View previous topic :: View next topic |
Author |
Message |
colt1954
Joined: 21 Dec 2010 Posts: 81
|
Posted: Wed Jan 26, 2011 4:18 pm Post subject: Power/speed of Fortran on modern PC's |
|
|
Sorry I posted this under a wrong section:
Have just resurrected an old fortran engine vibration programme of mine.
Amongst many subroutine pre and post preperation calculations it calls an FFT subroutine (based on Cooley Tukey algorithm) and does an FFT on 12 signals (2 x 6 DOF) each containing upto 65536 data points (2^16), it does this 6 times (speed increments), thats 4.7 million data points. But it takes merely seconds on the screen.
Ok perhaps its also a testament to modern PC's also !
Doing an FFT in Excel on a one signal of only 4096 data points takes c2 minutes !!! |
|
Back to top |
|
|
Robert
Joined: 29 Nov 2006 Posts: 445 Location: Manchester
|
Posted: Thu Jan 27, 2011 8:22 pm Post subject: |
|
|
I am determined to have my reply...
Modern PC CPUs are astonishing. Sadly, a lot of their brute force is hidden by modern UIs. Occasionally we get a glimpse |
|
Back to top |
|
|
JohnCampbell
Joined: 16 Feb 2006 Posts: 2554 Location: Sydney
|
Posted: Thu Jan 27, 2011 11:57 pm Post subject: |
|
|
I studied computer science when it was more numerical methods. I've never forgotten the lecture on Cooley Tukey FFT. It was at the time I was also learning (slow) FT in pure maths. I must have missed the tutorial, as I never managed to understand how I could use them. ( Isn't FFT plural, we only had one transformation in pure maths ?)
One of those things from the past I felt I should have understood.
I'm pleased to hear someone actually used this algorithm.
John |
|
Back to top |
|
|
DrTip
Joined: 01 Aug 2006 Posts: 74 Location: Manchester
|
Posted: Fri Jan 28, 2011 3:14 pm Post subject: |
|
|
I used Fast Fourier Transforms all the time when I was doing my Phd work analysing time series and spatial structures. I am likely to start doing so again when I start my new job next week.
but I have never been able to quite work out what the algorithm is (i.e to successfully code my self and understand what it is doing, I always just used matlab or some other library). I can do a simple DFT in 10 lines of code or so (sines and cosines etc). But the FFT has some some clever stuff that has always proved to be beyond me (given I never had to do it!)
A divide and conquer approach is used split the array into odd and even cells and repeat recursively. Wikipedia talks about this recursive definition but apparently this is not normally actually done.
The Matlab FFT is an implementation of the FFTW (the fastest fourier transform in the west!) Apparently there are people who just spend all day trying to make the FFT even faster. Which sounds like a pretty cool job!
Carl |
|
Back to top |
|
|
jjgermis
Joined: 21 Jun 2006 Posts: 404 Location: Nürnberg, Germany
|
Posted: Wed Feb 02, 2011 6:12 pm Post subject: |
|
|
We also use the DFT all the time. I never heard the name Cooley-Turkey FFT. So I had a look in my numerical recipes in Fortran. There is a small section on other FFT algorithms where this is explained - the NR book is available online.
I also (did) use Matlab for a while to do the DFT stuff. Since I discoved ClearWin+ I rather prefer to do everthing in Fortran - very fast and very effective With NR one gets really cool code. |
|
Back to top |
|
|
|