Chipsets and memory speed...

[Drew Hess]

On Tue, 12 Dec 2000, Mark Hahn wrote:

> it's interesting to consider the "efficiency" of various cpu/northbridge/dram
> combinations.  the above machine delivers 500-550 MB/s out of a theoretical
> (never-possible) 1066 MB/s: 47% efficiency.  good old PII/bx/PC100 boxes 
> would deliver 3-400 MB/s out of 800 MB/s: around the same efficiency.  
> i840 machines seem to max out at around 550 MB/s out of 3200, for a record
> of around 17% efficiency!  otoh, rumor has it that i850's deliver more like
> 1400/3200, back into the 40-50% range, and well into Alpha territory...


yup, here's mine:

-------------------------------------------------------------
This system uses 8 bytes per DOUBLE PRECISION word.
-------------------------------------------------------------
Array size = 1000000, Offset = 0
Total memory required = 22.9 MB.
Each test is run 10 times, but only
the *best* time for each is used.
-------------------------------------------------------------
Your clock granularity/precision appears to be 2 microseconds.
Each test below will take on the order of 7700 microseconds.
   (= 3850 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:        1465.2076       0.0113       0.0109       0.0118
   Scale:       1461.3146       0.0113       0.0109       0.0118
   Add:         1654.8314       0.0152       0.0145       0.0155
   Triad:       1640.9164       0.0151       0.0146       0.0154

This is a 1.4GHz Pentium 4, i850 chipset on standard Intel workstation 
mboard (D850GB) with 256MB of PC800 RDRAM, running a Linux 2.4.0-test11
kernel on Debian.  I compiled Stream with gcc 2.95.2 20000220 with "-O2
-march=pentiumpro".

The ASUS P4T might even get slightly higher numbers because it has a BIOS
option to run RDRAM devices outside the active pool in standby mode, as
opposed to the D850GB, which runs them by default in nap mode and has no
BIOS option to change it.  I'm getting a P4T soon to test.


-dwh-