🚀 go-pugleaf

In article <1993Jan10.232804.1030@qualcomm.com> antonio@qualcomm.com (Franklin Antonio) writes:
>I haven't seen any performance benchmark data on alpha-based machines
>posted by users.  Has someone done some measurements of speed of some
>program you've ported to an alpha-based machine vs same running on some
>model of VAX?  ..or even some other brand?

we managed to borrow an AXP 3000/500 for the two weeks ending Jan 6th.
ooohhhh... although most of our people we gone for the holidays, the few
 who remained reported fifty times faster performand than a VAXstation 3100/38,
 and five to seven times faster than an HP 9000/710.
These were biggy Fortran number-crunch, with lots of transcendental-routine
 calling deep in the loops. (which kills HP's)
The machines in question were: VAXstation 3100/38, 8mb
  Alpha AXP  3000/500, 96mb, dual 400mb disk (the OpSys and Fortran FILLED one)
  HP 9000/710, 48mb, 1Gb disk.
The program which got the most use repetitively runs thru a 50k block file.
 The Alpha showed 30% idle-time , which i assume was waiting for the disk.
Other users also reported the 50x factor over the 3100.
We also tried a program against an 11/780.  Due to an undiscovered bug,
 we had to run it non-optimized on the ALPHA.  it was still 68x faster.

But- if you want an Alpha to reaaaallllyyyy  drag, try WATCH under DEBUG.
They apparently don't have the memory-locking hardware trap a VAX does.
They -trace- every instruction as it executes to see if it modifies the 
 WATCHed locations.  very, very, VERY slow...

We also ran into a Open/Closed VMS difference: the LIB$FIXUP_FLT and its
 kin don't work under OpenVMS AXP (which is explained in HELP.. VERY good
 cross-platform HELP information... kudos to DEC)
We also tried joining it to our cluster (of course!), but LMF got upset.
It -did- manage to join (for a few seconds), and the cluster manager 
complained about differing VMS version numbers (what? OpenVMS V1.0 not 
the same as A5.5-2??)
--dick (i want one (or two, or three...))