August 29, 2008 5:01 AM PDT

Boxx fills in for a failing SGI

by Peter Glaskowsky
  • Font size
  • Print
  • 5 comments

I miss the old SGI. Silicon Graphics was widely regarded as the greatest computer company in Silicon Valley back in the 1990s. Sometimes forgotten--but not gone--SGI was one of our greatest success stories and one of our greatest tragedies.

Boxx Technologies logo (Credit: Boxx Technologies)

Apple may have had more revenue by virtue of shipping millions of small systems, but SGI's hardware spanned the range from video-game consoles (the Nintendo 64) to workstations to supercomputers. SGI's Unix-based operating system, IRIX, was one of the most sophisticated in the industry.

I used to lust over SGI machines. I'd obsess over lists of used SGI gear, looking for a great deal that would let me have my own IRIX box at home. In 2004, I finally bought an Octane with MXI graphics... but that was years after these machines were effectively obsolete, and I paid less than 0.5% (1/200th!) of the original retail price of the machine.

In the mid-to-late 1990s, SGI was not well managed, losing huge amounts of money because its leaders would not accept obvious and inevitable industry trends. Eventually SGI focused its remaining efforts on Linux-based supercomputers-- which remain some of the industry's best big systems and continue to bring in significant (but declining) revenue.

But SGI's deemphasis of its workstation business left a vacuum of sorts. Major PC OEMs like Apple, Dell, and HP tried to fill this niche with commodity systems, but these machines generally lacked the scalability of SGI's offerings. If you only needed one or two processors and a moderate amount of RAM, you didn't need SGI... but if you could justify spending six figures on a workstation, PC makers couldn't help you. SGI itself sells a few commodity-based workstations, but they're no longer anything special.

Today, I think one company does more to make up for the diminishment of SGI's workstation business than any other: Boxx Technologies in Austin, Texas. Though still limited to the same commodity processors and chipsets that other PC makers use, Boxx at least builds systems that maximize the potential of the available components.

I met with Boxx at Siggraph earlier this month and learned how the company is positioning itself vs. its competition in the workstation and render-farm markets.

Boxx's RenderBoxx and VizBoxx products seem well designed for professional users doing offline 3D rendering and interactive visualization (respectively), without the sense of overreaching ambition that used to pervade SGI's product lines.

It's rare to find a four-processor workstation anywhere (that's 16 cores with modern quad-core processors), but Boxx has both four-processor and eight-processor machines in its APEXX series. If there's another eight-processor, 32-core workstation on the market, I'm not aware of it. The APEXX 8 can take 256GB of RAM and more than 15 hard drives... the kind of numbers we used to expect from SGI. And fortunately, even the biggest APEXX doesn't have a six-figure price tag.

The APEXX 8 is only available with AMD Opteron processors because Intel's Xeon family can't match Opteron's inherent scalability. Although the Boxx people wouldn't comment, I expect to see 8-way systems based on Intel's new QuickPath Interconnect system architecture sometime in 2009.

QuickPath is the most significant feature of Intel's forthcoming Core i7 processors (aka Nehalem), as well as a new Itanium-family chip code-named Tukwila. I learned about these chips at the Intel Developer Forum, and I'll be writing about them here soon.

Peter N. Glaskowsky is a computer architect in Silicon Valley and a technology analyst for the Envisioneering Group. He has designed chip- and board-level products in the defense and computer industries, managed design teams, and served as editor in chief of the industry newsletter "Microprocessor Report." He is a member of the CNET Blog Network and is not an employee of CNET. Disclosure.
Recent posts from Speeds and Feeds
So long, and thanks for all the hits
Wrapping up Speeds and Feeds, part 5: Access
Wrapping up Speeds and Feeds, part 4: Security
Wrapping up Speeds and Feeds, part 3: Ruggedness
Wrapping up Speeds and Feeds, part 2: Reliability
Wrapping up Speeds and Feeds, part 1: Efficiency
Tilera's balancing act: 100 cores vs. market realities
The Gizmo Report: WikiReader--simple, singular
Add a Comment (Log in or register) (5 Comments)
  • prev
  • 1
  • next
by supoman September 8, 2008 6:32 AM PDT
The biggest single mistake they made was switching their workstations over to Windows NT. Instead they should have bought a company like Adobe or Macromedia (both were relatively small back then) and taken over the high end graphics industry. Their machines were already producing most of the special effects for movies back then. It would have been an easy transition. But they waited until studios saw how easy it was to create their own graphics rendering applications in Linux and left SGI the odd man out.
Reply to this comment
by Peter N. Glaskowsky September 8, 2008 11:51 AM PDT
Actually, SGI did buy Alias and Wavefront in the mid-90s, two companies that were in the forefront of the special-effects industry. Adobe and Macromedia might have been better choices in the long run, but at the time they weren't as highly regarded.

. png
by antialiasing November 24, 2008 3:34 PM PST
Peter, if you want an O2, I can give you one for free. I conserved a few when they were being kicked around (in the general direction of the dumpster) during the shutdown of the Mountain View campus.

As for SGI's own "alternative" (well, if a canon can be an alternative to a rifle) to either HyperTransport or QuickPath, check out UltraViolet in the future (or the current Altix). SGI was one of those who believed "if you want it done right, you gotta do it yourself." What this meant was that, in its long and painful history, it became the absolute pioneer twice (*), and absolutely failed to reap the ultimate benefit twice. The result, of course, is devastating. Although my association with the company was in the past, I fondly remember the days when my daily machine had 256 Itanium processors and 256 GB of memory.

* The first time in graphics, the second time in coherent shared memory multiprocessing.
Reply to this comment
by kalyanakrishna February 4, 2009 10:12 AM PST
There are a wide range of comparable systems here -
32 core, 256 GB - http://www.hpcsystems.com/workstationaw800.htm
32 core, 256 GB - http://www.hpcsystems.com/AMDQuadOpteron_A5808-32.htm
16 core - http://www.hpcsystems.com/AMDQuadOpteron_A1403.htm
Reply to this comment
by kalyanakrishna February 4, 2009 10:13 AM PST
Probably the biggest mistake on SGI part was switching to Itanium and betting the shop on it.
Reply to this comment
(5 Comments)
  • prev
  • 1
  • next
advertisement

15 sites that went kaput in 2009

Web sites launch all the time, but they also shut their doors. We highlight 15 that bit the dust this year.

Top 10 news stories of the decade

Let the debate begin: Was the iPhone more important than iTunes? Was anything bigger than Google finding a great business model? CNET offers its list of the 10 most important stories of the '00s.

About Speeds and Feeds

Silicon Valley-based computer architect and chip analyst Peter N. Glaskowsky attends a variety of industry conferences throughout the year to meet with industry thought leaders and dig into the future of computing technology. In Speeds and Feeds, he analyzes trends in system architecture and interface design, as well as market and political pressures surrounding those trends. He is a member of the CNET Blog Network and is not an employee of CNET. Disclosure.

Add this feed to your online news reader

Speeds and Feeds topics

advertisement
advertisement

Inside CNET News

Scroll Left Scroll Right