How Intel's Nehalem will benefit AMD

Although enthusiastic benchmark claims for Intel's Nehalem have been available for over a year now, the actual product ihas only recently made it into early production servers featuring the new motherboards and firmware needed to make use of technology AMD has had for years but couldn't get integrated into volume server production.

The most important new features in Intel's much praised "Nehalem" technologies are the integrated memory controller, the on board internal bus, and the "hyperthreading" improvements enabled by the level 3 cache - all of which AMD has had for a number of years.

So why do I think this should benefit AMD? Because the people who make motherboards for the PC industry make them to sell -meaning that these things are designed to be made as cheaply as possible and to work with the lowest common denominator CPUs the PC makers are likely to use.

As a result reviewers who wanted to fairly achieve pro-Intel results needed only to put the Intel and AMD CPUs into motherboards differing, at best, only in terms of the sockets used, to seem perfectly fair while silently handicapping AMD's performance.

With Nehalem on the way in, however, the board makers are now starting to produce boards designed to work well with those capabilities - and that will cut into Intel's reviewer advantage.

Now, of course, I could be wrong - and to test that I'd like to offer two short term predictions, one about what we will see, and the other about what we won't see, as production Nehalem servers finally hit reviewer stands.

The first prediction is that the reviewers will do what they always do: compare the new 45nm Nelalems to the older 65nm Barcelonas in functionally identical motherboards to find the Intels faster by just about exactly the margin predicted by the reduction in scale between the 65nm AMDs and Intel's 45nm products.

The second prediction is that not a single one of the major reviewers will mention the miracle plainly happening in the background of their tests: the Barcelonas, plugged into those new motherboards, will lose to Intel while significantly out-performing themselves as reported by the same people in previous iterations.

And, in that same vein, you won't see a single one of these guys apologize for having previously said that Intel uses less power than AMD now that Intel's integrated memory controllers cost it another 12 or so watts.

I'll also make a longer term prediction: as PC makers move increasingly toward system on chip technologies AMD's design advantage will grow in relative importance and visibility.

If I'm right what we'll see in the next year is the working out of the obvious, if rather counter-intuitive, conclusion: as Nehalem forces the motherboard designers to get with AMD's program, Intel will be pushed further and further back against the scaling wall: forced to accept more risk, and spend more money, just to get to large scale 32nm production before AMD consolidates its performance advantage at 45nm.