The Inquirer-Home

AMD announces Barcelona

Many problems remain though
Fri Jun 29 2007, 14:57
AMD IS FINALLY talking about Barcelona, and the news is not good. The situation is complex and messy, but it is getting clearer every day. The short story is 2.0GHz shipping for revenue in August.

That said, you probably knew that by now, just ignore the last two columns in Mike's story and act surprised. The announcement brings up more questions than it answers though, and none of them lead to a positive conclusion.

First let's parse the question of how it was announced, a Friday at 9:01AM EST embargo date. Well, not just any Friday, iPhone ( *YAWN*) Friday, think shiny geek distraction day of the quarter. Toss in that is the day before a holiday week, and you have enough people out of town and not paying attention to bury most bad news.

On top of that, almost all AMD press releases in the last year or six have come out at midnight, even if there is an event that happens later that morning. OK, sometimes it delays it for the event, but from an unscientific look at the releases, it doesn't happen often. Why do it at 9am without an event? Newspapers can't carry it until Saturday and you have plausible deniability. Buried deeper and deeper.

But enough about the PR side, what about the tech? Not much good news there either, 2.0 in September won't cut it, and 2.5 by the end of the year won't threaten Penryn much either. If there was a 2.5 out in May, Intel would not be sleeping much, but I don't think it feels all that threatened now.

AMD's view is that it offers a notable increase in performance over the current Opterons, and that it does, for some software. If Barcelona cores are only equal to the K8 clock for clock, and have little or no scaling losses as you go from 2 to 4 cores, then you end up with about a 33% speed increase with a Barcelona 2.0 over an Opteron 3.0GHz. (3.0GHz * 2 cores = 6 coreGHz, 2.0GHz * 4 = 8 coreGHz). We will ignore single threaded performance for now, the situation there is not pretty.

What you get is better than AMD's current offerings but not much to threaten Intel sales. Q4 will close the gap a bit, but between Stoakley and Penryn, well, that is about all you can hope for. It isn't grim, but it isn't a win either. The launch, when it happens, with the parts it is launching, is simply a better part than they have, not the parts they want.

The state of the tech is a bit more curious. Stepping B0 came back in early April with several characteristics that were notably better than expected. The problem? More testing revealed speed path problems, lots of them. AMD has the headroom for near 3GHz parts, but there are a lot of roadblocks in the way now.

What you end up with is 2.0GHz parts with great thermals, but hard to yield at that speed. At Computex, the show floor was littered with lower end parts, and the labs around Taiwan had the 2.0s working at speed.

The good news is that we hear most if not all of the logic bugs were quashed with B0, so only the speed path stuff remains. The bad news? That is the hardest part.

Stepping B1 came back last week or so, it was not back in the middle of the week before. That means it is about a 10 week silicon to silicon turn time for new steppings, but this is not a hard and fast thing. This puts the stepping after B1 at about the end of August or early September.

One scenario is that AMD puts out spin after spin, each bumping up the max clock one or two bins until it gets to where it wanted to be at launch, and things progress normally from there. If AMD is shipping for revenue in August, this puts the shipping parts as B1.

B0 had lower clocked parts and 2.0s in not enough quantity to launch. B1 seems to move the curve up enough to launch at 2.0, and if the next step adds a few hundred more MHz. With one after that the same, they could well hit the requisite 2.5GHz by the end of the year, barely. Figuring a 10 week turn time, B2 would hypothetically get AMD to 2.2 in Q4 and B3 the rest of the way in time for New Year but probably not Christmas.

The thing is, there are persistent rumoirs that this is not the way AMD is going to go with the chip. Unsubstantiated whispers say that there is a Cx stepping in the works which should fix the speed path problems all at once instead of patching them one by one. This will take longer than the Bx->By steps, but be more effective.

If you look at the dates, Cx seems the much more plausible path to take, and it fits the scheduled releases better. This is where the smart money is betting.

In the end, it all comes down to execution, Intel is, AMD isn't. Instead of taking the lead from Clovertown, it is not going to threaten it. Instead of a horse race with Penryn, they are, well, you get the idea.

Technical problems are just that, and can be solved. What is more worrying is the way it is being announced, bury the bad news as deeply as possible. The problems will eventually be fixed, the race will be on again someday, and this will be about as memorable as the last time it happened (K8 launch), or the time before that (K7 sampling). The more things change... µ


Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

INQ Poll

Heartbleed bug discovered in OpenSSL

Have you reacted to Heartbleed?