Workstations unplugged
Issue: Volume: 31 Issue: 3 (March 2008)

Workstations unplugged

HP workstations are dead. Long live HP workstations.

HP recently made official what we have all long known was inevitable: It announced the end of its venerable PA-RISC workstations. But while HP is bidding farewell to one of its two workstation lines, it is reveling in the success of the other.

The company’s PA-RISC workstations have trickled out of the factory for some time, but HP’s xw line—based on platforms from independent hardware vendors (IHVs) AMD, Intel, and Nvidia—has been on a roll. With a strategic and well-orchestrated effort to not only push its own branded products but develop the overall market as well, HP’s position in the market has been on a steady upswing for some time. Once just a speck in Dell’s rearview mirror, HP is now hot on the market leader’s heels, accounting for 32.7 percent of units in the second quarter of 2007 compared to Dell’s 42.5 percent.

In the third quarter, HP’s x86-based xw line outshipped PA-RISC by a factor of more than 100 to one, so dropping the older line isn’t going to take much steam out of the company’s momentum in the market. In fact, had decision-making been based purely on the numbers, HP might have pulled the plug long ago. But this particular business couldn’t be measured purely on dollars and cents, as there were other strategic concerns—not to mention unforeseen developments—that have led the company to hold onto the PA-RISC line far longer than expected.

We’ve Seen this Coming

Reaching its end of life doesn’t mean the PA-RISC workstation has been a failure. On the contrary: The line served the company well for years, outlasting many of its peers. Born essentially as a client-based spin-off from the minicomputer revolution of the 1980s, the early workstation delivered unmatched deskside performance, leveraging the power of the leading proprietary processors of the time, like PA-RISC, DEC’s VAX and Alpha, SGI’s MIPS, Sun’s SPARC, and IBM’s POWER.

But the emergence of the PC platform irreversibly altered the evolution of the workstation. More than technology, it was the PC industry’s economy of scale that slowly eroded the business model for traditional proprietary platforms. Over time, chips from IHVs, such as Intel, improved dramatically in terms of both performance and price to the point that both traditional workstation OEMs and customers alike had to take notice.

In 2004, HP delivered its last PA-RISC workstation, the c8000. Though competitive, by 2007, it had lost all its price/performance appeal.
By the late 1990s, it was clear to most that the clock was ticking on the traditional RISC/Unix workstation. The marriage of workstation demands and PC technology spawned a machine Jon Peddie Research defines as the PC-derived workstation, delivering PC-based components—or derivatives thereof—in a solution geared to the special demands of workstation professionals.

Accounting for 99 percent of units shipping, the PC-derived workstation today dwarfs the traditional proprietary workstation in volume. Long gone from the market are Alpha from DEC (later Compaq, then HP) and MIPS from SGI. Over time, IBM has relegated the POWER workstation to the support of servers and niche legacy applications. And Sun’s once-dominant UltraSPARC workstation shipments have declined dramatically, now roughly matched in number by the company’s AMD Opteron-based systems.

Itanium vs. PA-RISC

Among workstation pioneers surveying the ever-encroaching PC industry, HP had arguably the clearest vision of the bunch. The company recognized it wouldn’t be able to indefinitely—and unilaterally—keep its proprietary platform competitive.

HP had a few options in front of it. It could fight and take on the daunting task of trying to both broaden the appeal of its platform and entice partners to share in development—along the lines of what IBM has done with POWER (and Sun, to a less successful degree, with its offering). Or it could phase it out, then either adopt an existing high-volume platform or partner with a big PC IHV to develop a new platform. HP chose the latter, a choice that appeared sensible in strategy, but eventually proved flawed in implementation.

Now a virtual afterthought when it comes to client-based computing, it wasn’t long ago that Itanium was set to take over the computing world, workstations included. As most will remember, HP pitched Intel on partnering around a new computing architecture it had been kicking around called EPIC, which would eventually replace x86, PA-RISC, and any other high-performance architecture com­petitors might dare develop.

In a world where backward-binary compatibility wasn’t so important, Itanium may have just succeeded. But it is important, and Itanium’s inability to natively execute x86 was—among other issues—responsible for its inability to catch on in client-focused platforms that didn’t demand the absolute bleeding edge in performance.

It was AMD Opteron’s arrival in 2003 that would ultimately drive the final nail in Itanium’s coffin, at least with respect to workstations. Opteron made a splash introducing a 64-bit instruction set extension over the top of a binary-compatible 32-bit x86 platform and, in the process, pressuring Intel to do the same.

HP and Intel had envisioned the eventual transition to 64-bit computing as the inflection point that would culminate in a mass migration from x86 to EPIC. That transition, in retrospect, certainly turned out to be an inflection point, but instead of signaling Itanium’s rise, the emergence of 64-bit on x86 rather ironically marked the acceleration of its demise.

Once Opteron began turning heads, the few workstation-class ISVs that had Itanium on their road maps promptly dropped the priority on Itanium, or jettisoned plans to support it altogether. And of course, few applications will mean few customers. So not long after EM64T appeared, HP announced that Itanium workstations were discontinued effective September 2004.

PA-RISC Receives a Reprieve

Credit HP for eventually coming to terms with the fact that Itanium wasn’t going to be that promised land for its installed PA-RISC workstation base. Working a succession plan in midstream, HP repositioned 64 bit on x86 platforms—both Opteron and Xeon—as the default destinations for PA-RISC workstation emigrants.

But finding alternative 64-bit hardware solutions turned out to be just part of the complete solution. The 64-bit software had to be there as well, and that similarly proved slow to progress. Yes, ISVs knew they would eventually move to 64 bit on x86 (Microsoft’s plans for Longhorn, aka Vista, told them as much), but with a few minor exceptions, the sense of urgency simply wasn’t there. Longhorn kept moving further out, and Microsoft’s intermediate step of implementing x64 on Windows XP Professional was slipping as well. And to top it off, most bread-and-butter x86 workstation users appeared content with 32-bit capability (most notably those in the CAD industry), with few clamoring for 64 bit.

PCHP ramped up its xw line, which is based on platforms from independent hardware vendors.
Hit first with Itanium’s lack of traction, and second with a slow-to-develop 64-on-x86 software ecosystem, many workstation customers were inclined to stay with PA-RISC for the time being. And they told the company just that, requesting workstation road maps further and further out. The problem for HP was that building a whole new PA-RISC workstation from scratch wasn’t going to be cheap, and the last thing the company wanted to do was invest big dollars on a dead-end business.

The answer—as the other few remaining traditional RISC/Unix workstation suppliers have also found—was to salvage as much as possible from the platforms built and justified for servers. And that’s precisely how HP delivered its last PA-RISC workstation (the c8000) in 2004. Its dual-core PA-8800 CPU and companion Itanium-compatible chipset were already built and being used in server applications, and the graphics hardware came courtesy of ATI: Voila, a new workstation with relatively trivial NRE (non-recurring engineering) costs.

The c8000 was never going to lead in price/performance compared to a PC-derived workstation leveraging the latest components from Intel, AMD, Nvidia, and ATI. But it was relatively competitive and quickly dominated PA-RISC workstation sales, allowing HP to wind down older models. By the end of 2007, the c8000 had lost any shred of appeal with respect to price/performance. But it had served its purpose admirably, staying capable enough to allow HP to keep PA-RISC customers afloat until they could make a smoother transition to a more mature 64-bit platform on x86.

A Fond Farewell

So now the time has come for PA-RISC workstations to officially say good-bye. Vista is here and deployed (though that hasn’t gone as smoothly as hoped, either), Linux is out and has gained acceptance on x86, and virtually every major workstation-class ISV has a 64-bit capable application shipping. Most of the few remaining PA-RISC users have viable places to go, so there’s no point in HP wasting more money keeping the platform alive for another year.

The bulk of those who had already moved off PA-RISC ended up with Windows on Intel. A smaller number—some in oil/gas and the sciences, for example—have moved on to Linux, with many of those initially opting for Opteron. And for those few legacy installations that would just as soon stay on PA-RISC despite its obsolescence (government contracts with long qualification cycles, for example), HP filled the last orders on the c8000 before it is officially end-of-life’d it at the end of this past December.

Looking back, HP was never blindsided when it came to the PC industry’s encroachment of its workstation turf. On the contrary, HP itself ended up spurring on that encroachment with its own selection of EPIC/Itanium as the eventual platform to supplant PA-RISC. Well before Itanium saw silicon, its very existence on road maps helped persuade others to abandon development of their own platforms (RIP: MIPS and Alpha).

But what HP could never predict was how slow Itanium would develop in servers and how it would ultimately be totally abandoned by workstation OEMs. In the end, however, Itanium’s troubles provided only a short-term reprieve for PA-RISC workstations, not a permanent stay of execution. Years ago, HP made it clear that someday the plug would be pulled, and that day has finally arrived.

Alex Herrera is a senior analyst with Jon Peddie Research and author of the JPR Workstation Report series. Now in its fourth year, JPR’s Workstation Report has established itself as the essential reference guide for navigating the markets and technologies for today’s workstations and professional graphics solutions. Based in Tiburon, CA, JPR provides consulting, research, and specialized services for a range of digital media-related technologies and markets. For more information, visit