I had lunch with Hector and Randy

AMD HAD A ROUND table lunch with Hector Ruiz and Randy Allen before the Barcelona launch, and there were a few interesting tidbits that came out of it. In no particular order, with comments, here are my impressions.

The first and most important point was the difference between the 2003 and 2007 launches for AMD chips. When K8 was launched in 2003, as Hector said, vendors were hiding behind the curtain not wanting to be seen. I was there and can verify that first hand.

On 2007, there was a bunch of vendors splashing their name all over the event, several showing AMD only product lines, Dell being the big one there. What a difference 1/25th of a century makes, curtains to overflow crowds.

The next one came from Randy when he was asked more or less why quads now, why not a year ago? You got the usual "because customers wanted it that way", followed by some interesting bits. The 1 to 2 core transition was planned before day one, the historical roadmaps show that without a doubt.

For 2 to 4 there were a lot of additional things that needed to be done to the core and uncore. You could have done it in two incremental steps, but AMD said its customers did not want that kind of churn. For the enterprise side, this is quite understandable, but for the desktop, it is a bit more questionable. In hindsight, I am guessing it would make a different choice if they had to do it again.

AMD also promised eight cores in 2009, something that was promised (link) for 2008 not all that long ago. I wonder if this is a product slip, a 45nm slip, or both. In any case, it was a bit of a downer.

The topic moved to virtualisation and what effect that had on CPU sales. The short story is that it does hurt sales, but does not drop them. As you consolidate, you buy 'less more' servers, and pack them in tighter. Once everything is consolidated, demand then grows at the same old rate.

You end up with a sales curve with a flat spot on it, unpleasant, but hardly the end of the world. Anecdotal evidence on my part says that while you might buy less CPUs, the mix tends to be higher spec parts, and more importantly, more socketed systems. AMD would much rather sell you four Opteron 83xxs than eight A64 3800s.

Randy was quite right when he said that most customers have a rather insatiable demand for CPU power, if you give them more headroom with virtualization, they tend to buy more and use it for different purposes. There are several near infinite compute resource problems, the more you can throw at them, the more you are able to do.

When quizzed about how well Barcelona would do, they quoted a 54% faster figure right now with no infrastructure upgrades. If you have 2 cores at 3.0GHz, you have effectively 6 coreGhz. Four Barcelona cores at 2.0 gives you 8 coreGHz. 8 / 6 = 1.33 so 1.54x is not out of line assuming a few improvements with the new CPU.

Changes in the upgrade model were also noted, in the past, people would not upgrade for anything less than a 2x improvement in performance. That was then, now, businesses are doing more of a rolling upgrade with performance upgrades being put into service when they are available.

This is driven in part by the sheer number of servers most companies deploy, and also the rate of change in the x86 world. AMD is focused on the middle of the market, and outliers on the top and bottom, while they can be lucrative, tend not to have the volumes that x86 players need.

The session was closed out on a more legalish note. The lawsuit was brought up, and it was said to be still going on, and not slowing down at all. One difference between this launch and the K8 is that now AMD is in every market segment that Intel is in or at least in with x86 CPUs, there is top to bottom competition. This has lead to wins or customers, vendors, and just about everyone. AMD is now taken seriously in the corporate world, something that was not the case our years ago. µ