I work for a large government contracting company that deals with HPC. My question is what would one be able to expect as far as residual value for HPC technology over time? Say that a system costs on average from $10 million to $40 million, what would be the future value of these systems over the next, say, 1-5 years? I realize the answer is complex because of the introduction of new technologies and the update path that can be taken from deployed systems but I am hoping to be able to narrow it down to a reasonable percentage that can be quantified.
I suppose the resale value would be somewhat less than a new system of the same performance would cost, but the value of an installed working system continuing to do useful work may not drop so fast. For a first crack, look up some of the new system price and performance trends. Much of the HPC market value is more closely tied to performance than it is for other computer systems.
That's an interesting, but very complicated,question. I don't have a definitive answer. There are a lot of factors that drive down the value of a supercomputer over time.Tim suggested agood quantifier - performance relative to newer systems. I'd like to add some other considerations to the equation.
Let's say thatthe performance of your five-year old supercomputer is still competitive with newer systems. (Unlikely, but let's pretend.) How does the power consumption, including cooling costs, compare to newer systems? Power costs money.
System footprint is another consideration. Can the same compute capacity be installed in a smaller space? Floor space costs money too.
Here's another, less tangible consideration - system management overhead. Maybe the older supercomputer has automated management tools that increase reliability and lower the overall time spent keeping the system running correctly. A stable system does not require as much system administration time. The system administrators are free to do other things like research or revenue generating projects.
The relativesalaries of Linux, Windows, or proprietary UNIX administrators may also be a factor but I don't have any evidence to suggest that one is more expensive than another. Maybe someone else can comment on that.
The first question I would have is "the value to who?" Are we talking a value to the owner or to someone who would want to buy the system from the owner.
The value to the owner would be dependent on it's overall usefulness as Henry and Tim have mentioned in other replies.
If the owner is selling the system they might have to look at its depreciated value per the tax code. I'm not sure but I think the current code allows a computer resource to be depreciated over a 5 year period. That would mean that the residual value of the system on the books would be $0 after five years.That wouldn't be the sales price, but it would figure in the sales price. The prospective buyer would look at theoverall computational value, maintenance, etc costsjust like the original owner.
I don't know if this will help, but it's a different angle on the issue.