Porting software to OpenVMS AXP platform

Wladek Minor xte at mace.cc.purdue.edu
Thu Mar 11 08:59:46 EST 1993


I am afraid that you did not ported your software correctly or you did
not set-up machine correctly. I am running OSF1 and I found that ALPHA
is 3-6 times faster than DECstation 5000/240 and DECstation is ~3
times faster than vaxstation 4000/60. So you should have factor of 9
or better over VAxstation 4000/60.

Our preliminary results are following:

Virus X-ray data processing ( M. Rossmann program, over 8000
reflections) takes 58sec cpu time.  Program was not optimized yet to
run on the ALPHA. I do not have exact numbers in front of me but the
same data need 3.5 minutes on DECstation5000/240, ~3.5 minutes on IBM
RISC 6000/560 and 6 minutes on CYBER-205.

Weissenberg type of data. DENZO program. Over 4500 reflections.
Interactive mode (slower than batch mode) - 67 seconds. Again, program
was not optimized. It was just recompiled with only one line changed
(OSF1 is using 64 bit pointers - important when you mix C and Fortran).


TNT is running 4.8 times faster than on DECstation 5000/240. 

Li Yang rotational function program is:

              3.3 times faster than on DECstation 5000/240
              3.4 times faster than on IBM RISC 6000/560
             11.0 times faster than on SUN IPC

People at Yale has similar results for OPEN VMS. I will ask Zbyszek
Otwinowski to post his benchmarks for DENZO. 

If you would like to take full advantage of the ALPHA speed, you have
to RECOMPILE your programs. The MX utility to convert executables works
very nice but does not allow you to take full advantage of incredible
ALPHA speed. But even with use of 'translating utility' you should be
much much faster.


Wladek Minor
Dept. of Biological Sciences
Purdue University
xte at mace.cc.purdue.edu 




More information about the Xtal-log mailing list