By Charlie Demerjian: Saturday, 08 December 2007, 1:22 PM


THE SITUATION SURROUNDING Barcelona is quite puzzling. Not what happened, that much is very clear, but how people are reacting, and the ulterior motives they uncover.

Barcelona is turning into a running joke. From hard SKUs last February at 2.9GHz to badly bugged parts a year later. The roadmap situation has slipped from comical to changing days after release, and that shows little sign of improving. The entire community has gone from eagerly awaiting parts to not believing it even after you have one in your hands.

The situation is bad for AMD. Really really bad. Stepping B1, the supposed release part was revved to BA after B2 was in the oven. B2 is the TLB bugged stepping, and the supposed B3 'fix' won't be out until February in any sort of quantity. Given the track record of the last half dozen steps, the world is not holding their breath. We will see in a few months.

Now, the part that bothers me is why the current furore happened. AMD botched this one badly. For the third or fourth time. Anyone with a memory long enough to remember the rather sad gestation of K8 will have eerie flashbacks. Before that, the last major architecture was Palomino, the less said there the better. The more things change...

Basically, it is the same old mess they are always in, as one person from a much bluer company said, it looks like their tocks are better than their ticks. Amen.

This latest TLB fiasco is only one problem with the B2 stepping, it has two big ones. First lets look a little deeper at what B2 is. BA was the 'release' version, 1.9 and 2.0GHz parts released in September. The reason for the odd nomenclature is that a bug, strongly rumoured to be in the north bridge, was discovered and fixed after B2 was done.

On the up side, B2 raised the speeds available from 2.0GHz to 2.6+GHz with one problem, they couldn't be made. The production ramp was, to be charitable, running behind schedule by a tad, with tad being defined as a couple of quarters. Word is that it will be fixed some time in Q1. Again: maybe.

Did we say one problem? Just kidding, there is that TLB errata as well, that would be the second. That one is distinct from the ramp problem. The TLB problem is fixable, but the fix comes at a rather heavy price, a speed bin or more. Not only can they not be made, but they have to be cut down on top of that.

So, what is the problem? This news blew up again last week, the better part of a month after word of it first hit. It was hinted at during the launch of the chips in early November, and people grudgingly wrote off another quarter for AMD at that point. 2.4GHz parts were killed minutes before launch due to ramp problems. Speed was hammered after the problem hit the news again. Between those two, it became headline news again for no real reason that we can tell.

There have been dire rumours of the chips being recalled, scrapped or shipping stopped. None is true, but the stop ship is the closest to the truth. There are only small quantities of Barcelona out there, and those are still shipping. The thing is that they are only going to very specific customers that are aware of the TLB 'oopsie' and can work around it or better yet don't hit the patched code. Word has reached us that main recipient of these parts are HPC customers, specifically Redhat-based ones. This probably explains why the initial public non-BIOS patch is for Redhat.

In the end, nothing new has come out since mid-November, and now the world is about to end? Nope, not with Palomino, not with K8, and not with Barcelona. AMD squandered its best chance to regain any sort of parity, much less lead, until Bulldozer ships. That is another story though, and one to be told in 2010 or so if all goes well.

The headlines are being fueled by people and companies with very specific agendas. The people, most of them rabid board trolls, are doing their best to keep things fired up. Certain companies are sending out baiting mails to second tier gullible journalists (By this we do not mean the tech people who broke it and who restarted it, but mainstream media who don't understand the technical side) to keep the flames going, and it is turning into a mess.

In the end, what do we have? The usual AMD year-long nightmare launch, a second hosed-launch product, and a rehash of old problems. Add in people who want to see it kept alive for self-interested reasons, and top it off with AMD putting about every possible foot wrong, and you have a nightmare. Wonderful, but really, get over it people.

From this point on, what needs to be done? For AMD it is quite simple: shut up and deliver. Make damn sure your roadmaps are 100 per cent accurate. If this means only putting out roadies that look one month out and taking baby steps for that month, great, that is what you have to do.

Credibility is easily chucked away and very hard to regain. There isn't much to chuck away at this point, so the only way to go is up. When Intel was at the bottom of the Prescott disaster, it promised to deliver roadmaps with 90 per cent or more accuracy. It did. It promised to deliver better chips. It did. It promised to deliver on time. It did. AMD has to do exactly that.

Ironically, just as ATI gets its house in order, AMD's falls apart. AMD kept to a rigorous schedule and hit almost all K8 roadmap points dead on for about four years. It has the engineering skills and discipline to do things right. ATI was a mess a year ago, and now it is completely back on track. Both sides can do, so now they need to simply shut up and do. It is all about the execution. µ

Credibility is easily chucked away and very hard to regain. There isn't much to chuck away at this point, so the only way to go is up.

(Blue numbers mean Intel is faster; green show a victory for the AMD chip)

AMD vs. Intel Performance Summary

General applications - Opteron 2360SE vs. Xeon E5365 - Opteron 2360SE vs. Xeon 5472

WinRAR 3.62 23% faster 6% faster

Fritz Chess engine 24% slower 26% slower

HPC applications

LINPACK 4% slower* 9% slower*

3D Applications

3DS Max 9 19% slower 25 % slower

zVisuel 3D Kribi Engine 7% faster 14% slower

zVisuel 3D Kribi Engine (AA) 2% slower 23% slower

Server applications

SPECjbb (Sun) 28% faster 11% faster

SPECjbb (BEA) 12% faster 12% slower

MySQL 14% faster Equal

* Faster LINPACK binaries from Intel were available at the time that we finished this article.

We still cannot draw any solid conclusion on the server performance of AMD's quad-core as no MS Exchange, SAP ERP, TPC-C, or TPC-H results have been published. In fact, with the exception of the SPECjbb and MySQL numbers in this article, all server benchmarks on AMD's third generation Opteron are MIA. This situation will probably continue for a few more months as most of these benchmark results traditionally come from OEMs and not AMD.