SGI: Discussion

jimmer wrote: As to the original post asking to comment on: "chips is where it's at, mobo's are obsolete."

I'm not sure I understand what this greg chap is on about in his blog. As I understand things, chips need interconnects - no? Mobo's are simply surfaces that support and implement interconnects - no?

If so, if you've got chips you'll have mobo's. Though it's unlikely that the classic PC _architecture_ will still be around in 2010.

On the other hand, wasn't the whole point of the Origin 3xxx architecture to make a system with upgradable interconnects? Even so, somewhere along the line, a 'chip' will have to be stuck onto some kind of connector-to-the-rest-of-the-system thing, and that's a Mobo in my book.

As to the original blog entry, the point was about the accelerated integration of more and more of the interconnects, subsystems, etc. onto one or a few pieces of silicon in order, partly, to keep up with the advancing speed of processors. But the key phrase was "add value".

squeen wrote: So I now add: "Where did SGI go wrong?"

Continuing this line, SGI stopped innovating in silicon. If all the component parts of the system are off-the-shelf devices, there's obviously a limit to what value you can add. As squeen points out, for instance, Prism's ATI gfx are a retrograde step in a market which is supposed to be moving forward, and never mind that SGI once owned the market.

For my own part, the "Where did SGI go wrong" is a little more complicated. I think they fell into a classic trap of not knowing what business they were actually in. All their early work involving gfx involved throwing large amounts of data around to support visual output, so they got very good at that. However they seem to have interpreted that as being in the business of high-bandwidth data processing when much of their market thought they were in visual computing. "Their" perception took them towards HPC, whilst many of their customers expected them to stay in visual computing.

Having bought Cray, they owned 2 iconic brands. They could easily have taken all of the gfx knowledge, kept it branded as Silicon Graphics, and been what nvidia are today. Or they could have extended the Indy approach and owned what is now Apple's market. Meanwhile who wouldn't want to have a Cray in their datacentre? The way they've gone they've just squandered decades of IP, hard work, development, testing and customer loyalty.

And from here? Well, they don't seem to be able to innovate at anything more than motherboard design (or bolting on RASC, or refining old interconnects) so where is the value they're adding? Right now they're good at making very large HPC machines, but all it will really take is for IBM say to decide to own that market for PR purposes and sell machines at no margin for that market to vanish.

Now, if somebody wants to port Darwin to Itanium then I'll happily pay big bucks for a "big iron" machine with an SGI cube on the front and MacOS X on the desktop to run some video editing jobs, even if it isn't the best thing out there

clavileno wrote: For my own part, the "Where did SGI go wrong" is a little more complicated. I think they fell into a classic trap of not knowing what business they were actually in. All their early work involving gfx involved throwing large amounts of data around to support visual output, so they got very good at that. However they seem to have interpreted that as being in the business of high-bandwidth data processing when much of their market thought they were in visual computing. "Their" perception took them towards HPC, whilst many of their customers expected them to stay in visual computing.

I have to accept that you have taken an step higher the SQueen's analysis with this paragraph, and what you describe there seems to be a very good approach to their problem, and left the door open for further analysis too.

clavileno wrote: Having bought Cray, they owned 2 iconic brands. They could easily have taken all of the gfx knowledge, kept it branded as Silicon Graphics, and been what nvidia are today.

Of course, even if they decided to stop their own production of custom graphic chips, what one could expect today from a brand with the importance of SGI is at least something as an onboard 256-chip graphic cluster based on NVidia, to have on the size of an Onyx4 or Prism Deskside a graphic engine board with 256 graphic cores working on a collaborative SMP way, but as a single high-power graphic pipeline... then just add the possibility to add more of those baords to have a multi-pipe graphic monstruosity... then of course, just write excellent drivers, and you're set.

And that's what I could call a relatively innovative thing.

clavileno wrote: The way they've gone they've just squandered decades of IP, hard work, development, testing and customer loyalty.

Indeed.

A comment by Sweetman:

The R10000 was a major departure for MIPS from the traditional simple pipeline; it was the first CPU to make truly heroic use of out-of-order execution. Although this was probably the right direction (Pentium II and HP's PA-8x00 series followed its lead and are now on top of their respective trees), the sheer difficulty of debugging R10000 may have set Silicon Graphics up to conclude that sponsoring its own high-end chips was a mistake.

Oskar45 wrote: A comment by Sweetman:

The R10000 was a major departure for MIPS from the traditional simple pipeline; it was the first CPU to make truly heroic use of out-of-order execution. Although this was probably the right direction (Pentium II and HP's PA-8x00 series followed its lead and are now on top of their respective trees), the sheer difficulty of debugging R10000 may have set Silicon Graphics up to conclude that sponsoring its own high-end chips was a mistake.

That's a very interesting observation. However that's a little like climbing, say, K2, reaching the top, and then deciding that Everest should be left to somebody else. The next chap up Everest just picks over how you got up K2 and then takes all the glory. I remember sitting in front of Indys a dozen years ago, reading R10K information, and thinking "someday all computing will be this good".

GeneratriX wrote: Of course, even if they decided to stop their own production of custom graphic chips, what one could expect today from a brand with the importance of SGI is at least something as an onboard 256-chip graphic cluster based on NVidia, to have on the size of an Onyx4 or Prism Deskside a graphic engine board with 256 graphic cores working on a collaborative SMP way, but as a single high-power graphic pipeline... then just add the possibility to add more of those baords to have a multi-pipe graphic monstruosity... then of course, just write excellent drivers, and you're set.

And that's what I could call a relatively innovative thing.

Agreed... but that's also something which could easily be copied by others if there were a market for it. It would help reinforce SGI sales, but it wouldn't necessarily be noticed fast enough, or for long enough, to sustain a high price or deliver conquest sales. And, forgive me, but isn't the market for really "HPG" just as small as that for "HPC"? Now if they were still cutting Silicon, and could put that network of cores down, they could use it themselves, license it to others, all those revenue-enhancing things that companies with good IP can actually do

Interestingly I happened to find this http://sunsite.uakom.sk/sunworldonline/ ... ation.html showing SGI with 10% of the 1995 sales volume (but about 13% of revenues) from workstation sales in the marketplace - larger than, say, IBM in that market. Sun, unsurprisingly, had 40% of the market.

Today Sun still sells a lot of Unix workstations, IBM still sells a lot of Unix workstations, SGI decided the future wasn't in Unix workstations... I wonder if the turning point was when SGI management stopped believing they were right to innovate, that they were too small, that they had to "follow the market" - and their competitors didn't read the same book, so just carried on innovating...

All the best

clavileno wrote: And, forgive me, but isn't the market for really "HPG" just as small as that for "HPC"?

Sincerely, I don't know. I'm not following the statistics for the above facts; but probably, you're right.

clavileno wrote: I wonder if the turning point was when SGI management stopped believing they were right to innovate, that they were too small, that they had to "follow the market" - and their competitors didn't read the same book, so just carried on innovating...

Sadly, and again; you're probably quite right. The company screams for the fifth-escence from a brave soul to lead these change again.

managed resistance wrote: I have to admit that I lost count of how many times I laughed while reading this thread. Then I realized that this thread pretty much sums up the type of people that use and like SGI computers: creative, independent, free-spirited, intelligent, cynical, and witty.

I read with great interest how people seem to have come to the same conclusion about SGI: They've lost the plot.

I just took a very quick look at SGI's website prior to surfing on over to Nekochan, and I was rather befuddled. It seems that SGI has adopted this very patronizing community responsibility position, while remaining focused on markets like defense and government contracts. It's just insane.

I wonder if the good folk at SGI even really have a clue as to what they're doing. I mean, it's nice that they have this wonderful program of hiring welfare recipients and feeding the homeless, but um... how does this fit into the grand scheme of things. Philanthropy aside, SGI should have as its objective, the broadening of its target market... or, failing that, diversifying into new markets that will provide them with the scope and depth that they will require in order to remain a viable corporate entity.

I mean, it just seems ridiculous what the folks at SGI are doing. There should be little wonder as to why SGI is failing: They haven't a fucking clue!

Well, you don't have to be a rocket surgeon to figure that one out...

Maybe we need to email this thread to the boneheads below:



Executive Team

Below are the SGI executive profiles that provide details about our key personnel across the business groups worldwide. If you would like to obtain an executive profile not listed here, please contact SGI Press Relations.


Robert Bishop
Chairman and Chief Executive Officer


Warren C. Pratt
Executive Vice President, Chief Operating Officer


Sandra M. Escher
Senior Vice President and General Counsel


Eng Lim Goh, Ph.D.
Senior Vice President and Chief Technology Officer


Jeffrey V. Zellmer
Chief Financial Officer


Anthony K. Robbins
President of SGI Federal and Senior Vice President of SGI North American Field Operations


Brian Samuels
Senior Vice President, Global Sales, Service & Marketing

I'm not expecting anything, but I think we should try it.
At least we would have tried

Bluefan wrote: I'm not expecting anything, but I think we should try it.
At least we would have tried

Sure! ...after all, who knows?

GeneratriX wrote: Well, let's start with something different on this thread. I want to talk about the power of the marketing, and how you can make look better what in fact is the same, almost the same, or plainly worst.


1) When Apple uses Intel processors, we say "-wishes do come true!"
2) When Apple uses ATI graphic pipelines we say: "-Oh!!! ...That means it’s screaming fast. Up to 256MB GDDR3 graphics memory!!!"
3) When Apple uses to have on their site the yin-yan logo, with the legend: "UIniversal Binary: Software just works", we said: "-oh geez!; it is a miracle... they managed to run Mac apps inside the intel-OSX!!! Geniouses!!!"
4) When Apple uses SATA drives we say: "-bloody hell!!! ...they will provide an amazing 250GB 7200-rpm Serial ATA hard disk drive!!! Cool!!!"


...Oh Steve; what a magnificent genious you are, and what a magnificent genious are your marketing chieffs!!! ...You can make almost the same thing that the rest, but it will look different, and people will think that they are "Thinking Different" ! ...Heheheh... we gotta love you, mediatic Guru!

untrue. I think macs are no more macs. THat whole intel crap is crap. APple externalized the chipset and motherboard design. It is not just a "processor" change, it is a philosphy change.
intel macs may sell, they may be geek, people may like them, but I dislike them.