Skip to content

No performance increase from dual-core processors?

2 messages · Adam Petrie, Brian Ripley

Adam Petrie
# 
Hey all,

I was planning on getting a new computer for the new year to help with 
my dissertation research, and recently had an opportunity to compare the 
performance of my 1.6 GHz Pentium M laptop, and a 2.8 GHz dual-core 
Pentium processor (both running WinXP professional 32-bit).  I run a lot 
of long simulations, so I was hoping to get something that would speed 
them up.  I ran a few quick computation tests (e.g.  generating 500,000 
normals), and found the performance increase of the 2.8 dual-core over 
my 1.6M laptop to be negligable (and in fact sometimes slower).  One 
thing I did notice that if I look at the CPU usage of my laptop when 
it's performing the simulations the laptop is at about 100%, while the 
dual-core reaches 50% (and seems to refuse to go higher than 50% no 
matter what).  Of course if I load up 2 instances of R I can get them to 
run the simulations simultaneously in about the same amount of time, but 
this doesn't help me get to the end of a very long simulation quicker.  
This got me thinking that no matter what kind of processor I get, I'm 
not going to be getting a large speed increase over what I already 
have.  Does anyone have any insight into various setups/processors that 
would help me speed up my work (e.g. maybe run R through linux, Pentium 
extreme edition, etc)?

Thanks in advance,

Adam Petrie
Rensselaer Polytechnic Institute
Brian Ripley
# 
R only runs multiple computational threads as part of a BLAS/LAPACK addon 
on Unix-alikes, so no speed-up is expected from dual processors (which 
includes Intel's HyperThreading, as well as dual-cored systems).

A faster processor would give you considerable increases even single-core,
and a 2.8GHz Pentium is quite slow compared to an Athlon64 or Opteron
(as you have found: P4 systems are slow for their clock speeds compared to 
PIIIs (such as Pentium M) or almost anything else).  My 2GHz Pentium M 
laptop is faster than my 2.8GHz P4 home desktop, and an Opteron 252 is 
considerably faster again (even running a 64-bit OS and build of R).

BTW, testing 500,000 normals is a single R command and is for me fast 
enough to be hard to time accurately:
[1] 0.10 0.01 0.10 0.00 0.00

on an Opteron 252.
On Mon, 26 Dec 2005, Adam Petrie wrote: