[Rpcemu] Comparison with Iyonix.
jcgl at audiomisc.co.uk
Wed Apr 7 06:54:37 PDT 2010
Out of curiosity I've just tried a comparison between my Iyonix and my new
Linux laptop running RPCEmu.
To do this I used of of the 'sound' apps I've written and which have
appeared in 'Archive' magazine. I wanted something that read in data from
files and did a fair amount of double precision number bashing. So I chose
!WAV_FFTScan that does a series of FFTs and reports the results.
Did this with the same settings on the same Wave file on both systems.
On my Iyonix a series of 345 FFTs took about 5 mins. On my new (Xubuntu
9.04+ROX) laptop RPCEmu took about 10 mins. So on that basis RPCEmu was
doing the work at about half the speed of the Iyonix. Overall I was quite
impressed by that as I think it means that RPCEmu was somewhat faster than
an actual RiscPC doing the same task! :-)
One curio to report which I'd welcome comment about...
If I then compared the series of values produced I did find some occasional
The process produces a 'report' file that lists various values for each
FFT'd chunk of data. So I had 345 lines of these for each run. In almost
every case the values on the line were identical. But for a few lines one
value differed from the equivalent by one in the least significant (fourth
I'm not sure if this is a rounding difference when turning doubles into a
displayable decimal, or due to something else. Anyone have any idea or
explanation? The difference is tiny enough to be irrelevant to me, but does
mean that the two processes weren't computationally identical. (No
surprise, I guess! :-) )
FWIW All the FFTs, etc, in the app are done using double floats to minimise
errors. The input was 16 bit integer as it came from LPCM Wave. But I turn
these into doubles before any calculations are done.
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html
More information about the Rpcemu