SGI Sets Sights On Turnaround

grub wrote to us about an article about SGI, and its ongoing battles to turn its corporate fortunes around. The company's been doing interesting stuff for a long time - here's to hoping they stay around.

Comments

SGI still around?

[BWJones]

Other than its ability to run on cheap (price and often quality) hardware, I still don't understand SGI's movement to Linux. I guess that I am showing my ignorance here, but it seems to me that Apple and SGI are in similar situations right now in some respects. Both companies historically have relied on income from the hardware side of things while making a closed OS/hardware system that for each of their respective markets is very effective. The difference between Apple and SGI however is that SGI already has a UNIX OS with a GUI (however difficult it is to manage but VERY extensible and powerful), and Apple is developing UNIX with a GUI (easier to manage, more powerful in some respects, etc etc etc...). Both companies need major transitions to survive, but why Linux/Intel? IRIX is already mature, stable, fast, with great graphics capabilities and IO capabilities, so I ask again, why move to Linux and Intel?

Both companies obviously want to benefit from the open source paradigm while still remaining in business with proprietary OS's. (I am guessing here for SGI as I assume that they will make their OS on a proprietary linux model) The approach Apple is taking certainly makes sense to me by developing a UNIX OS that includes the opensource Darwin, but I am totally clueless as to what SGI is doing here. What makes Linux more attractive than simply continuing to develop IRIX and putting more effort into improving, simplifying some features, and pushing development for IRIX on perhaps less expensive hardware? (among other changes to their business model) Again it seems to me that SGI is making another crucial mistake here as the developers that have tapered off work for IRIX have not for the most part started developing for Linux (although I know of a few examples), primarily they have lost ground to Wintel. (thus
their misguided attempt at Wintel/SGI boxes I guess)

In short it appears that they are trying to make Linux/Intel into what they already have in IRIX/MIPS, only with cheaper hardware which seems awfully dangerous to me for both end users and the company.

I believe that by 2005 SGI will no longer be in the low to mid-range workstation market. This market will belong to perhaps Linux/Intel or OSX/PowerPC. Right now for what my maintenance contracts cost me for a single SGI Octane, I can purchase a new G4 WITH a 22in Cinema display YEARLY! This is not even talking about the $40k initial acquisition costs.

SGI will survive in the server market and high end visualization market if they are not acquired by someone else. After all SGI's market cap is only around 585 million last time I checked.

Re:SGI still around?

[foobar104]

I believe that by 2005 SGI will no longer be in the low to mid-range workstation market.

I'm not sure how you define that market, but for my money SGI isn't in it now.

SGI's workstation products include O2+ (a special purpose machine), Octane2 (definitely high end), and Onyx 3000 (the highest of the high). There's going to be a new product coming early next year-- this is not a secret, although the details and case color are-- that sits below Octane2 on the price graph, but it isn't going to be a low- or mid-range desktop computer either.

Re:SGI still around?

[skotte]

I still don't understand SGI's movement to Linux

hey, here's a couple reasons fFor a company to move to linux:

1) sales fForce: "it looks good! it costs nothing to buy, and you only make money on selling it!"

2) marketting: "it's The Next Big Thing, and we wouldnt want to miss out on it, now would we?"

3) management: "surveys say more people know what linux is, than irix. let's go with the established winner!"

takeover candidate = ibm? Re:SGI still around?

[StandardDeviant]

A takeover of SGI does seem pretty likely at this point. Here's some reasons why I suspect IBM would snap them up:

1. IBM has a much larger market cap and a fairly healthy balance sheet (this gives them the resources)
2. One of IBM's most profitable business sectors (or so a little birdy told me) outside of consulting is big iron and high performance computing, which is right up SGI's alley.
3. IBM has a large product base and customer base that center around Unix already
4. if IBM is serious about a Linux push, SGI is a source of skilled engineers who have also been working on Linux in the enterprise
5. it'd be one less competitor for all those cushy big-gov/big-edu contracts that IBM loves

Just idle speculation on my part. Sun is more of a pure-play Unix vendor, and thus might seem more appropriate as a takeover initiator, but I don't think their financial reserves are high enough to do it. Further, they're more of a "one-os, one-platform" company than IBM and would probably have a harder time assimilating the SGI folks/products.

Jeez, if SGI goes tits up, how many Unix (commercial) vendors will be left? Both HP and Compaq seem to be treating their unix offerings as an afterthought compated to cheap shitty PCs and winprinters. I guess just Solaris and AIX. God save us all from AIX being the only Unix out there...;-)

SGI's Failing Points

[Lethyos]

Yes, I definitely agree that SGI has done some killer things. They revolutionized the graphics industry and demonstrated that computers can be made to do visual effects never dreamed of. Their systems are powerful tools for research. All in all, they've been quite swell.

But, we don't need them anymore. Nor do we want them.

SGI offers big, expensive, proprietary solutions, that like Microsoft, lock you into their product line with little or no hope of escape. Let's discuss the reasons.

1. Lack of extensibility. SGI boxen typically do not scale well, and if they do, much of your hardware has to be replaced to accomplish any scaling. Ever try to upgrade an Indy? And O2!? I certainly understand that in any upgrade, sacrifices of existing hardware must be made, but they are no champions of modularity.

2. Proprietary hardware. SGI hardware, for its consumer price-level equivalents is not all that great. You can spend $16,000 on one or maybe two decent SGI systems, or you can buy 10-15 high-powered PC's and cluster them. You get the advantages of redundancy too. Another problem here is repair work. Nobody but SGI and SGI certified technicians can repair their hardware. Worse still, only SGI and a few licensed vendors manufacture SGI hardware replacements. More money here. And then there is Irix...

3. Proprietary OS and software. Irix is a disgrace. Certainly, it's a great performer, but because it's geared specifically to SGI hardware. Take Linux and optimise it to the same level and write good drivers, and you'd have not just a strong contender, but a superior OS. However, it's just not there and SGI doesn't want it to be. They're too proud of their OS and they want Irix tools to remain Irix-only so SGI software vendors can't take their products to other markets without tough costs. Since everyone does servers these days, SGI doesn't mind having Linux run on Challenge or other volume servers. Besides, everyone who wants Big Iron for www.hugefuckingcompany.com uses Sun anyway.

All in all, what SGI does for huge costs can be done in the PC scene with a fraction of the price. Perhaps not in Linux yet, but certainly in Windows with products from NewTek and ReelMagic for example. With nVidia around pumping out killer graphics hardware, what do we really need SGI for? I guess the only reason I can see is that they produce big solutions (who else will build a C.A.V.E. for you?). Can anyone clue me in on what it is exactly SGI does that we can't do everywhere else these days?

Re:SGI's Failing Points

[Anonymous Coward]

I'm posting ananymously to avoid trouble however I really do have to respond to this as and SGI employee (Standard legal disclamer I speak for myself not employer etc...)

1. Lack of extensibility.

There is a difference between upgrading and scalability. Scalability is not about adding an extra harddrive it is about adding CPU's and IO. Take the 3800 for example. That goes to 1024 CPUS' running off one kernel. That looks like it scales well to me. Yes you pay but it sure is impressive

2 & 3 Proprietary OS/Hardware. So what? Sun does and get aways with it. So does HP/IBM and every other UNIX manufacture. The hardware is designed for certain function and the SGI does it very well.

Adding preformance to a system is not as easy as you think. Its not about just writing some code and say look how great this is. Look at from the point os Sparc Linux for Sun. The scsi layer get only arouind 1/2 the performace of Solaris. Why? Because solaris is designed to make the most of the hardware

I think you should wait to see what SGI can really do

Re:SGI's Failing Points

[Graymalkin]

What the fuck are you talking about? Do you know anything about SGI hardware or software?

1. Saying SGI boxes do not scale completely invalidates any thought you've ever had in your life. Have you ever seen an Onyx/Origin system before? You add a brick and have twice the processing horse power or add 10 PCI (or any periphrial connection you ca nthink of) slots to do anything you want. You don't even have to turn the system off to add anything to it. Origins scale from 2 processor to 128 processor systems at the drop of a fucking hat. Using the Indy and O2 for your examples is just jackassery. Neither of which were designed for easy upgrades. Later revisions of the O2 when the Octane was released were easy as could be to upgrade. You turn the system off and slide out a processor modules and replace it with a new one. Voila.

2. Clusters fucking suck. Jesus get it through your Beowulf worshiping head that clusters are not the answer to all computation problems. LANL might build a bix cluster of PC systems to solve some embarrassingly redundant set of equations big fucking whoop. The big SGI systems are cache coherent and have messaging latencies measured in nanoseconds. A processor on one end of a 512 processor machine can talk to another processor with similar latency that a PC based system can talk to its own RAM. It is trivial to write software that spawns processes onto multiple nodes in a cluster, ccNUMA systems run as if their far flunt components were a single machine.

3. Linux cannot do that. If it COULD do that it would no longer be Linux and certainly would not be portable outside of SGI's hardware. Do you seriously think you can get good performance out of off the shelf software? Hell no. Optimizing software and good drivers doesn't mean anything. You've bitten far too deeply into Linux hype if you think it is even a contender to the grand daddy Unix systems.

Bringing up NewTek and ReelMagic cards is just fucking ridiculous. Wow you can have a hardware video processor that uses a PCI bus. That's rad...until you've seen an Origin in action running more than a dozen pipelines feeding 8 display channels. Try that with a PCI bus and see what happens. Nobody buys SGI for the logo.

Re:SGI

[pqbon]

Given how the company is doing... I would say they can't. They just didn't know they couldn't...

Re:No mention of linux

[skotte]

Nice post. well spoken. i have one point of difference tho. this is old information, but it's information.

about a year ago, when SGI fFirst started pushing linux, they held a number of meet-n-greet seminars about linux and what it can do fFor you .. you know, bring people in, talk all morning, give people coffee and donuts and some cheap handouts. all that. and of course they talked a great deal about oss.sgi.com.

After the event, I asked one of the event organizers whether linux was to replace IRIX. he told me fFlat out, "oh no, IRIX will always be the big engine. it handles larger machines .. yadda yadda .. Linux will be fFor desktops, and small networks."

so, according to what i know, they werent going to replace IRIX, just accompany it.

But i dont know your sources or their reliability. :) hell, you might be Bob Bishop hisself :) in which case, it's safe to say your source is better than mine :)

Re:SGI

[skotte]

nah. they just didnt know they couldnt sell wintel.

in fFact.. get this: the visual workstation 320 came out fFirst, a couple years ago. solid kick ass machine. hardwired kick ass graphics card, the whole nine yards. one problem tho: if you wanted a different graphics card, you basically couldnt, or it was a lot of work. and the add-ons and stunning graphics made it fFairly expensive, and hard to sell. so -- and this is the fFunny part -- they downgraded the next year's models, put in slower dumber cards, and actually dropped the system's version number to 230 indicating it's a lesser machine.

thats right.. their intended equipment was actually TOO GOOD. not many companies can seriously claim this.

even if the wintel strategy was a bad move, by golly they did it up right

Re:SGI was killed by it's greedy salesmen

[hetz]

I really don't know if that wouldn't affect their financial bottom line - look at the upcoming ATI Q4 profits - they made nice amount of money with their contract with Nintendo about the graphics chip.

Case in point - SGI got some great hardware, totally impressive machine, and I haven't heard many flames about Irix. Yet it reminds me what was written by a british computer magazine about the Amiga 1000 - Dream machine - Nightmare price - I guess it's the same for SGI.

So - you could think "ahh, market conditions now are not the best, they'll lower their prices on their workstations/servers" - well, you are more then welcome to look at the prices of their Visual Workstation machine's prices - FAR and ABOVE any competition! who would be nuts to buy at those prices?!?!?

Bali and Odyssey... *sigh*

[Anonymous Coward]

Bali and Odyssey -- Two reasons right there why we no longer use SGI (and really, why we're now in a totally different line of business).

When we bought our first Octanes in 1997, we were excited about the totally new gfx due in "about 18 months". Shucks, the only upgrade we got for YEARS was the simple "e series" speed tweak. By the time Odyssey gfx (VPro) shipped, we had already shifted gears and changed platforms.

Same goes for Onyx2 and its graphics. At the time we bought our first Onyx2, it came with original IR (InfiniteReality) graphics. We were told that IR2 was due soon, and to be followed by something totally groundbreaking (Bali). Hell, Bali never did ship. Bali was never even finished. Here we are at the end of 2001 and the current high end graphics offering is just IR3, another minor speed boost.

SGI can build some damned impressive machines, offering GOBS of thruput--bandwidth from hell. But what can we use it for? Only Bill Gates could afford enough disk subsystems to swamp that much bandwidth, and person can only make use of so many HDTV I/O streams. My company used to work on "photorealistic" 3D simulations for a wide variety of clients. Over the years we had used and abused many different platforms, constantly desiring more performance. Our Onyx2 systems served us well, but the lack of a real graphics upgrade left us scratching for more. We tried E&S, we tried 3DLbas, we tried nVidia. Some speed boosts, many new features, but total kluges when it came to driving more than one display or trying to feed the graphics pipeline. For us, there really was no solution. SGI canceled Bali and the only other alternatives were halfbaked. After a stint with non-realtime (rendered) graphics, we eventually branched off into the world of physics sims.

Cutting edge graphics, where did you go? Please tell me there's more to the 3D world than IR, WildCat II, and GeForce3. Has *nothing* (other than cost) really changed over the past five years? It's almost as though I haven't missed anything in the 28 months I've been away from 3D.

Re:Bali and Odyssey... *sigh*

[Anonymous Coward]

There's been much more than dependant texture reads in the past few years. I design hardware, and my company has put more than that in our last few products.

New texture blend modes, like dot product for bumpmapping were added. Texture compression support is universal now. Cubic environment maps also hit hardware, useful for reflections and phong-like lighting among other things.

The latest n' greatest feature are the programmable pipelines. Vertex and pixel shaders are here, which can enable all sorts of new effects.

What hasn't changed in the past few years is opengl, which has been stuck in committee. One really needs to look at the extensions by the various vendors, or check out the dx8 sdk from MS to understand where graphics has been going.

There's more than floating point pixel formats in the near future as well. Displacement maps, bi-cubic texture filtering, and adaptive tesselation of high order primitives. Of course, the bandwidth internally and externally will make a leap to support these features.

Make a pretty case, sell on the name.

[Snowfox]

Is it just me, or should SGI still be able to sell based on the name, if they take a rational desktop approach?

A pretty, curvy plastic case with the SGI logo prominently displayed, and they could probably compete with Dell for workstation products, while adding 10-15% just for the name.

I work in video games. Many of us, especially my artist coworkers, have worked with SGI extensively in the past. They miss the SGI platform, many with a fondness on par with that of the typical Linux, Mac or Amiga fanatic. And these people do have a voice when it comes to purchasing. If these guys thought they could get "An SGI that runs Windows," but at a sane price (they missed this part with their Windows endeavors), they'd jump on it.

Hell, I'd probably get one too, just for the novelty of it. A bona-fide SGI running Linux just feels cooler than generic PC hardware, even if I know the internals are identical.

There's probably a lot of money to be made in selling branded PC hardware. When Gateway bought Amiga, they could have probably sold thousands more units just by replacing front panels with something stylish and Amiga-esque, flashing a set of BIOSes with a snazzy "Amiga Phoenix" or similar logo & tossing a UAE CD and a Boing! mug in the box. There was no need for them to look into reinventing the PC, just like it was silly of SGI to go about trying to reinvent the PC when they tried shipping Windows products. Commodity hardware is rocketing forward so fast that most any attempt at creating custom hardware for your own PC products is purely daft. It's all about presentation.

Certainly, pretty cases wouldn't have to be SGI's only business line, but it could certainly be a source of safe & easy revenue to help turn things around.

When will I see a SGI graphics card...

[Gat1024]

for my pee cee?

Hell, "graphics" is in their name! What's wrong with giving NVidia and ATI a little competition, especially at the gamer and prosumer level? How about a sub $1,000 card that does digital video in and out, accelerates OpenGL with *precision* and spanks NVidia at games?

Hell they can tweak on of their old boards and milk it all it's worth. And it shouldn't cut into their fat margin business.