I haven’t said anything about SGI recently, been too busy with other things. This is good, but much has been happening in SGI land.
First, I had wondered whether or not SGI stake holders would get anything out of the company … that is, whether or not it would survive the bankruptcy proceedings. It does appear to be navigating those waters well, but it is worth noting a few things in case you haven’t followed the story.
When SGI went chapter 11, all common stock (equity) was canceled. If you held any notable amount of stock, you now have a nice little write down for next years taxes. Not fun. I know several friends that had several thousands to tens of thousands of shares that they had purchased while at SGI. I worked with one of the original SGI SE’s who had been a millionare based upon the value of his shares. I sold off all of mine before I left SGI, so while I did lose money, I didn’t lose all of it.
That is bad, but in any bankruptcy proceedings, the question on everyone’s mind is not what happens to the equity share holders, that is an answered question. The question that the courts care about is what happens to the debt holders? SGI was in the hole for 345M$ US. Chapter 11 seeks to protect the company from the people who own the 345M$ that SGI borrowed. SGI is basically trying to negotiate a lower debt, that is, to get the right to have bought 1$ and return much less than 1$. So how many pennies on the dollar of debt does SGI get to pay? About 20.3.
Yup. They got their creditors to agree to shave about 4/5 of the value of the debt off against guarantees that they will pay. See this article.
I am glad I am not a debt holder for SGI.
Ok, that it? Well … no.
SGI had to get out of Dodge, or in this case, its very nice facilities at the ATC as part of the Chapter 11 settlement.
Ok, that it?
No. SGI is entering the cluster market. Just like it did 6 years ago. Which it subsequently left 5 years ago. Anyone remember SGI 1100’s?
That means … it is competing with Dell, HP, IBM, Sun as top name suppliers, LNXI, Penguin, Appro, … as 2nd tier suppliers, and lots of smaller highly specialized shops like Scalable. It is competing using Woodcrest. Note my earlier benchmarking on woodcrest. Since it cannot leverage the volumes of the other folks, it won’t get the discounts of the other folks. Which puts it in a difficult price crunch.
The only possible thing worse than that that I could imagine, is if they decided to use 32 bit processors in the Intel line (the 5000 series), rather than the Woodcrest series (5100). That would be … an error of monumental proportions, not one they could easily recover from. Given where they are financially, this would be terminal for that product.
It is a shame that they didn’t go Opteron though. My impressions of the woodcrest are that it is nice but not an obviously better platform than its competitor. This is unlike the impression that Opteron gave me 2 years ago. Woodcrest has some nice features. I would like to see more performance delta though, and more specifically, it would be great if older code ran faster. This is what made Opteron sell well in the first place, that it was a faster Xeon than a Xeon without recompilation, and you could get even more benefit by recompiling. Here there is a performance detriment from what we have seen to date unless you recompile, and that is with a modern compiler at that.
Add to this the HTX bit… it would be natural IMO for SGI to integrate an HTX to NUMAlink bridge. HT3 is much faster, at least on paper, than PCIe: you would need something like PCIe x64, yes, 64 lanes, to match on paper what HT3 will give, and you still won’t get the latency advantages.
I hope SGI emerges. They just need to be aware that the market they are emerging into is not one that would be called friendly. Or forgiving.
Will they survive even if they emerge? They need to do their product transition quickly into something that has a fighting chance of selling in volume. The new cluster is a step in the right direction, but they need to ditch some of their older baggage. Fast.