GM performance variabilities

I performed some identical tests on three different system types to 
see how performance varied.  As a point of reference, the Intel Xeon 
system is 2 years old, the UltraSPARC system is 1 year old, and the G5 
PowerMac is brand new.  The results are very interesting.  One thing 
interesting is how the DPX read and write performance compares.  For 
the PowerMac the read and write performance is very close, but for 
UltraSPARC, and particularly Intel Xeon, the DPX read performance is 
much better than write performance.

It is disappointing that Apple's latest hot hardware produces run 
times which are similar to 2 year old Intel hardware.  The Intel 
hardware is also cheapest, and is a Dell server system.

1.2 GHz UltraSPARC (dual processor)
% gm convert -verbose sample-10bit.dpx -level '10%,.9,90%' crap.dpx
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.650u 0:01 (4.7m pixels/s)
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.280u 0:01 (10.9m pixels/s)
sample-10bit.dpx=>crap.dpx DPX 2048x1556+0+0 DirectClass 10-bit 11.4m 1.740u 0:02 (1.7m pixels/s)

2.5GHz G5 PowerMac (dual processor)
% gm convert -verbose sample-10bit.dpx -level '10%,.9,90%' crap.dpx
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.380u 0:01 (7.8m pixels/s)
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.090u 0:01 (27.6m pixels/s)
sample-10bit.dpx=>crap.dpx DPX 2048x1556+0+0 DirectClass 10-bit 11.4m 0.440u 0:01 (6.6m pixels/s)

2.4GHz Intel Xeon (dual processor)
% gm convert -verbose sample-10bit.dpx -level '10%,.9,90%' crap.dpx
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.258u 0:01 (12.2m pixels/s)
sample-10bit.dpx DPX 2048x1556+0+0 DirectClass 10-bit 12.2m 0.094u 0:01 (32.4m pixels/s)
sample-10bit.dpx=>crap.dpx DPX 2048x1556+0+0 DirectClass 10-bit 11.4m 0.680u 0:01 (4.4m pixels/s)

These image rotation numbers are interesting:

SPARC
gm convert -verbose sample-10bit.dpx -rotate 45 crap.dpx  15.22s user 0.83s system 98% cpu 16.347 total

G5 PowerMac
gm convert -verbose sample-10bit.dpx -rotate 45 crap.dpx  12.20s user 1.00s system 100% cpu 13.143 total

Intel Xeon
gm convert -verbose sample-10bit.dpx -rotate 45 crap.dpx  10.33s user 0.98s system 99% cpu 11.412 total

Which shows that SPARC closed the gap, but Intel Xeon is still 
fastest.

And some image scaling numbers:

SPARC
gm convert -verbose sample-10bit.dpx -resize 200% crap.dpx  20.51s user 0.54s system 98% cpu 21.290 total

G5 PowerMac
gm convert -verbose sample-10bit.dpx -resize '200%' crap.dpx  13.38s user 0.54s system 98% cpu 14.062 total

Intel Xeon
gm convert -verbose sample-10bit.dpx -resize '200%' crap.dpx  10.23s user 0.57s system 99% cpu 10.851 total

======================================
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/


-------------------------------------------------------
This SF.Net email is sponsored by: New Crystal Reports XI.
Version 11 adds new functionality designed to reduce time involved in
creating, integrating, and deploying reporting solutions. Free runtime info,
new features, or free trial, at: http://www.businessobjects.com/devxi/728