Bang for the Buck, comparing Celerons to Xeons (and PIIIs?)

[Alan DuBoff]

I was comparing 2 different SMP machines just now.

On both machines I'm running Debian potato which was updated, dist-upgrade'd,
and upgraded before starting the test. Each machine is running the latest
compilers, 2.95.1 from unstable.

Machine A ('da hotrod)
---------
ABit BP6 motherboard
Dual Celeron 366mhz CPUs overclocked to 550mhz (128k L2 cache)
256mb memory (CAS 2)
Symbios 895 controller (U2W)
U2W 7200rpm drive

Machine B ('da cruiser)
---------
SuperMicro S2DGE motherboard
Dual 550mhz Xeons (512k L2 cache)
256mb memory (CAS 3)
Symbios 895 controller (U2W)
U2W 10000rpm drive

As you can see, both machine are similar, but not identical. Memory is
actually faster in the Dual Celeron, but disk is slower. Memory is much easier
to swap out actually, it's marginally faster for this type of process. Cache
is the area I think would be where cycles are spent.

The test is pretty simple. I have a script which I will include below, which
compiles the kernel 15 times, and spits out some time as it does it. This is a
typical "real world" process, compiling the kernel.

The original intention was to run the kernel compilation for 15 revs to see if
heat could be built up on the Celeron CPUs. I have high performance cooling
heatsinks and fans on my Celerons, which were purchased from a company that
specializes in overclocking the Celerons. They run pretty good, but not as
good as the Xeons. As I started doing this, I decided to give it a run on my
Dual Xeon box and compare the time it takes to compile the 2.2.12 kernel. An
areas that is commonly complained with overclocking Celerons is when heat
builds up the machine will reboot mysteryously. With the ABit mobos you can
select the voltage which can help in this case, but I leave mine at 2.0 volts
as they run fine.

If anyone else has a 2.2.12 kernel, especially with a 550mhz PIII (single or
dual) I would be most curious to compare numbers...The PIII has 512k L2 cache,
but it runs at half the processor speed.

Machine A
---------
~4m25.000s (for a compile rev)

Machine B
---------
~3m30.000s (for a compile rev) 


Here is the script I was using
------------------------------

#!/bin/bash
rm ./comp15.log
time make clean >> comp15.log
echo "About to make Rev1"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev1"
echo "About to make Rev2"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev2"
echo "About to make Rev3"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev3"
echo "About to make Rev4"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev4"
echo "About to make Rev5"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev5"
echo "About to make Rev6"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev6"
echo "About to make Rev7"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev7"
echo "About to make Rev8"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev8"
echo "About to make Rev9"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev9"
echo "About to make Rev10"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev10"
echo "About to make Rev11"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev11"
echo "About to make Rev12"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev12"
echo "About to make Rev13"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev13"
echo "About to make Rev14"
time make >> comp15.log
time make clean >> comp15.log
echo "Done with Rev14"
echo "About to make Rev15"
time make >> comp15.log
time make clean >> comp15.log
echo "Finished! (15 Revs)"

-- 

Alan DuBoff
Software Orchestration, Inc.