Tell me more ×
Electrical Engineering Stack Exchange is a question and answer site for electronics and electrical engineering professionals, students, and enthusiasts. It's 100% free, no registration required.
up vote -3 down vote favorite

Assuming unlimited space and power what is the cheapest way to buy off the shelf processing power (GPUs, microprocessors, etc.)?

I am looking into doing a fluid dynamics simulation, with an artificial intelligence component to intelligently design wings/fuselage/propellers.

share|improve this question
It depends. What type of processing is it you are wanting to do? This question isn't a specific design problem that you have. Give us some details of why you are asking and it might be an OK question. – Kellenjb May 15 '12 at 2:40
The answer is still it depends. What all goes into the simulations you want to perform? Lots of multiplying? Lots of ram? What about development cost? Do you know how to program for every option out there? Some will require a deeper understanding and thus cost more in labor. – Kellenjb May 15 '12 at 3:42
2  
How many computers? If one, migrate to superuser.com. If more, migrate to serverfault.com – posipiet May 15 '12 at 4:24
1  
This question makes no sense. Suppose, you have a gazillion of CPUs, but your computations happen to not scale well. The gazillion CPUs will be idling and waiting for network exchange to finish. You can't plan for hardware until you know all the details of the computations performed. – sharptooth May 15 '12 at 5:38
@sharptooth good point. i once built a small cluster as a hobby project. the first prototype was severely limited by the network connection. you definitely need to do benchmarks, tests and probably prototypes for testing how it scales. – noah1989 May 15 '12 at 7:28

closed as off topic by Kellenjb, Leon Heller, stevenvh, tyblu, Kortuk♦ May 15 '12 at 6:10

Questions on Electrical Engineering Stack Exchange are expected to relate to electronics design within the scope defined in the FAQ. Consider editing the question or leaving comments for improvement if you believe the question can be reworded to fit within the scope. Read more about closed questions here.

1 Answer

up vote 1 down vote

From a professional standpoint, determining the optimum for a medium sized cluster is months of work.

First you have to determine the load type and devise a benchmark. This alone is no easy task and can take weeks because you have to find out what architecture best fits your computing problem.

Then you have to test all available components and test their speed and cost including space, power, cooling.

Normally you would call for bids, specifying the load case, benchmark, power and space cost. The bidders would then try to find the optimum.

For a system consisting of about 150 1-unit computers we usually spent three person-months to determine the real optimum when power and space consumption was a factor. The cost for space and power usually was in the vicinity of the purchasing price.

Since your question contains none of the required information, it is impossible to answer.

share|improve this answer

Not the answer you're looking for? Browse other questions tagged or ask your own question.