Original URL: http://www.theregister.co.uk/2006/06/10/sgi_plans/

SGI chief outlines roadmap out of mortuary

Yesterday's Google has plans for tomorrow

By Andrew Orlowski

Posted in Servers, 10th June 2006 01:22 GMT

Dennis McKenna has been in one of the technology industry's most thankless jobs for five months. No, he's not in Itanium marketing, or even Wikipedia's press officer. The CEO has the task of nursing one of Silicon Valley's most storied names, Silicon Graphics Inc., out of bankruptcy. And as you might expect, he reckons the only way for yesterday's Google is up.

McKenna was in London outlining his plans this week, and he dwelt tantalizingly, and briefly, on some of the issues we've highlighted recently. Particularly Rick Belluzo's brief trail of destruction through the company. But more on that in a moment.

In the early 1990s, SGI was an icon of technological omnipotence. It was feted by politicians and lauded in the popular press as the technology company that had the potential to do anything, and be everywhere. It was simply cleverer and more stylish than any rival, we were told, and had the profits to prove it. SGI filed for bankruptcy protection last month, and its Mountain View offices are now occupied by its successor in the media iconography, Google.

So where did SGI go wrong, then, Dennis?

"SGI made three acquisitions - spent a lot of time integrating them, and they all were sold... They bought high and sold low."

SGI also abandoned its MIPS architecture for Itanic - although McKenna was careful not to give the impression that he thought IA-64 was a dead end. SGI sold just 230 IA-64 systems in the last quarter, down from 244 a year earlier.

SGI was bullish about Intel roadmaps now, and thought its loyalty would finally begin to pay dividends.

"Opteron has won the last twelve months' fashion show. But people don't make buying decisions based on the past. We think Intel's Woodcrest..." - and brace yourselves, here, dear readers - "... will take Opteron to the woodshed."

(We wish we hadn't heard that either)

"We will see what will happen in two years," he added, sounding a fairly pragmatic note of caution.

It's clear, however, that McKenna sees much more growth in commodity Intel chips than in the boutique 64-bit behemoth. SGI's future is in bundled clusters, tuned for vertical markets.

Which verticals? Archiving and storage, he said. Storage was SGI's best kept secret he claimed, something that's been true for a long time. Selling the kit has always been the problem.

We wondered why with the post 9/11 "pork boom" - government defense contracts focusing on data mining - as well as an added urgency to the business of energy resources exploration, SGI hadn't capitalized on what should have been a promising time in its traditional markets?

SGI had been outdone by cheaper solutions, McKenna said, bluntly. But some customers had tried commodity Linux clusters and just didn't get the performance. Collaborative tools also offered SGI some potential growth.

Now to the interesting bit.

McKenna said he was astonished that SGI had failed not just to capitalize on its rich IP portfolio, but had waved the white flag as soon as the shooting started.

"I am surprised by SGI's reluctance to defend its IP," he said.

This we interpreted as SGI's capitulation to NVidia, in 1999. NVidia and commodity Linux clusters became SGI's nemesis - and few staff at the time could understand the logic of Belluzo's decision to turn a surefire court win, and potential licensing bonanza, into a defeat. One can't imagine, say, a Qualcomm surrendering its CDMA assets for a flaky joint venture with a tiny upstart. This isn't how business is done.

McKenna added:

"SGI has a legacy of selling stuff for a small one-time fee. That's not going to happen again," he said.

And this we interpreted to be a reference to the day SGI met a giant, and agreed to give it many of its vital graphics patents. The deal, with Microsoft, was first brought to light here in early 2002, and the net value to SGI was just a handful of beans. Or $62.5m, to be precise.

"We have a hell of a lot of IP left," McKenna insisted. To stop the rot, he said he had lawyers writing letters in his second week.

He confirmed, however, that the OpenGL trademark was up for sale, along with other "non core" copyrights.

When we checked back to the 2001 10Q in which SGI disclosed the Microsoft deal, it also described the IP assets as "non core".

We wish Dennis and his 1,800 staff all the best. But recommend he has a "core inspector" on hand, to be safe.

SGI's history is full of strange echoes like this. Remember that the hasty Microsoft deal was arranged because of a dispute between Nvidia and Redmond over the first generation Xbox that threatened to delays its launch. But wasn't SGI at one time in the business of designing games consoles? Indeed it was.

And why is SGI now excited by a processor based on an Intel core that's twelve years old? The Woodshed is based on the P3 core, which is little more than the Pentium Pro with added marketing instructions. It's undergone a brief, two year emergency surgery to bring it up to date. Yet twelve years ago SGI owned an industry leading RISC chip that was both widely OEM'd and future proof. This was MIPS, which was spun out, and with Itanium on the road map, mothballed. You can speculate on how much advantage SGI might have gained from just two years' worth of similar focus on MIPS.

But back to now, and McKenna said SGI saw the opportunity to spin out of some of its software products and letting them benefit from a smaller organization. McKenna didn't say which, but he said SGI would retain a minority stake.

As for SGI's outlook, McKenna acknowledges the grim reality for the company's shareholders, but sees if not a sunny day, then at least some light.

"When I saw The Register's headline about our Chapter 11 filing, I thought, 'It's better to be titsup than titsdown'," he said.

"They'll use that," a PR minder warned the CEO.

We have no idea what he meant. ®