Replacement rides on software selection
3000 world loses a point of technical light

Throwback: When IA-64's Arrival Got a Pass

IA-64 SequelDuring a summer of 15 years ago, the reach of HP's final processor foundation became obvious. Rather than take over the computing world, the project that started as Tahoe and eventually became IA-64 was labeled as an incremental improvement. Hewlett-Packard said this was so while it started talking about IA-64's lifespan and impact. It would be a gradual change.

This story is instructive both to today's migration planning as well as sustaining homesteading of the HP 3000. Processor power doesn't matter as much as a vendor claims. The pass that HP gave IA-64 in 2000, labeling the technology as years away from the datacenter, proved that chips wouldn't make a difference much more. When it comes to chip futures, the only ones that make a difference come from the timelines of Intel. HP partnered with the vendor, but it wouldn't get a marketable advantage out of the alliance.

In July of 2000, not a single IA-64 system had shipped, even though Hewlett-Packard annointed IA-64 as the successor to the PA-RISC chips that powered servers like the HP 3000. PA-RISC performance remains the leading edge of Hewlett-Packard's MPE hardware. But 15 years ago, making the leap to IA-64 processing looked essential to keeping MPE/iX competitive.

In 2000, though, the technology based on Explicitly Parallel Instruction Computing was just being dubbed Itanium. HP's Integrity brand of servers hadn't been introduced, and HP was supposed to be farming out Itanium to niche markets. The vendor's Unix servers, being sold by the same resellers who offered 3000s, ran on the same PA-RISC chips. And those chips were in no danger of being lapped by IA-64.

Up at the CNET website, an interview with HP's Duane Zitzner included a comment from HP's marketing for IA-64. In 2000, IA-64 computers were "a development environment," said Dirk Down. "You're not going to put this stuff near your datacenter for several years."

In the Newswire, we did the translation for a customer base that seemed certain that leaving IA-64 off the MPE roadmap was a bad fork in the road. Zitzner said PA-RISC would still outsell IA-64 for another five years.

His comments explain why few people in the HP 3000 division seem to think of IA-64 as nothing more than a future. In one interview after another, lab experts and general managers praise the new architecture, but point out that it has little to do with meeting customer demands for performance. Now we seem to know why: the stuff won't be ready for datacenter-level performance for years.

While one analyst thought these delays might be a problem, we think they're a blessing in disguise. There's nothing so broken in today's PA-RISC that it must be replaced. And if PA-RISC's successor is still on the drawing board, that lets the 3000 lab focus. Considering how tough it is to staff development labs, nobody's engineering effort needs the distraction of having to build more than one version of an operating system at a time.

IA-64 looks like it's going to have about a 10-year history of being a future at HP, considering that it was first announced in 1994. (Of course, back then, HP was calling it Tahoe, and then Merced, and so on.) Since HP has four more generations of processors in the wings for the PA-RISC line after the PA-8500 rolls out next spring, it looks like IA-64 might have more impact on PowerPoint slides than in any HP 3000 for the next five years. 

Like HP, we were just guessing on when IA-64 computing would be ready to assert itself in the datacenters. We couldn't see a future where HP would lose faith in the 3000 customer and the MPE ecosystem — not any more than HP could see that IA-64 would become more of a boutique for computing instead of the superstore the vendor imagined five years earlier.

Only two generations of PA-RISC were ever produced that pulled ahead of the top 3000 processors. The 8800 and 8900 would both work in what HP was still calling the HP 9000. The 8800 arrived in 2004's 9000s, mostly being the driver for Superdome. The 8900 showed up in 2005's HP servers.

IA-64, when it was called the Merced project, was supposed to arrive by the end of the 20th century and become the replacement for x86 computing. Instead, HP's partner Intel doubled down on the x86 to make Xeon, an extension to IA-32 created when IA-64 took longer than expected. Intel didn't give IA-64 a pass. It passed it by.