The big news this week is about how AMD has managed to split itself in two. On Tuesday AMD announced that is divided itself into two companies — one that that designs semiconductors and one that fabricates semiconductors. A significant part of the deal was the injection of at least $6 billion in the two companies by two Abu Dhabi investment firms. This news represents a big change in the CPU arena. Intel is now the last fully integrated (design and fabrication) CPU company. And, I don’t believe it is bad news for AMD, rather it reflects the times.
Fabricating modern semiconductors — particularly CPU’s — is an expensive endeavor. How expensive? A modern FAB may cost upwards of $4 billion dollars. In order to justify that cost, you have to do a few things. First, you have to fully utilize the FAB. Because the cost is so high, any down-time means you are losing money. Second, the cutting edge moves rather fast. Today’s modern FAB is tomorrow’s average FAB. If you make cutting edge CPU’s, you need the latest and greatest FAB — though there are still plenty of average things that need to FAB’ed along the way. And finally, you need to sell a boat load of what ever you are producing in your FAB.
Of course, controlling your own FAB has its benefits and is probably the best way create CPU’s. Intel has successfully demonstrated this advantage since the introduction of the Woodcrest. Having the designers and processes guys at the same same lunch table is a good idea. If you are not Intel, today’s economies suggest another approach which is cost sharing. AMD will assuredly be the fabrication company’s biggest customer, but not the only one. And, with this move AMD will be be able to focus on CPU design. As I understand it, the new company is called the “Foundry Company” which must come from one of those PR think tanks. (I mean come on, even I could come up with some better names, like Foundrosity, MyFoundry, and even Foundrication. Well that last one may have some problems.) In any case, AMD now joins the ranks of NVidia, Texas Instruments, and host of other chip manufactures that do not have their own foundry.
The “Foundry Company” will be free to find more business in addition to AMD’s commitment. Some may suggest that AMD should have kept the foundry business and thus become a “Foundry Company” in its own right. I suspect that the current balance sheet and the need to finance new business units would make that almost impossible.
The need to cost share FAB’s and process technology makes good economic sense. Indeed, IBM has found that instead of working secretly in their FAB’s, opening the doors and inviting others in makes the most business sense. Of course it flies in the face of the traditional business model which one could delineate as follows:
Shoulder all development costs
Carefully protect IP
Sell an exclusive product
Sharing seems to make sense in cases where the cost to make something is expensive. In the case of open source software and Linux, this lesson has been demonstrated over and over again. A recent Information week post observes that Linux Will Be Worth $1 Billion In First 100 Days of 2009. That is, if you were to start from scratch, it would cost on the order of $1 Billion to recreate the Linux kernel. A typical shared development model might look something like:
Share development costs with others
Build products faster, cheaper, better
Sell more products
Let’s take the whole analysis a little further in an over simplified HPC scenario. Suppose you were going to create a new supercomputer from scratch. Your idea might be brilliant, but there would be many costs, the largest would include the design of a new processor, the design of an interconnect, the cost to fabricate the chips, the cost to create an OS, and finally a facility to assemble everything. Note, I assumed you were smart enough to not build your own FAB. A back of the envelope calculation tells me that you may be looking at several billion dollar investment. Let’s be wildly optimistic and assume $1 Billion. Let’s further assume that your system sells for $500 thousand (and provides a 20% competitive performance to commodity clusters in that price range). That means you need to sell 2000 systems before you start making money. The markets size for these type of systems is currently about $2-3 billion. Fortunately, in the five years it will take you to create your masterpiece the market may grow to $4-5 billion. So in order to win your billion dollar bet, you will needed to see at least a 25% market share in five years. Good luck with that.
How can one lower this cost and get into the market quicker? Well, how about licensing and modifying an existing CPU design or using an existing CPU. The same goes for the interconnect. And when it comes to the OS, there is Linux. At this point, the economics start making more sense. Sharing or leveraging what already exists may not get you the top end performance you want, but it gets you in the game a lot quicker with a lot less risk than doing it all yourself. Of course,
this is a really a history lesson.
I expect more sharing models will develop as companies see the advantages to cost sharing R&D. The take-away from the AMD split is simple. When it comes to HPC, leverage commodity and hope the FAB gets one of those game console deals.
Fatal error: Call to undefined function aa_author_bios() in /opt/apache/dms/b2b/linux-mag.com/site/www/htdocs/wp-content/themes/linuxmag/single.php on line 62