How to calculate computer speed

Question

I have a high-speed ADC data capture/analysis program which performs poorly on older computers. This was discovered in beta testing in my customer's lab when one test engineer reported the application was hanging. It turns out that she had an old computer in her lab (single core P4), and the 'hang' was the computer taking a very long time to complete some calculations.

I would like to calculate "computing power" at startup, and warn the customer that some functions will be very slow if the calculated power is below some cut-off point. Note that CPU speed isn't what I'm after (the P4 was running at 2.4 GHz).

I thought that I could just get the CPU family/model/stepping and show the warning if the family/model was below some cut-off point, but I don't think this approach is workable because the family for the P4 is higher than the family for, say, an i7. Using a table is out because the table will have to be maintained.

I could use a benchmark algorithm such as whetstone/Dhrystone/whatever, but I don't want to add any more time to startup than necessary.

Is there another way I can accomplish this without taking an inordinate amount of time at start-up?

TIA

Answer 1

It's probably best to run a small actual calculation (with some included sample data) and time the operation.

As for the startup time concern, you could either do the testing during the installer, or run it on startup only the first time that it hasn't been run (and then store a flag indicating that it has been run).

Answer 2

Rather than running a check at start-up, create an install for your application and run the check and requisite warnings then.

I agree maintaining a table of suitable and unsuitable chips would be a pain. I would stick with which ever benchmark algorithm most closely tracks with the type of calculations your program will be performing and set a minimum score that will prevent installation.

Answer 3

Have a progress bar or similar so that there's some feedback that the app hasn't crashed.

(e.g. in the future your computer may be twice as fast but your dataset 30 times larger. The computer that you once considered fast now looks like it's died)

Answer 4

In combination with some of the ideas of running a little benchmark function of your own using your real calculation and sample data, I think it would be clever to offer the user the chance to upload the results to your web site. Your web site could then display charts showing the results of many combinations of CPU type and speed. That would enable new users of your program to know what sort of system they need before installing the program.

Answer 5

If we designed the front-end (UI) and the back-end (functionality) in the right way, even if the calculation is slow, neither the program, nor the computer shall hang.

I would suggest that you use multi-threading to overcome this hang issue, rather than the warning, which you can be sure that most of the people will just ignore, till they see the actual hang.

This answer may be way off your question, but just a suggestion.