Intel's Geneseo makes no sense at all

THE MORE I THINK about Geneseo, the more it makes no sense, at least for the next two years. The people at IDF I talked to, the upcoming products, and the general enthusiasm level swirling about are just not giving me the success vibe.

Let's just start out by saying that this article is not about the technology at all, it is about market and timing. I will assume Geneseo does everything perfectly, is better than Torrenza in every way, kicks butt and butters your toast just the way you like it. That said, in this case, tech does not matter. What matters are three things - timing, CSI and partners.

Intel announced Geneseo at IDF last week, and assuming it is fully defined, fully baked and ready for the partners to pick up and run with today, it is still too late. Running around the show floor and talking to people who are likely targets for such a coprocessor, socket, or infrastructure, I found zero people saying they were implementing it. In fact, no one I could find said they had seen anything solid.

Again, let's assume that I missed the dozens of people who have been working on it since the middle of the year, three months or so, for the sake of argument. If it takes them a year to imagine, design, make and write software for the Genesea widget, that will mean at best you will see one in Q3 of 2007.

Let's say that it takes another quarter for customers to understand what you are doing, evaluate it and write the purchase orders. You are now talking Q4/07 before anyone doing this starts getting revenues from it. When CSI, the second of our three horsemen, comes out on X86 in Q4 2008, it will be a far more appropriate infrastructure to crate a Genesea type coprocessor. Call me crazy, but I somehow don't think parts for one will play well with the other.

This leaves you with a one year window to make your money back. If you are a customer, and you simply must buy the latest and greatest, and you truly need to have whatever the hot new Geneseo widget brings, would you invest in the infrastructure? Would a one year cycle make sense?

I can see mainstream customers holding off for the one year until CSI/Nehalem comes out, and then spending the money. If you are bleeding edge, you will probably have racks of Torrenza boxes already. The sales from meagre middle ground is not a place I would want to bet my company on.

Last come the partners. Intel has fewer every day. The argument up until now is that when Intel guesses wrong, if there are no partners, there is no plan B, and no other direction. Intel has systematically shut them out one by one. Just ask yourself who makes a Xeon DP chipset(1) other than Intel now?

Anyone partnering with Intel has to be quite afraid of being shut out, something that Intel can and will do without hesitation if it thinks there is a reason to do so. I am not saying this is the wrong thing to do, or Intel is a bad company for doing so, business is business. The problem is that when you have a track record like this, regardless of how it came about, people are going to think long and hard before they leap in.

Until Geneseo is an open and widespread answer, I'd say partners won't be willing to rush in and spend the engineering dollars to develop parts. Yes, there were partners announced at the launch, yes they will do what they said, and I am assuming it will be a smashing success, but again, would you spend your money to develop for it?

Geneseo is launching into the perfect storm, time to market, time to live, and friends with toys. None of these things are pointing in Intel's direction, and I think that is more than enough to negate any technical advantage, large or small that the platform may have. While I am eager to hear the details, I can't see it taking off until Nehalem .µ

(1) Several smart people pointed out that IBM makes the X3, other people make high end parts, and several make embedded chipsets.