Slashdot Mirror
Reliving The Glory Days of SGI
devin15 writes "Remember in the '90's when the tech boom was in full swing and SGI was the darling of the 3D graphics industry, whatever happened to those days? Wired is running an article about a group for whom the glory days of SGI have not yet gone. From the article:" If the Mac community is dwarfed by the Microsoft horde, the number of SGI users amounts to a rounding error.""
The whole 'UNIX workstation' market is gone.
Sun? SGI? HP? DEC?
Computers became powerful and inexpensive too fast. Clusters killed the big servers.
They're cool looking computers, but in the end that entire stack of SGIs shown in the fellow's home office probaby has about as much power as the Nvidia/AMD box sitting on my desktop.
:-)
A few years ago now, I had access to an old Silicon Graphics machine - a Indigo 2, or something like that. It was quite fun being able to mess around with what had originally been an incredibly expensive machine, and of playing with another UNIX I hadn't used. I even got Blender running on it...
Of course, the machine (well, IRIX) promptly killed itself, and nobody knew the equivalent of the BIOS password to allow reinstallation from the IRIX CDs and bootable SCSI CD-ROM drive we'd spent weeks hunting down. There turned out to be no way of resetting that password, at least not without wiping the MAC address too. Given that the machine was only useful as an X terminal and web browsing machine, it didn't seem worth doing.
Looking inside, at the multi-boarded graphics subsystem covered with huge custom-built chips, it seemed rather sad that even a bargain-basement PC of the time would have massively outperformed it. And now, when I run Half-Life 2 on my current, elderly PC, complete with all sorts of per-pixel shaders and suchlike thanks to its inconceivably powerful (yet obsolete) Geforce 4, I think about how impressed I'd been by a couple of gouraud-shaded polygons...
The only thing I really miss is the screensaver. I forget what it was called, there's an attempted simulation in Xscreensaver called 'stonerview' or similar, but it's nowhere near as good as the original.
Tedious Bloggy Stuff - hooray?
SGI faced the innovator's dilemma big-time; it was tricky to cannabalize their $2 billion workstation business for a $300 million graphics card market. And to move from being a full-system vendor to being a graphics card vendor. And even with all the management and business-issue problems, I noticed three problems their engineering effortsg never overcame:
- trouble with quality and shipping on time (see IMPACT)
- couldn't match/switch from 3-4-year development cycles of the workstation business to 6-month product cycles of the PC graphics card business
- engineers were loath to give up control of the chipset/box/OS in order to settle for just controlling the graphics subsystem. They tried to be a full-system player in a PC world. Given that Compaq couldn't really do it (something that was at least semi-obvious at the time), its not a surprise they, coming from the workstation space, couldn't do it with their integrated NT workstations.
- The engineers were delivering product that was differentiated but not in the areas that the biggest customers cared the most about. The benefits of UMA (unified memory architecture) graphics just weren't in sync with what the market most wanted: the fastest 3D at the cheapest price. And in the classic workstation space, polygon-pushing was what was most needed. Half their business was CAD workstations and in the end they lost that to Sun/HP/IBM who didn't have the sexy texture mapping stuff but could render polygons "good enough".
SGI also benefitted from many years from the other workstation vendors under-investing in 3D graphics. When that era ended, even the workstation business they were in got a heck of a lot more competitive.
Anyway, that's what comes to mind when I remember back to SGI in the mid-90s. In hindsight, I don't know of any silver bullets that would have gotten them out of the situation; it was death by a thousand cuts. At the time, I wondered if a merger with Apple would have made sense but it wasn't clear that the disfunctionality of the two organizations at the time would have melded into something better. Maybe a damn good CEO could have helped them carve out a more defensible role in the industry; that's the only thing that got Apple through as far as I'm concerned.
There are still a very large number of situations where a "cluster" does not work. In some modeling and simulation area's they need a lot of CPU's on ONE OS image.. To make a blanket statement saying that clusters are the answer is either someone lacking experience or bias.. Before deciding what the solution is you have to figure out what you are trying to do..
In our shop we have many large SGI's (128CPU +), IBM Regatta's, Sunfire's, Linux Clusters, and Sun Clusters. They each solve the requirements for the task at hand. We have found that there are numerous areas' Linux is lacking, so for that on our infrastructure support servers we use Sun and HP.
But, here we have some graphically intense stuff we do and they have done all the tests on available x86, IA64, and MAC hardware out there and they just can't do what our current SGI's are doing.. Various vendors have thrown a lot of money in the studies to try and get a larger foothold over SGI.
I think they failed because they didn't sense the change in the PC market. Back in the early and up to mid 90's professional 3D graphics and visualization was synonimous with SGI. I worked for a company that developed one of the major CAD/CAE/CAM product and everyone on their desk had an SGI. If you were a co-op like me you had an older one, if you are the manager you had a R12000 one with 1Gb or ram. In the basement in the "vault" we had a quad R12000 with 4 Gb or ram to crunch huge matrices for CAE. Then around the year 2000 consumer 3D graphics cards and CPUs became more and more powerful and caught up with SGI's products. I could spend about $1000 and get a PC that was 3 times as fast as the SGI on my desk at work which was probably bought for $4000. SGI just couldn't stay ahead of the market and they never lowered the prices to make their machines competitive with PC. I still don't know many people who have or had an SGI at home, they were just too darn expensive.
Another thing is, after the tech bubble burst companies that before had plenty to spend all of the sudden had to cut corners, and one of the corners were the very expensive SGI workstations that could be replaced by Linux boxes or Windows PCs.
I ran old SGI systems for a few years and really loved them. I was never really into the graphics stuff, because most of the software was way too expensive, and the cheap software (Blender) was incomprehensible. But I loved the machines and used them for web stuff. Sure, they were allegedly insecure, but you could tighten them up pretty easily, and nowadays all the breakins are automated exploits of commodity systems. So now I'd say a SGI is a lot more secure than the average system just because they have such a tiny market share.
I thought the operating system and GUI were really slickly designed at the time. They certainly had the most attractive implementation of virtual desktops I've ever seen. Linux has them, but not with the style SGI does and I have to admit that style wins points with me - especially when Linux was still lost in the world of horrible, unreadable fonts while SGI did a great job making them legible and attractive.
But then came Apple and MacOS X, which really showed the world what a truly slick Unix desktop could be like, and I switched almost immediately, leaving my Windows, SGI and Linux machines in the dust. After all, Apple could do it all in one slickly designed system.
I'm sorry SGI never took off; I think they could have been a nice consumer alternative if they could have figured out how to keep costs down. I tried to install Mozilla on my old Indigo2 about six months ago and I got bogged down in dependencies and quit, so it's just sitting in the corner.
People talk about proprietary systems being bad, and the future being in open systems and commodity hardware. And there are bad things about proprietary systems, but I love the spirit that created them, the desire to create something that was designed, not built out of tinkertoy blocks. The desire to create something where the operating system and hardware were built together in one seamless, coherent way.
Because of this, I shed a tear for the proprietary systems, built when men were men, women were women, and computers were something special instead of crudely-designed commodities.
Those days, of course, live on in the Apple world. Which, if you think of it, may be the best of both worlds - the price has been forced down by commodity machines, but it's still very much a sleek, designer experience.
Because after all, that's what I want a computer to be: Something special.
D
No, that's really the opposite of what happened. "Rocket Rick" Belluzzo saw the shift in the market, but he reacted to it in precisely the wrong way. Rather than trying to develop subversive technologies to undermine the PC market shift, Belluzzo decided to try to outsource SGI's workstation business, turning it into just another PC manufacturer.
Right now, companies like ILM are tearing out SGI workstations and replacing them with ultra-cheap desktops. They're taking advantage of the ability to work with low-resolution proxies in real time and then render jobs overnight on the big iron. That's a good workflow for that environment.
SGI should have been their first. They had the big iron --nobody has bigger iron, even now; SGI's supercomputers are more scalable than anybody's. They should have developed software frameworks that facilitate remote rendering of graphics operations. How? I don't know; I'm not a graphics expert. But they should have been first on that block. Then SGI could have gone to a company like ILM and said, "We'll sell you a thousand server processors and a thousand one-processor desktops for five million bucks."
Instead, SGI said, "Fuck the desktop. The server business will boom forever!" Which was a huge mistake.
SGI's failure is that they tried to adapt to the dominant paradigm instead of recognizing its limits and engineering ways to get around them. They reacted instead of created. And they lost vast sums of money in the process.
I write in my journal