Bolding is mine...
For the desktop, Agena or Kuma is not likely to beat Penryn, and unlikely to beat Conroe. Barcelona's big changes deal with FP and SSE processing and quad-core optimizations, but the integer side is not radically different from K8.
Integer single-threaded performance is likely better than K8, but there is little indication and no definitive statements from AMD that it will be better than Conroe, and a number of hints and ass-covering vague statements to indicate that it will not be.
AMD's all about the platform and scalability, none of which will affect the desktop for quite some time.
Bingo! Some people will perhaps disagree with that, and AMD certainly might not like it either, but IMO there really is no way to repeat that enough. The hardware enthusiast and financial communities keep blabbering about how good Barcelona might (and most likely will be) in some market segments, but in the end, that doesn't even really matter.
Barcelona is not going to double AMD's server market share compared to the one they had in H2 2006. In an absolute best-case scenario, they'll increase it slightly and keep good margins and ASPs there. That's good, but as I said, it doesn't really matter because servers currently aren't more than 40% of AMD's CPU gross profit. So, even if that increases by even as much as 50%, but desktop/laptop revenue halves, they'd still only have 90% of today's gross profit. And given the operating loss they've got right now, that'd be rather catastrophic to say the least.
In the end, what AMD needs is a very strong line-up in all market segments (at least within 5% of Intel at decent margins, and ideally better than that), and I see no evidence whatsoever that Barcelona can give them that. Perhaps I am wrong. Heck, I hope I'm wrong for AMD's sake and for the industry's sake. But annoyingly enough, I very much doubt I'm wrong at this point in time.
Here's a simple way to think about it: what does a dual-core Barcelona have that Penryn, or even Conroe, doesn't have architecture-wise? Besides for minuscule things that are really implementation-dependent like TLBs and most likely balanced properly in Conroe (and, if they weren't, you'd expect them to be in Penryn!) there really isn't much that's better than what Conroe proposes.
The IMC is good, but sadly I think it's pretty much the only advantage AMD will have against Conroe for desktops/laptops. On the plus side of things, and unlike with K8, they won't have as many disadvantages. Also, FP performance might be good, but it matters less than in servers, and it
shouldn't be overestimated.
In the end, if Intel decides to release an unlocked high-end dual-core Penryn, it could be clocked at 3.43GHz+, and there's no real way for AMD to keep the desktop performance crown then. I'm also not very bullish about their 2008 laptop platform (Griffin) because it'll be 65nm-based and that Penryn looks to be a formidable competitor in that area.
Either way, I hope that AMD's management realizes that their problems are deeper than they might first appear. It's one thing to say that you understand your problems, but it's another to actually understand *all* of them rather than just the most obvious and/or problematic ones in the short-term.
Also, I sure as hell hope that their 2009 roadmap is more than a souped up Barcelona derivative with an on-chip GPU, because that'd a recipe for disaster against Nehalem. I believe it will also be interesting to see if Intel releases a Penryn derivative with a smaller die size. Looking at the
die shot, it would be rather straightforward to halve the L2 cache and create a new ~80mm2 chip for the value segment to replace Conroe-L.
Anyway I guess it's fairly obvious that I'm not too optimistic (read: I'm extremely pessimistic!) about AMD's 2007/2008 CPU roadmap. They might do OK in the server market, but that's only one small part of the picture, and it is a massively insufficient revenue source for the company as a whole. I think it's about time everyone, and especially the financial community, started to realize this.
P.S.: And obviously this opinion is too opiniated and perhaps subjective for me to ever write something about it in an article or a news post, which is why I decided to just do a rather long forum post instead...
