Dell defection from Intel is a brain twister

TWO THINGS THAT SOME people missed in the Nvidia Dell press release yesterday - but not us - were the fact that it wasn't an Intel chipset in a Dell box and the release of a true 2x 16x PCIe slot motherboard with a single CPU.

The Dell bit is huge, not because of the fact that it is using, gasp, something non-Intel, but it shows how powerful a marketing message SLi really is. A while back, ATI put that into a quarterly report, begrudgingly pointing out that it was a problem. The new fire lit under its chipset division and Crossfire also shows how critical this whole market segment is to box builders and graphic chip makers alike.

Dell defecting, or at least letting some light in through the cracks shows that even the mighty Dell bows to market pressure. It also shows that Intel has no 32+ PCIe lane chipset in the works for the near future, nor a multi-part chipset with PCIe on the south bridge. The volume isn't going to be high, Intel does not have a competitive gaming CPU, and won't for another year, so this is more a low volume face saver than a threat to Voodoo.

Still, if Nvidia can meet the stringent Dell supply chain requirements, there is an upside to leverage this into something more lucrative. Nvidia seems to know what it has, and is not shy about twisting arms to get what it wants. One has to wonder if this is what ATI is trying to counter than the occasional benchmark loss.

On the tech side, there is also big news. The new chipsets are the more significant thing, and have the potential to outstrip everyone else out there. It is called SLI X16, and it potentially provides a true 2x 16x PCIe slots, plus two 1x lanes, one slot comes off the northbridge, and the other off the southbridge. Unlike the NForce Pro line, the 2200 and 2050 chips, the new X16 does not need a second CPU and a second HT link to work right. In fact, it even works on Intel, getting them closer to being back in the gaming game.

While this may seem like a no-brainer, just put more ports off the southbridge, it is a lot more complex than that. The NB to SB connection has to be a fairly high speed link on the most vanilla chipset. Each PCIe channel is 2.5Gbps, so 16x is 40Gbps of bandwidth, not something you can just slap on because it meets a feature requirement checkbox. By comparison, the IDE, SATA, GigE and sound are a collective bandwidth pittance.

This is a major engineering exercise, you don't do this on a whim. If you screw it up, or don't provide enough bandwidth to the SB, it will be quite visible, literally. So, high pressure, high stakes, and high use of obfuscation. NVidia would not comment on the speed of the link between the NB and SB, so something might be up there, keep an eye on the benchmarks.

Both AMD and Intel motherboards should be out using the new architecture in August, so you won't have to wait long to buy one. A side effect of this is that it puts Crossfire into an also-ran position with only 2x 8x PCIe. The engineering involved says that ATI won't have a competitor out for a while, and Nvidia is again playing it for all it is worth, they thought the whole SLI thing out from the beginning, and it is bearing fruit. Barring any NB to SB fudging, it does not look like anyone will have a real competitor until next year.

What will this do to the prevailing balance of power? If the full 34x PCIe channels are really there, this would make an astounding low end server chipset. It may not have the RAS features of Blackford, but for those that don't need memory retry, it has the ability to add more I/O than anyone else.

Additionally, it may give an advantage to Intel in gaming. OK, the gap in sheer horsepower is to great to overcome, but if you look at the block diagrams, AMD has to push memory traffic over the CPU to NB link. That link is probably pretty strained with 2x 16x PCIe slots running at a high percentage of absolute capacity. Intel passes memory traffic from NB to GPU instead of CPU to NB to GPU. It may be enough to make a difference, we will see when the benchmarks hit.

Overall, we have some marketing, some arm twisting and a lot of engineering. Most of this was lost in the headlines, but it is definitely worth paying attention to. Once again, things get interesting when you least expect them to. Good. µ