How to launch Nehalem when Penryn is doing so well

Intel's CPU transition troubles

WE'RE A MONTH AWAY now from the upcoming Fall Intel Developer Forum in the cool, windy, hilly and beautiful (still love the place -Ed.) San Francisco.

Especially this year, where Intel is pretty much expected to unveil its final Nehalem rollout plans for the initial 4Q '08 rounds: read - DP workstation, HPC and UP gaming parts.

The Bloomfield UP and Gainestown DP Nehalems, both using varieties of the Tylersburg chipset platform, are doing well, and seem to have gained a bit of speed over the past few months too - at least enough to have their launch speeds matching the best GHz launch clocks of the Penryns at the same time, while hopefully providing quite a bit more performance.

At the same time, Penryns have managed to keep AMD at bay when it comes to speed, except in big MP boxen and HPC supercomputing apps, where Opteron's memory bandwidth and Hypertransport still give it an edge till presumably... yeah, till the Nehalems are in.

So, Intel seems to be in a kind of dilemma: start pre-marketing the Nehalems early to extend the mindshare domination over AMD, but at a risk of cannibalising the current healthy Penryn sales; or 'balance out' the rollout, focusing strictly on the high-priced extreme segments first to gain platform recognition without affecting Penryn revenues?

Well, it's not all so clear. If, say, limited "extreme" Bloomfield and Gainestown benchmarks were to surface, with results drastically above the best Penryns, even many mainstream buyers could be impressed enough to consider delaying their upgrades till the mainstream Nehalems are out and about in a year from now. Those sales right now have very little risk of going to AMD, so delaying them is a pure revenue loss to Intel.

Keep in mind also that the QuickPath interconnect and on-chip memory objectively give the least advantage on uni socket desktop machines, where the existing FSB1600 is far from being saturated. Also, in a typical X48 or 790i chipset scenario, the Crossfire or SLI GPUs have fast direct access to very high bandwidth DDR3 memory banks without going through chipset->CPU hops like they have to do on the current AMDs - or future Nehalems.

Then, we have the E0 Penryn stepping coming in, sometime this month probably. Expect performance and power gains over C0 similar to what the famous G0 stepping did for the 65nm generation. Either way, E0 should enable Intel to launch at least 3.6 GHz parts at about the same declared TDP and voltages as the current 3.2 GHz ones. Whether they decide to go ahead with that, or stick with the 3.2 GHz but possibly lower TDP, depends to a great deal on the decided Nehalem launch speed.

I personally feel Intel would have no problems having 3.2 GHz UP and DP Nehalems at launch in 4Q, and, if they really push it, 3.6 GHz limited Extreme parts aren't impossible either at the same time - as far as I've seen in early tests. So, Nehalems should have the launch speeds identical to the top Penryn speeds at the time.

My feeling is: go full steam ahead with the new processors and make things clear at the IDF - the promise was made to have them out this year, and it better be kept, at least for the workstation, HPC and gaming markets which are always the best early adopters.

The sales overlap problem with Penryn, and any resulting Osbourne effect possibility, can easily be removed by placing the initial Nehalems in a price class above the best Penryns, at least until early next year. By that time, Nehalem should be ready for the mainstream.

Does it mean cheaper high-speed Penryns? I hope so. More pressure on AMD CPU lines? Yeah, definitely. More opportunity for AMD GPU lads to bundle pairs of 4870 X2 cards with each high-end Penryn and Nehalem sold? Obviously. And, in absence of Nvidian Nehalem SLI chipset, more "value sales" of Nforce cum GeForce SLI bundles on Penryns? Sure - look at the GTX280 pricing drop for an early taste of it.