A Validation Snag
The switch from FBD to on-board buffers in the accompanying platform is what ultimately catalyzed the change in scheduling for Tukwila. On one hand, Intel could have shipped out the Tukwila platform with FBD in 4Q08, but they would have to change the platform eventually to work with on-board buffers. Intel and Itanium OEMs decided that they wanted the Tukwila platform to be compatible with both Poulson (the 32nm successor to Tukwila) and Kittson (a 22nm design). The two main drivers were the investment reasons outlined previously and also providing a smoother upgrade trajectory for end-users and OEMs.
The decision to move the Tukwila platform to on-board buffers was made in the latter half of 2008. Unfortunately, this was also well after Intel and their partners had started system validation.
Given that Tukwila taped in September of 2007, it likely taped out in early 4Q07 and the first silicon would have probably returned at the end of 2007, or early 2008. At that point, the platform bring up and validation began in earnest. So when Tukwila’s platform was retargeted in late 2008, everyone was already several quarters into system validation. The unsurprising result is that system availability slipped by 1-2 quarters to ‘mid 2009’ instead of 1Q09. While large swathes of system validation had to start over from scratch, it turns out that Tukwila itself was just fine.
According to Intel, the delay in availability for Tukwila should not impact the timeline for Poulson or Kittson – which sounds reasonable. While the target date for Poulson is not public, it is not hard to hazard a guess. The earliest that Poulson could be scheduled is the fourth quarter of 2010; which would bring Itanium to 32nm at the same time as the Xeon MP line – a year behind standard desktop processors.
Currently, Intel is competing against 65nm CPU designs from IBM and Sun (not to mention their own 45nm CPUs), with a 90nm Itanium product. The whole point of skipping the 45nm node was to bring Itanium to process technology parity with Xeon MP and the rest of the server world. Thus either the last quarter of 2010, or the first quarter of 2011 seems to be a good guess for Poulson’s target date.
In the mean time, Tukwila validation will go on and in the middle of next year and IPF users and OEMs will have to wait till summer or fall for the biggest improvement in the Itanium line-up since McKinley. Perhaps the biggest remaining question is whether Beckton, the 2010 Xeon MP, faces a similar challenge, as it too had an FBD memory controller. It’s most likely that ISSCC 2009, which is next week will clear up that mystery.
Discuss (224 comments)