Dear All
I'm busy looking at an Alpha PC AXPpci33 (an DEC CPU board put in a box
by a supplier), and I think its a 100MHz Alpha chip. It's supposed to
be slightly faster than a 4/100 AlphaStation.
I'm running 3 different numerical models on it:
1) Fish Model with a fair bit of I/O to disk.
2) Water Column model, mainly CPU bound.
3) An iterative numerical problem, CPU bound, spends 97% of its
time in a 10 line procedure, but does have X based graphical
output.
I'm also running these on a AlphaStation 200 4/233 as a comparison.
Here are the runtimes from 'time':
pci33 4/233 Speed Up
1) Fish 82 27 3 times
2) Water 30 10 3 times
3) Iter 230 218 about 5% !!!
As you can see the Fish and Water models run 3 times faster on
the 4/233 (as expected), but for some reason the Iter model
runs at a remarkable pace on the pci33 (it isn't the 4/233
running it slow since I tried it on a 4/100 and it had a run time
of about 600 secs).
What is going on ? I've disable the graphical output from the model-
no improvement. I've ran 'pixie' and looked at the stats - both
take roughly the same number of CPU cycles yet the slower pci33
almost matches the 4/233 - how can it ?
Investigatng further, it seems that the problem is to do with
Pascal - models 1) and 2) are Fortran, model 3) is Pascal.
A simple 20 line pascal program runs roughly the same on both
machines, but the same program converted to C runs at half the
speed on the pci33. Both machines are running version 5.20 of
the Pascal runtime library, with everything compiled on the
4/233 with version 5.20 of the compiler.
I appreciate there is lots of information I haven't given you,
but if anyone has any ideas about where to look for the source
of the problem, please let me know!
Many thanks
Ian
Received on Wed Nov 22 1995 - 12:22:57 NZDT