Factlet: Bubba2 calculating speed

jws · Post by **jws** » 28 Oct 2008, 10:18

Get pi.c from http://www.jw-stumpel.nl/pi.html

Do

cc -o pi -O2 pi.c

(in the -O2 optimisation directive, O is a capital letter O, not a zero).

You may need to apt-get install gcc on the bubba2 before this works.

Do this both on Bubba2 and on a Linux desktop machine.

Then run on both machines

Code: Select all

time ./pi

You'll get pi to ten thousand decimals. My results:

Bubba2: 6.6 seconds
Pentium 4: 0.67 seconds

So the Bubba2 is almost exactly 10 times slower than a not-so-very-modern desktop. Of course this is fine, because the Bubba2 is not a calculating monster, but a home server, aimed at minimising energy use and noise. What results do you get? Bubba 1 machines? Other desktops?

Clive · Post by **Clive** » 29 Oct 2008, 05:27

EEK !

My Bubba 1 takes 1m31 seconds to do this

Guess my Bubba is not suitable for heavy number crunching yet !

Hammer · Post by **Hammer** » 29 Oct 2008, 09:13

Bubba1 is now used as paper holder, i.e. put on top of a stack of papers to keep them from flying away in the speed draft of B2...

jws · Post by **jws** » 29 Oct 2008, 10:17

Hammer wrote:Bubba1 is now used as paper holder, i.e. put on top of a stack of papers to keep them from flying away in the speed draft of B2...

Did you get the same calculation time as Clive got? That seems very slow indeed.

Post by **DanielM** » 29 Oct 2008, 14:23

1.20 on my B1.

/Daniel

ahoff · Post by **ahoff** » 29 Oct 2008, 15:20

B2 = 5.88 s
Athlon 1.8 = 0.87 s

/ahoff

pa · Post by pa » 29 Oct 2008, 17:44

Hi guys,

Bubba 1 is as you have noted very slow at calculating pi. The reason for this is not only the clockspeed, but more that B1 lacks a floatingpoint "co-processor".
So comparing B1 and B2 in this case is actually not fair for B1, floating point calculations is not used that much.

/PA

Cheeseboy · Post by **Cheeseboy** » 29 Oct 2008, 17:56

I also got 1min 30 secs on B1.
The old one with only 64MB memory, and it was quite busy too...
Nice.

It will not become a paperweight (once I have sorted out the issues with B2/my network and it takes over), it will be a backup station.
I just wish I could start it with Wake On Lan...

jws · Post by **jws** » 30 Oct 2008, 05:20

pa wrote:Hi guys,

Bubba 1 is as you have noted very slow at calculating pi. The reason for this is not only the clockspeed, but more that B1 lacks a floatingpoint "co-processor".
So comparing B1 and B2 in this case is actually not fair for B1, floating point calculations is not used that much.

The pi program uses only integer calculations, so that can't be it. But I read somewhere that the ARM has no (integer) divide instruction (it has a multiply instruction). Division is done in software with shifts and subtractions. The pi program does lots of divisions, so that might explain the speed difference.

Clive · Post by **Clive** » 30 Oct 2008, 05:51

I also found a small C programme to calculate prime numbers (remember to remove the line breaks if you copy+paste into vi or whatever) -

http://bellard.org/mersenne.html

This took over 10 hours on my Bubba 1 ! I had a similar programme on my 1MB 286 computer years ago and that took 4-5 days to compute something similar, so regardless of slow Bubba, things are still alot quicker then they were....

jws · Post by **jws** » 30 Oct 2008, 07:26

Clive wrote:I also found a small C programme to calculate prime numbers (remember to remove the line breaks if you copy+paste into vi or whatever) -

http://bellard.org/mersenne.html

This took over 10 hours on my Bubba 1 ! I had a similar programme on my 1MB 286 computer years ago and that took 4-5 days to compute something similar, so regardless of slow Bubba, things are still alot quicker then they were....

Oh yes. My own 286 took many hours to run the pi program. I had to let it run overnight to see the results. It is in fact amazing how quickly the speed of processors has improved.

Eek · Post by **Eek** » 30 Oct 2008, 15:10

B2: 0m5.981s
B1: 1m21.222s