>>floating point operations (but moderate memory and small i/o). The
>>RS/6000's are 2 to 4 times faster (at best). A Cray YMP achieves about
>Once again, which RS/6000? There is a substantial difference between
>a 220 and a 560...
factor of 2 to 2.5. How much faster (and more expensive) are the fastest
A bargain ba*t clone.Quote:>Also, what 3/486 system? It's not really fair comparing a
>20-processor Sequent i386 system to an RS/6000. :-)
You guess wrong! (and even a low end powerstation is a lot moreQuote:>If you mean an RS/6000 only outperforms a PC clone by a factor of 3,
>I'd guess that either your code was poorly written, you were using a
>very low-end PowerStation, or both.
expensive than a cheap 486 clone). Apparently you did not
read the footnote regarding the code. It would be easy to make it
perform at higher mflops rates but then it would take *much* more time.
Since it vectorizes to the point of reaching 100 mflops on a Cray YMP,
I *doubt* it is poorly written! This rate qualifies for the "well-optimized"
This crunching *was* in C!Quote:>The quality of the compiler has a lot to do with this as well. Even
>though I slagged on xlc in a previous post, it (and gcc for that
>matter) DO optimize code well, whereas I have no idea how good the f77
>compiler for the RS/6000 is.
The RS/6000's are very good on certain kinds of number crunching, but
they don't exactly shine on a whole range of problems.