Slashdot Mirror
SGI Faces Bankruptcy
Richard Finney writes "
The stock chart tells the story: One time Silicon Valley high-flyer and contender for the Unix crown, SGI stock price dropped 20% on Friday ... deep into penny stock territory ... after releasing fiscal fourth quarter results. The Mountain View, California maker of high end computers is '
exploring financing alternatives with its lender and other sources.' With mounting losses and investors giving ol' Silicon Graphics the thumbs down, things aren't looking good."
How can you say that after all SGI have done for Linux, and open standards in general?
SGI:
- Gave Linux XFS, arguably its fastest and most robust filesystem to date. Far, far more robust than reiser, and quicker than anything else except reiser4 (and then only sometimes), except on deletes where it is slow by design - SGI realised earlier than most that if you need a simple rule, it's pretty safe to assume that people just don't delete files often (excluding short-lived temporary files, which XFS handles _incredibly_ efficiently.) Just check out the low rate at which XFS volumes become fragmented to see how you can take advantage of putting a little thought into deleting files.
- Scaled Linux beyond 32 CPUs for the first time ever. And years later they still hold the record: 1,024 CPUs in the one computer with a single memory space. Nobody else comes close, and I do mean nobody. And this isn't just SGI lab stuff any more - NASA bought 20 of these computers to build the fastest computer on the planet that uses commercial microprocessors.
- Invented OpenGL (hint: what do you think the "Open" in "OpenGL" refers to? bonus marks: compare and contrast OpenGL and DirectX) together with the surrounding (open) glue like GLX. This is pretty much the only reason Linux boxes and Macs have decent 3D, and the only reason you can actually have a decent game of quake even if you're using a dumb terminal. Try playing Quake when connected to a Citrix box. Fun? Didn't think so.
- a bunch of other things I don't know about personally, but here you go anyway.
Anyway, since SGI's main role these days is selling IA64-based supercomputers and workstations, I hope Intel just buy SGI but let them continue to run independently so they can just keep on with all their good work. They provide a useful service to the Linux community, even if you never pay them a cent - this probably has something to do with their current share price (sadly). You might not use OpenGL, Itanium, massive shared memory systems or XFS but the odds are good that at least one of these is helping you, or at least some bugs SGI fixed while getting one of these working.
IRIX *was* way ahead of its time, back in 1995. It had some security issues, but they were eventually fixed too. But that's it folks, aside from new hardware support, IRIX has hasn't had many updates since it recieved the IndigoMagicDesktop and 64-bit support about 10 years ago.
SGI's MIPS hardware went on a similar path. The fastest SGI MIPS CPUs available today are 800 MHz and 1 GHz (and maybe 900 MHz?) these are called R16K but are based on the oldschool R12K design. Still very impressive in terms of performance per MHz and performance per watt, but they are far away from UltraSPARC III, Power5, and even PowerPC and Intel Pentium. The same can be said for SGI's graphics hardware, they lost their competitive advantage over the past ten years to the point where they just started using a bunch of ATI FireGL GPUs instead.
What's their future? Itanium2 and SuSE???
SGI has treated its largest customers well over the years, but those who buy less than $5 million of SGI gear a year have basiclly gotten the finger. Those who buy less than $100 thousand a year aren't even recognized.
Sure, SGI still has some good technology, like OpenGL Performer (which is perfect for multi-GPU simulators and can run on Windows and Linux, thankfully!!) but for the most part, the company is a has-been.
So SGI is about to tank. Does this surprise anyone? I think the only surprise is that SGI has remained in business for the past 10 years.
Gave Linux XFS Scaled Linux beyond 32 CPUs In regards to OpenGL vs Direct 3D, I have heard that D3D has gotten way better since Carmack made those comments. HOWEVER, it's still just Microsoft that controls all of DirectX. But just look at the orgs and people that are on the OpenGL board, even their emails addresses are public: http://www.opengl.org/about/arb/overview.html http://www.opengl.org/about/arb/notes/meeting_note _2004-12-07.html
Or just download a source archive and a documentation archive from http://www.sgi.com/tech/stl/download.html
You're correct that IBM left the PC business (sold the Personal Systems Group to Lenovo last year) but IBM is still making -- and selling -- plenty of hardware. From page 22 of IBM's 2004 Annual Report,
ftp://ftp.software.ibm.com/annualreport/2004/2004($ in billions of US dollars)
Systems and Technology Group 2004: $17,916 2003: $16,469 Yr to yr change: 8.8% zSeries: 14.9% iSeries: (17.2)% pSeries: 7.3%
Almost $18 billion in hardware sales sounds pretty decent. A 14.9% increase in mainframe sales from the year before doesn't look "all but dead", and a 7.3% increase in pSeries (AIX/Linux) machines is more than "a few unix boxes." Especially since Gartner reports IBM leading the worldwide Unix server market last year,
http://www-1.ibm.com/servers/eserver/pseries/newsYou make some very good points in your post and I agree with most of them, but please understand that IBM hasn't completely left the hardware business. We (yes, I work there) are having too much fun kicking Sun and HP around. And by the way, we sold over $15b in software last year, so we're not just a consulting company.
Quantum mechanics: the dreams that stuff is made of.
Gave Linux XFS, arguably its fastest and most robust filesystem to date.
Sorry, you are wrong. XFS is robust ON SGI machines, and nothing else.
XFS uses a direct memory-to-disk scheme. This makes it fast, but not robust on common x86 machines. On these machines, the first thing to go out on a power failure is the memory and later the harddisks. So, on power failure garbage will be written to the disk. On SGI machines, they added little capacitors to the memory, so it will survive more than the harddisk (and write will be correct).
I learnt that on the FISL.
Ummmm....
A modern mainframe is still a IBM Mainframe.
They never went away, they never died. People still use them and people still need them.
Hell I WORK with a freaking mainframe. It runs 24/7. All day, all night, for months and months and months.
And my laptop could kick it's ass in a linpack contest, but proccessing power was never the point of mainframes. It's all about the Input and Output. Try proccessing data, running queries, and updating 10 3 gig-size databases from dozens of disks and dozens of tape drivers.. simutaniously.. on a nice new P4 3.6ghz with dual scsi disks and see how far it gets you.
You can't do it.
Mainframes were always been there, they are here now, and will be here for decades to come.
SGI never made mainframes. They made Unix servers and Unix worksations.
They made high end Unix servers and workstations which were once very commonly used in holywood movies and such and are now being replaced by Linux rendering clusters.
Mainframes don't run Unix. Unix/Linux CAN run on them, but thats not what you buy a mainframe to do.
They also make a few supercomputers.
Mainframes can't do 3d stuff. Hell they only have a monitor for monitoring the hardware and still use text based menu-driven interfaces.
The only reason I mention this is because your the third person so far that confuses mainframes with other completely different types of computers, and those are only in posts that have been modded up.
It's a reasonable question, all right, with an unexpected answer: I/O. This is the one area where IRIX still stands out among the other Unix flavors, and nobody outside the supercomputer world knows it, even though it holds true on all their hardware platforms. If you look under the hood, you'll see that the IRIX kernel's I/O layer can move bits at a higher percentage of available bus bandwidth than any of the others. The OS does an amazing job of getting out of the way of the hardware.
When I was working on HP-UX, we used them as our benchmark goal, and never met it.
If you're NASA, you probably find the Altix supercomputers pretty compelling. If you're an iBank, you probably find the 8-24 way dual core (48 cores in the big ones) Sun boxes pretty useful for processing all your data and trades.
Sun boxes are about the same cost as x86 boxes in the high end, and they have all the stuff you really need. 64-bit, lights out management (you can discover problems in the hardware even after it has crashed, because it contains a little computer on a chip designed just to report the statte of the hardware, power cycle it, etc....), lots of PCI cards, SSL accelerator cards, lots of ram slots, disk slots, raid cards, etc....
Your average 8 proc US-IV system (16 cores) from Sun costs about the same as an 8 proc (8 cores) Opteron system from HP, for similar configurations. It (supposedly) has much better support for things like SSL cards and massive multiprocessing/multithreading, especially under java.
Someone probably should buy SGI and Cray. There is a market for high end (top 500) supercomputers and other high end data processing systems.
Around 1997, I went down to Sony Imageworks in Hollywood to talk to them about physics engines. They were almost entirely an SGI shop back then, but had just purchased some NT systems running Softimage|3D. I was asked whether some NT software was going to be ported to SGI, and, realizing that was a dead end, replied "Resistance is useless. You will be assimilated".
Three years later, I visited again. Everything was NT except for some of the same SGI machines I'd seen three years ago.
SGI just couldn't cope with graphics becoming cheap. Around 2000, they dramatically announced some NT workstations, priced from $7000 upwards. They just didn't get it.
SGI's supercomputer side developed some interesting hardware, but there's no real market for supercomputers. It's all government, and mostly pork anyway. Lousy price/performance has forced them out of the server farm business. What's left?
This is far scarier than the stock price right now. When they're carrying $264 million in debt, have $148 million in market cap, and only have $83 million in the bank that's a bad sign...
Key Statistics
Oh how the mighty have fallen...maybe somebody will buy them, but somehow I doubt it. The shared memory parallel systems market will have a gaping hole if SGI does fold. Could be a golden opportunity for someone new to enter the market.
If SGI goes down, what's going to happen to OpenGL and the OpenGL Architecture Review Board that's responsible for advancing OpenGL?
Presumably the ARB will be run by nVidia, ATI, and 3DLabs, just like it is now.
But once the pro driver has been written, it's a sunk cost and ATI/nVidia could afford to bundle that driver with every card they make.
First, it's not a sunk cost, but rather a continuing expense. Second, it's a different driver -- it generally optimizes quality over speed and undergoes extensive compatiblity testing. Third, my point was that you don't want to release your driver as open source because your competitors will take advantage of your generosity and force you out of the market. Finally, companies exist to make a profit and will do whatever benefits them. Doing otherwise would be unfair to their shareholders.
Gee, 'cos it isn't like HP have lightsout, Dell have a remote access console and even Sun's own v20z/v40z have that. Of coruse, the reason Sun have it is because those boxes are Newisys reference designs, and they've put it in place.
But please, can I have some of what you are smoking!
This is just total crap. To get a 8-way US-IV, you need at least a V890. Which comes in at $155k!
Sun's pricing
Now, you say 8-way Opteron. No tier 1 makes them, but I presume you mean 4 seeing as you can get dual core Opterons only. Do they cost $155k? Ummm, no. They cost $39k from Sun and half configured (2 processor box) from HP cost $17k. Somehow, I can't see HP being that much different than Sun.
v40z pricing
HP DL585 pricing
And when it comes down it, an Opteron is way faster than a USIV anyway so you don't even need that many processors. And yes, I do use these processors everyday so I do know what I'm talking about. Which apparently you don't.
If you really don't want to run Linux, you can of course run Solaris 10 on the v40z.
I joined the company shortly after the "Gee-Whiz" article and after Jim Clark left.
4 _sales/ in which you find the seeds which made Forest Basket (now a VC) make the fatal decision to abandon the good stuff in favor of the good ship itanic.
... the Origin was in fact called a J90 killer, which it was). Rocket Rick's little jaunt into PCdom could have been successful as it turns out, as right when the units started getting traction, the plug was pulled.
...
In 1994++ SGI had some absolutely killer chips on the boards, code named Beast and Alien. Beast was a hellacious monster. About 5-6 Gigaflops, 6 GB/s main memory bandwidth (bad old memory giving up on me). Due out in 98 or so. Really insanely fast design. Alien made beast look like a game boy processor.
In 1998, with the company defocused from its mission by the unfortunate (and now effectively fatal indigestion due to the) purchase of Cray, it decided to pursue a new path to use the much hyped and over marketed "Merced" processor. Remember this? Have a good long look at http://www.theregister.co.uk/2005/02/28/itanium_0
If a long term SGI stock holder digs hard and far enough, they might find a really good reason to fire up a lawsuit against Intel for doing what IBM used to do many moons ago. Convince your potential competition to drop their products and use yours, by pre-announcing so far out, and claiming strong sales numbers that never materialize.
Rocket Rick, an accountant, completely defocused the company. The Cray indigestion was taking hold, Cray took over SGI (BTW Cray-Link was renamed from NUMA link right after the acquisition, it did not come from Cray, contrary to urban legend
Mis-step after mis-step after
The story of SGI is a sad sad tale of failed management, still largely in place, failed or completely neglected marketing, failed sales efforts, failed relationships.
In 2001, Warren Pratt killed off the linux cluster push that a few of us had been pushing hard. Warren only wanted to make MIPS based machines. He just sold off a bunch of shares in May 05. Know anything we don't there Warren?
Later in 2002/2003 they killed off Irix.
Before, during, and after the Rocket Rick fiasco, we had massive infighting between Crayons and SGIers. This infighting took the focus off the customers and placed it in the internicine battles. We had Crayons attend meetings with SGIers and customers, where the Crayons would shout down the SGIers in front of the customers (and vice versa).
Massive egos lay in the way of doing the right thing. In the end, many good people left. Many stayed, only to be rewarded with a RIF.
SGI is a great place to have been from. Many friends who left SGI tell me how much they miss it. The people were the best. Really, it is was honor to be an SGIer.
Just sucks to be here now.
SGI uses chips that show every likelyhood of being killed off by their supplier as a non-performer. They are costly to design and build, and they are not breaking even for Intel. The designs of machine are based upon 10+ year old ideas, and while that is not a bad thing, they are about to be massively eclipsed by other CPU vendors.
It pains me to say this, but this game is over. Linux won. X86, or more precisely, x86_64 won. There are some SGI management I would have sadly said "I told you so" to, but they have left a while ago.
You can buy Sun Blades with UltraSPARC IIi's up to 650MHz, and with UltraSPARC IIIi's up to 1.6GHz CPUs. You certainly pay for it, though.
IIIi 1.6GHz = 7195$ and up
IIIi 1.5GHz = 3195$ and up
IIi 550MHz = 1395$
IIi 650MHz = 1995$ and up
"I was actually at the event that started the complete destruction of SGI. It was summer 2000 in New Orleans. This"
SGI's destruction started around 1995-1996. Jim Clark being driven out was the first nail. SGI was on the cover of Businessweek which is a suprisingly strong indicator of companies who have peaked and are on the way down.
The pivotal SIGGRAPH was the one right after SGI bought Alias and Wavefront. Microsoft was showing Softimage running on Pentium Pro under Windows NT. As soon as it came out smart people stopped buying lame ass O2's(except for video processing which is the only thing it was good at, I.E. low end Avid). The Achilles heal of O2 was its memory bandwidth to the CPU. It had 1/3 of the memory bandwidth in a Pentium Pro and it made the machines run like complete dogs on most applications.
The Cray merger was in this same time frame and it was a complete failure of a merger. The two teams completely hated one another, fought like cats and dogs, and back stabbed each other to customers. Not to mention supercomputing is a business you can turn a profit in only by carefully managing expenses and landing big government deals. It is NOT a growth business. Well SGI did have a surge of growth when they came out with the R8000 mini's, but that growth came at the expense of Cray and after carving up Cray's low end they for no obvious reason bought them though they already had the profitable part of their market.