[nSLUG] Software Freedom Day
jwark at eastlink.ca
jwark at eastlink.ca
Sun Sep 13 15:09:24 ADT 2009
It was not just the visual feedback provided by gkrellm. The CPU temperature
was definitely higher and this caused the fan to run faster, so that there was a
definite audible sound increase, indicating that the CPU was working harder
with the 32bit distros.
I am too old to want to go investigating hardware architecture details, but I
believe that the data path widths with 64bit systems are 64 bits wide instead of
the 32 bits with 32bit systems. That means that data flow rates between
memory and CPU are higher with 64bit systems. To me, that would adequately
account for the observed reduction in CPU utilization with the 64bit distro.
I don't see why 64bit pointers would have a slow-down effect. With 64bit pointer
fetch the time to fetch a 64bit pointer would not be greater than for a 32bit
pointer to be fetched in a 32bit system.
----- Original Message -----
From: Ian Campbell <ian at slu.ms>
Date: Saturday, September 12, 2009 9:14 pm
Subject: Re: [nSLUG] Software Freedom Day
> On Sat, Sep 12, 2009 at 08:54:25PM -0300, jwark at eastlink.ca wrote:
> > I have a laptop with an AMD64 processor and 1GB of RAM on which
> > I run Debian 64bit. I use gkrellm to monitor the system. I have
> tried> various 32bit distros on this system and find that CPU
> utiliztion is
> > significantly *less* with the 64bit distro.
> Saying I'm skeptical is perhaps a bit harsh given that... I mean
> you're seeing what you're seeing, but this:
> shows approximately what you'd (general you, not you specifically I
> suppose) expect.
> 64bit gives you more address space and more general purpose registers.
> The latter, under some circumstances, *can* give you a performance
> boost although in most cases it won't. As you can see from the charts,
> GCC gets a bump, with everything everything else it has a minimal
> effect (positive or negative.)
> On the other hand, the former can easily worsen performance. Pointers
> are now twice the size, with predictable effects... I have a client
> with a perl script that fetches all of a poorly designed database
> (think hundreds of columns, tens of thousands of rows) into an array
> of hashes. The machine he was on was replaced with a 64bit install and
> his RAM usage absolutely exploded.
> I'm not sure why you would see a lower load.
> Actually that's not true, one possibility is that the x86 packages for
> most distros are compiled with a 486 or 686 target, which means none
> of the newer instruction set extensions. x86_64 guarantees that at
> least SSE and SSE2 etc. will be there.
> nSLUG mailing list
> nSLUG at nslug.ns.ca
More information about the nSLUG