[Wien] Machines

[Pablo de la Mora]

Dear WIENusers,
    I have some money to buy good machine or few P4 boxes to make a 
small cluster. With my colleagues I was able to try a Xeon (3.0GHz) with 
two processors with HT (Hyper Threathing), a P4 (2.8GHz) with one 
processor with HT and some not so new P3 (1.3GHz), these last in a 
cluster and each machine with two processors.
    I did a test in these machines, I loaded the latest WIEN2k with the 
PRECOMPILED programs and ran a simple system with many k-points, the 
results were quite interesting;
-------------------------------------------------------------------
-1-, -2-, etc. refers to the number of lines in the '.machines'
-------------------------------------------------------------------
    In the P3 at 1.3GHz
times:

--1--
lapw1: 92:14
lapw2: 11:14

--2-- one machine
lapw1: 74:14
lapw2:  6:20

--3-- one machine
lapw1: 73:26
lapw2:  6:29

--2-- two machines
lapw1: 66:26
lapw2:  5:48

relative times between:

-- 1 and 2 -- one machine
lapw1: 1.78
lapw2: 1.77


-- 1 and 2 -- two machines
lapw1: 1.98
lapw2: 2.01

------------------------------
P4 at 2.8 GHz

times:

--1--
lapw1: 71:36
lapw2:  6:32

--2--
lapw1: 59:29
lapw2:  5:57

--3--
lapw1: 57:57
lapw2:  5:49


relative times between:

-- 1 and 2 --
lapw1: 1.20
lapw2: 1.10

------------------------------------
Xeon at 3.0GHz
times:

--1--
lapw1: 86:52
lapw2:  7:48

--2--
lapw1: 55:50
lapw2:  5:36

--4--
lapw1: 41:21
lapw2:  7:48

--6--
lapw1: 39:35
lapw2:  3:52

relative times between:

-- 1 and 2 --
lapw1: 1.56
lapw2: 1.39

-- 2 and 4 --
lapw1: 1.35
lapw2: 1.45

------------------------------------
The results show that;
The Xeon is slower than the P4, and the not so new P3 are relatively 
faster than P4.
The parallelism in the Xeon is not so good
The HT (Hyper threading) does not duplicate the processor, that is, it 
does not work as two processors.
The performance in the P3 is better when the load is distributed in two 
different machines compared to the case when it is calculated with the 
two processors in the same machine.
-----------------------------------
So my questions are:
    If I compile the programs would I get a better relative speed?
    What is better, one good machine or several cheaper P4?
    Dual or single processors?

       Yours

             Pablo de la Mora

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: P3-log
Url: http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20050613/ba5a1ba2/P3-log.ksh
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: P4-log
Url: http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20050613/ba5a1ba2/P4-log.ksh
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: Xeon-log
Url: http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20050613/ba5a1ba2/Xeon-log.ksh