In this post, the author indicates that SGI committed suicide, or at least attempted it twice. Their rationale was that the NT porting bit was the first phase, and that the Itanium choice was the second. Further they posit that there is no value left in the company.
I disagree with the first and third points. SGI acquired Cray during the time when we were busy taking away their business. This was IMO, one of the first fatal mistakes. To correct some misconceptions/misperceptions, the O2k (Origin) was called many things internally, including the “Cray-killer”, and its design was complete before the acquisition. Renaming the Numa-link to be Cray-link was a (misplaced) fob to Cray. SGI built the machine based upon the Stanford DASH project work (hired key DASH architect for this). SGI knew large processor count machine work, but building an OS that happily worked well across a single system image of more than 64 processors was hard. Took a while. Very few people can do this today, a decade later.
There were many reasons why this acquisition was a bad idea. There were many mistakes made during the acquisition. Many afterwords. SGI managed to give away technology for a song that made a direct competitor a fortune. SGI managed to avoid focusing upon a massive growth market, as it didn’t understand web servers, and its Challenge S machines blew the doors off everything out there. SGI let Cray managers take over, and this resulted in a larger version of Cray, burning through resources, screwing business decisions up, missing opportunistic and strategic business that they had no personal interest in, and finally retreating to the high end (also known as abandoning the growth markets).
The NT bit was a minor footnote, really. The porting of NT to a new architecture was interesting, but SGI blew it by not recognizing (and enough of us were telling management this) that the market was rapidly being commoditized, which meant you needed to reformulate the plans. Build better video cards. Capture a much larger market.
The old diehard business types in SGI poo-pooed this. There was no way, in their expert opinion, that 3D graphics would ever get commoditized, and no way that desktop PC cards would ever be able to do what SGI systems did.
Ok, you can stop laughing now. They really believed that.
Shortly after these decisions, Dave Orton, Dan Vivoli and others bolted and joined nVidia, taking lots of graphics folks with them. Kai-Fu Lee bolted for Microsoft, and with the hiring of “Rocket” Rick Belluzzo, SGI appears to have bartered a large chunk of IP over to Mr Gates and crew for cash on the barrel-head. Oddly enough, “Rocket” Rick later wound up working for Mr Gates. “Rocket” Rick wasn’t a complete disaster for SGI, the momentum had been established, and the death spiral had begun. He simply hastened it.
The NT machine was actually a good idea (if not a poor design). By being on the forefront of commoditization of graphics, SGI *could* have been where nVidia/ATI are today. Huge market, quite profitable, with a nice supercomputer division on the side. You create a new market by cannibalizing the old. This is a fact of business, you have no choice in the matter. It will happen, again and again. The foolish company will deny reality and pretend it will never happen. Then they will watch in horror and inability to act while a new company comes along and destroys their market for them, creating a newer one, and a larger one in the process. The agile company will prepare for it and be ready to cannibalize their own market once they see it is ready for it.
SGI could not deal with PC graphics adapters being able to replicate the sort of work that their big cash-cows could. So an 800M$/year market was destroyed. In favor of a $10B+/year market (100M PCs sold per year, with at least $100 graphics adapters in them).
As I build our business, these lessons are not lost on me. And I will not likely hire many former SGI marketing folks.
As for the value left in SGI, there is actually significant IP in the Altix. It is frankly one of the best designed systems out there. 10+ years later, this design is still exceptionally good. It is just the CPU choice that needs to change.
Itanium was (obviously) the wrong choice. Hindsight is 20/20. If SGI had followed suggestions from some of us on the inside to focus upon building the best platform and adapting the best CPUs to it (Alpha, Itanic, K8, …), as well as focusing upon reducing the cost of the platform by exploiting volume technologies, it could have done some very interesting things. As it is now, if it survives, it will become another relatively undifferentiated and expensive cluster player.
The value is (for the moment) in the design. SGI’s competitors have not been standing still though, and the architectural advantages will not be exclusive to SGI for much longer for large processor count systems. We just bid a 16 core single system image with 128 GB of RAM for an opportunity. Next year this will be much larger. System is designed on a cubical hypertransport linkage. And thats an issue for SGI. NUMA link gave them some interesting capabilities. SGI needs to look at HT3. Again, if they make the right choice to adopt it, then they can exploit economies of scale and get to market faster and cheaper in the Altix platform. If not, well, George Santayana’s quote is apt. Sadly, it would be their own history repeating, and I am pretty sure this is their final chance.
SGI did not willingly commit suicide. It was lead by marginal leaders since Jim Clark left, ignored market directions, acquired and then spun out competitors, and ignored opportunities. At a venture conference recently I saw a discussion on stranded assets. One of the commentators was talking about stranded assets in the context of the big-3 automotive companies here in Michigan. A question was on how to get access to these assets (think of the folks who decry conspiricy theories to suppress good technology, not that they have a point, but good technology often does not make it out of large companies for any number of business related reasons). The person doing the presentation gently suggested that a bankruptcy and asset sale was a good way to access stranded assets. Since they were talking about the big-3, this caused some nervous laughter in the room, but his point is correct. If you ceased SGI operations today, there is still (for the time being) value left in the company that could be sold. If SGI keeps grinding until the last possible moment, that value won’t last and probably won’t survive.
The people you have to thank for this are SGI management from 1995 onward.
Not suicide. Just really bad decision making.
Viewed 17083 times by 3466 viewers