! This definitely won't compete with the Nforce, and frankly it won't stand a chance against Intel's upcoming chipsets few months down the line.

What interest me most is - why such a long release delay since the RD600 chipset demos seemed to work well at Computex and the guys in charge were saying it's about launch ready? Maybe, after the DAAMIT, the 'new boss' re-evaluated the positioning and asked for a revision to, say, halve the number of PCI-E lanes and do a "down differential" vs the offerings for its own chips - and that would need some time? After all, its not as it they really counted on Intel customers for the future chipsets... µ