[OT] Maximum performance on single processor ?
landman at scalableinformatics.com
Fri Jun 20 09:11:27 EDT 2003
Marc Baaden wrote:
> Right now, even on simple test cases, the "response time" (eg calculation
> time for a single step) of our program is on the order of the second.
> (this is for an athlon MP 2600+)
> We need to get that down to a fraction of seconds, best milli-seconds,
> in order to be usable in real time. (makes it a factor of roughly 1000)
You are not going to get a 3 order of magnitude increase in performance
by switching CPUs. You won't get an order of magnitude over the Athlon.
> So my question was more, what is the fastest CPU I can get for $20000
> at the moment (without explicitly parallelizing, hyperthreading or
> vectorizing my code).
Before you pursue this any more, lets try to reset expectations to be
more in line with what would need to occur.
1) the CPUs are generally within single digit factors of performance on
a given code, and this is as much dominated by memory speed and/or
latency for various codes, as it could be by CPU "speed"
2) As it does not appear that you have profiled the code (just a guess),
I would strongly urge you do that as your next step. This will tell you
(if you chose not to optimize your code) where the code is spending most
of its time. If this is in fetching data from memory then a "faster"
CPU will not help, as the CPU is not the bottleneck.
3) to get multiple orders of magnitude change in performance, you will
need at minimum an algorithm shift. The algorithms used in your code
now are taking time, and you need to know how much time. It would be
terrible if your code takes 1/2 second to set up the calculation in the
first place, and another 1/2 second to perform it. In this case, the
first half second would impede any faster algorithm. If you comment out
the computation, how long does the setup/teardown take for the code?
Though you don't want to work on this code, I am not sure you will be
able to get more than some percentage points with a new CPU without at
minimum a profile of the code, and a hard look at what it is doing, and
how it is doing it.
> Marc Baaden
Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://scalableinformatics.com
phone: +1 734 612 4615
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf