Welcome to iraf.net Monday, May 20 2024 @ 03:03 AM GMT
Anonymous: Guest |
09/23/1988 10:39PM (Read 571 times)
|
|
|
|
Hi Joe. I was impressed by the work you put into your benchmarks and I agree
that for many applications, the 3/60 is already fast enough and one won't
see much improvement with a Sun/4. For particularly compute intensive
operations, however, especially those which do a lot of floating point,
the Sun/4 can be a lot better (up to 8 times faster in some tests I have
done). In addition to having a faster integer cpu and lots of registers
(which favors complex loops), the Sun/4 has the Weitek floating point which
is much faster than the 68881 used in most 3/60s. Hence it wins out for
floating point intensive, cpu bound applications.Our policy here is to buy mostly Sun-3s, plus a few Sun-4's, and put those
users which need the extra compute power on the 4's. Certainly IRAF does
have applications which can make good use of the speed - it's just that these
are too complex to be used for benchmarking purposes, where one wants a
simple program so that it is easy to run the benchmark, and there can be
some hope of interpreting the results. At some point we need to revise our
benchmarking suite so that it includes some more compute intensive
benchmarks... The main reason I haven't done so is that there are already
things like the Linpack benchmark which do a pretty good job of measuring
that aspect of a system. Doug TodyPs: Good luck on figuring out your Sun-4 problem. It looks like Suzanne
has some good ideas on what to try. I will check back and see how things
are going in a couple of weeks.
|
|
|
|
| |
|