Whither SGI

Obviously SGI has existential challenges ahead. This means something quite simple. No cow is sacred. No ego’s can get in the way of doing the right thing by the shareholders.
I will be frank. This is about 9 years too late.
First, will SGI recover? Possibly, though I am not going to bet on it. The challenges are not just internal, their competitors have wasted no time in making use of their situation. High performance computing is an unforgiving market. Single mis-steps can be fatal. Running a marathon of past mis-steps like SGI did should be fatal. I wouldn’t count them out, but the odds are against them regaining good will and customer trust.
The fundamental question that is going to be on every customers’ mind is, if I buy this proprietary machine from this vendor with shaky financials, will they be around when I need help with it, or need to update/upgrade it? I remember visiting customers when I was at SGI 7 years ago where the first thing out of the customer’s mouth was “Prove to me that you will be there in 2-3 years when we need support for the machine”. They were only 5 years off, but this question remains. SGI needs to address the existential question to make sales. They need sales to recover. This suggests the appropriate survival strategy, and it is not as an independent entity.
Second, what will SGI’s product portfolio look like in 90 days? Again, I am going to be frank. They need to ditch platforms/OSes that are not selling. They need to do a really rapid product transition. They can thank the now departed-from-SGI Warren Pratt for this. In 2001, Warren pulled the Linux cluster rug out from us. This was the final straw for me, and I left then. SGI, in 2001, right before the explosive growth of the Linux cluster market, decided to stick with its Itanium2 strategy (still a ways away from realization), and not do x86 based clusters. Note: the linux cluster market has been growing 60-90% per year for the past 3+ years, and accounts now for about 1/2 of the overall $9B HPC market. It shows no sign of slowing down. The majority of this market is x86 and x86_64 chips, with the latter appearing to replace the former.
This was a terrible decision. SGI previously had made other really bad decisions. But this one took the cake. Sure, hindsight is 20-20. Clusters have done to the super-micro market what the super-micro market did to the supers. They effectively destroyed it while creating a more active and much larger market. For those who are thinking ahead, what is going to destroy the cluster market? That is a good question: at first “The Grid(tm)” was supposed to do this, though it turned out to be more hype and hot air than high performance computing. It is good for some things, though it is a far smaller subset of what clusters are good for. The next thing will either destroy the market by creating a new one or create an augmented market. I am betting on the latter. SGI needs to be thinking about this, as it is coming far faster than they might think.
SGI needs to have a competitive product portfolio that can address the entire gamut of HPC, not just the upper end of it. Remember, according to IDC and others, the market where SGI, and Blue Gene, are playing is *shrinking* while the rest of the market is growing rapidly.
Right now, SGI sells Itanium2 units. They are expensive, they are large heat producers and power consumers, and they aren’t that good on a wide swath of codes. SGI’s product transition needs to start with pulling those out and putting in AMD64 based chips. Dr. Eng Lim Goh (someone I worked for a long time ago) CTO of SGI, recently said that he didn’t know any chip that could address a terabyte or more of physical memory. Dr. Goh, please have a good hard look at the Opteron. I mentioned something before about sacred-cows. The Itanium2 in the Altix is such a beast.
The SGI CEO said several interesting things. In the story on HPCWire, this is approximately what was said:

  1. The company was slow to adopt new technologies in the marketplace.
  2. As a result of various acquisitions that it made, it lost it’s core focus.
  3. The management team expended a lot of its effort dealing with the outcomes of these points.

My question is, why did SGI have the winners it did after Ed McCracken, and where the heck was this guy? This is the person we needed after Ed. Lots of former SGI folks blame Rick Belluzzo for the problems of the company, but to be quite honest, the problems predated Rick by a few years. Rick wasn’t that good. Bob Bishop following him didn’t help either. The ego’s at the top fo the company, and some of the real winners we had in various roles did real damage to the company, the shareholders, and the customers.
This new CEO has a clue about what happened in the past, he understands. He must have been reading sgi.bad-attitude posts from a while back. He needs to be ready to fire anyone on the spot who does not line up with this new reality. This has been a problem in the past within SGI, and he needs to take control of it.
SGI will try to become a player in clusters again. Great. Unfortunately they are going to stick with the Intel chips only according to this announcement. This would make sense if Intel bought a piece of them, or invested in them, but as of the last set of measurements we and many others have done, this puts them at an immediate performance disadvantage in this space. They need to broaden their horizons. This in and of itself won’t be fatal, but this puts them in direct competition with Dell as an Intel only shop. Dell will likely be able to leverage better discounts and therefore better pricing. Unless Dell buys SGI or starts reselling their gear (unlikely).
They are going after enterprise customers. This is a dangerous market for them, they don’t have the background for it, and enterprise customers are more interested in stability, ability to meet needs over a long period of time, and the security of dealing with a financially sound company. Betting the company on this is dangerous. It also had been tried before at SGI, and failed miserably. My concern would be the massive defocusing that this will cause. This would need to be a highly targetted highly specific entry into one subsegment of this market.
Add to all of this that there are companies out there (who shall remain nameless for the moment) building Altix killers among others. Using non-Intel chips. Large numbers of CPUs, and lots of memory. Think terabytes. And the cluster market is highly competitive, the margins on the low end are in the 3-4% region.
SGI needs to be prepared to destroy some of their markets to create larger versions. I am not sure they are all aligned with this, and this is where the CEO is going to have to wield a rather large sledge hammer.
As the Chinese curse goes, SGI lives in interesting times.

1 thought on “Whither SGI”

Comments are closed.