[rescue] Advice on Octanes

[Joshua D. Boyd]

On Wed, Sep 24, 2003 at 01:09:01AM -0400, Dave McGuire wrote:

>   I do have a great deal of respect for your knowledge, but please keep 
> in mind the fact that getting a PhD proves that you can please a 
> professor...not that you know what you're talking about.

Actually, I believe it means that you can please several professors.

> > SGI Octane2/R12k-400 :  514.1 MBps
> >
> > Pentium III/733-CM(pgi) :  685.7 MBps
 
>   In this case, however, having put them side by side, I've seen very 
> different results.  As have several others here.  PhD or not, "them's 
> the facts".

It is worth knowing exactly what task is done under each case.  I find
my Octane amazingly faster for some tasks, and somewhat sluggish for
others.  Apparently Francisco is talking about a raw memory bandwidth
test.  I'm guessing doing sequential bulk transfers.  Perhaps just
straight copying one large array to another and timing how long it takes
to do it.  Perhaps doing it using memcpy and assuming that that function
will just do the right thing (which is reasonable in my book).

According to these numbers he isn't getting the performance that SGI
claims their memory system is capable of.  Perhaps it is the R10k being
unable to keep up with the memory here?  I don't know that much about
how the Octane is set up in this way, but it appears that the SGI is
getting half the bandwidth SGI claims while the P3 is getting fairly
close to the performance typically claimed for PC133 memory.  I wonder
why.  Could expand on your results more Francisco please?