Things aren’t much prettier on the server front, since Intel’s new Core architecture was in large part designed to address its weakness in this direction, with features like Quick Path Interconnect. Yes, QPI is faster than the Front Side Bus used with Core 2 and there is a benefit to the desktop user, but there are far larger benefits on servers using several CPUs. Each CPU has a dedicated QPI for each other CPU, so they can speak directly to one another, as shown in the diagram to the left.
The FSB was a real bottleneck in these cases, allowing AMD powered servers to compete despite having weaker CPUs. Core architecture replaces the current Core 2 Xeon family in early 2009, and should be at least beginning to gain traction by ’10. This is likely to at least equalize the ground Intel had always lost to AMD’s HyperTransport technology.
The Red Team is planning a new platform for 2010, codenamed Maranello, using CPUs codenamed Magny-Cours and Sao Paulo. It seems to largely be a move to DDR3 and more cores per chip. That may not cut it against the Blue Team’s Core, or by then Sandy Bridge CPUs, particularly if you have to buy a new platform anyways.
Exactly, and we probably haven’t forgotten about that TLB incident that held up the last Phenom yet. In AMD’s favour though, they have pushed a lot of their dates back with the most recent roadmap, so they may be trying to get the delay related disappointment out now instead of later. Not a bad idea if you figure people are more interested in the Phenom II that’s coming out next month at this point in time.
Giving lots of warnings about product delays is also appreciated by shareholders, who will soon be voting on a huge set of changes at AMDfavor. With things looking a little grim in 2010 and unclear after that, it is a good thing that these changes should net the Red Team a lot of cash as they spin-off their manufacturing assets. We look at the proposed spin off, along with its business and technical implications in the rest of the series.