ServerWorks chipset

Don Holmgren djholm at fnal.gov
Thu Oct 12 19:51:27 PDT 2000


In case this helps...

I have a pre-release server under NDA.  I'm not positive if this is the
chipset you're interested in.  The output from lspci:


00:00.0 Host bridge: Relience Computer CNB20HE (rev 06)
        Flags: bus master, medium devsel, latency 48

00:00.1 Host bridge: Relience Computer CNB20HE (rev 06)
        Flags: bus master, medium devsel, latency 48



Here's a streams run (I have a modified streams which uses the Intel cpu
clock, in case you're wondering about the extra lines.  The higher
precision from this clock gives more accurate results w/out resorting
to huge array sizes):


lrep3:~/benchmarks/streams$ ./streams
-------------------------------------------------------------
This system uses 8 bytes per DOUBLE PRECISION word.
-------------------------------------------------------------
Array size = 4000000, Offset = 0
Total memory required = 91.6 MB.
Each test is run 10 times, but only
the *best* time for each is used.
-------------------------------------------------------------
Cycles/second = 731012878.228893
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 119432 microseconds.
   (= 119432 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function      Rate (MB/s)   RMS time     Min time     Max time
Copy:         405.2265       0.1580       0.1579       0.1584
Scale:        403.5475       0.1586       0.1586       0.1587
Add:          512.4053       0.1874       0.1874       0.1874
Triad:        353.2453       0.2718       0.2718       0.2719


Currently the machine has 4 128-MB 133MHz DIMMs.  Sorry, I don't have
the precise timings at hand.  I'd originally run streams w/only 1 DIMM
installed, and got a disappointing number (about 350 for Copy, I think).
I'd guessed the machine had interleaving when I added the additional
DIMMs.  I'm away from the lab until Monday, but I'd be happy to report
additional measurements.

Don Holmgren
Fermilab




On Wed, 11 Oct 2000, Greg Lindahl wrote:

> > I will check on this, but as far as I know the board is not shipping.
> > Can you actaully buy one from this company?
> 
> When someone does get one, PLEASE run stream on it. We have no idea if the
> interleaving gives any significant benefit.
> 
> -- g
> 
> 
> _______________________________________________
> Beowulf mailing list
> Beowulf at beowulf.org
> http://www.beowulf.org/mailman/listinfo/beowulf
> 





More information about the Beowulf mailing list