sunset's gemlog

Eulogy for a Beautiful Disaster - Itanium, 2001-2013

In November 2012, as part of the release of the new "Poulson" Itanium processors, Intel also announced details of Poulson's successor, "Kittson" - and it was exciting. Kittson was supposed to share a socket and uncore with future Xeon-EX processors, and was intended to be built on Intel's shiny new 22nm lithography. Specs were few on the ground, but Intel was making noise about doubling performance generation to generation, as they had with Poulson. Kittson seemed like a perfect fit for the future DragonHawk Superdomes, with a mix of Itanium and Xeon cells, and it seemed like Itanium was getting an extended roadmap and a soft landing, even without an HP-UX port.

Ten years ago today, 31 Jan 2013, Intel cancelled Kittson, and with it, the era of new Itanium development came to an abrupt end.

Kittson's sudden death came in stark contrast to what Intel had been saying less than three months before - that there was a long-term commitment to the IPF platform, with multiple interested OEMs - and revealed the stark truth that Itanium's future roadmap depended entirely on whether HP saw enough profit in their Business Critical Systems division to bother funding new development. RISC/UNIX had been declining for a long time - Oracle effectively terminated new SPARC R&D in Autumn 2017 and cancelled its SPARC M9 at a fairly late stage of development, and Power is likely now in its twilight years - but the combination of persistent mismanagement by HP, underdelivering silicon by Intel, and deliberate kneecapping by Oracle meant that Itanium sales had declined faster than expected. As a result, it got an accelerated trip to a farm upstate, despite having been a four-billion-dollar unit just four years before.

I was 21 at the time, not terribly bright, and had been under the misconceptions that companies actually cared about the tech they sell and that HP was the last of the truly great computer companies. That illusion had been partially shattered when a large number of internal HP emails was released as part of the HP v Oracle lawsuit, and Kittson's cancellation sealed the deal. There is an object lesson here about being a zealot of any corporation. They do not care about what they are selling except as a vehicle for profiting their ownership. This seems very obvious to me now, but it was not at the time - and it is not obvious to most people who really believe that Apple exists to develop revolutionary innovation or that Google exists to further the cause of an open internet.

Five years ago, Itanium got one final new SKU - a minor rev of Poulson, with a ~5% clock bump (from 2.53GHz to 2.66GHz) on an identical die. Rather cynically, this was called "Kittson" in marketing so that Intel and HP could say that they met their roadmap and delivered Kittson. At the end of 2020, HPE ended sales of Integrity systems, and HP-UX goes EOL at the end of 2025. An era has been slowly coming to an end. The customer base is a ghost of its former self - most have migrated, some to Linux, some to AIX, and perhaps some to HPE's Integrity/HP-UX emulator. VMS and Nonstop have been ported to x86. NT/IPF is a distant memory, and Linux support is barely hanging on.

It didn't have to be this way. Even Poulson systems are still fairly powerful today, and capable of performing the same workloads as when they were new. I quite like mine, and actively use it. But Integrity, for all its merits - and it had merits, from the scalability to the hardware reliability features to the excellent virtualization capabilities - failed to spark joy by sufficiently enhancing shareholder value. Even if Itanium was deemed unwanted, HP could have delivered the same featureset on another processor. We know they considered it, with the Kinetic project.

But that would have cost money.