Bear Flag Republic Radio Weblog

InfoWorld.  IBM's $10 b gamble with on-demand computing.  Although it is not explained in the article, the idea is to build a software infrastructure that can "flip" to handle a myriad of tasks.  It also includes an interesting twist.  Given that desktop PCs represent the majority of majority of most company's computational horsepower, IBM plans to slave them to a Grid computing architecture.  So, an enterprise with 100,000 desktop PCs and 5,000 servers would effectively improve by an order of magnitude the amount of work that can be done.

If they can pull this off, it will be good news to Intel since it would provide a basis for companies to continue to upgrade their PCs.  It would also enable IBM to infringe on Microsoft's turf by getting more per desktop than Microsoft gets from companies that adopt this strategy.  Is there a broad-based play here that could go beyond SETI at-home and other freebie projects?  There might be if IBM or someone else can find a way to pay individuals, corporations, and universities for the use of their spare cycles.  How much would the spare cycles on a 2.5 GHz desktop be worth?  The key to this may be to look at the cost of replicating this capability in-house.  The cost to purchase, house, and power a similar machine could easily cost $4,000 a year over three years (two cycles of Moore's Law).  So, given this target cost, would $400 a year in "rental" fees be possible?  It may be.  This would effectively make buying a computer a freebie.  In this model, $4 m a year would buy you the majority of cycles on 10,000 PCs each able to provide multi-gigaflop performance.

[John Robb's Radio Weblog]
comments < 6:27:05 PM        >