Re: work-load imbalance

From: Luca Tornatore <torna_at_sissa.it>
Date: Thu, 2 Nov 2006 18:02:21 +0100

Hi Pedro,

this was just my first little idea.. actually i don't know how the opteron are
built, and it actually sounds a little bit stupid to built-up a system with
that bottleneck (though some of the first dual core had this kind of
drawback).

of course "dual core" roughly means "double procs".. just "roughly" because
actually it means "two processing units", that both resides on the same chip.
for this reason, the two cores share a number of services: some dual core
chips implement a dual channel access to the memory, and some others do not.
In case the two cores are forced to use the same channel (or whatever like) to
access the ram, the net result is a bottleneck in the performances.
in spite of this, due to the increased calculus power, your gain would be more
than simply a factor of 2 rescaled by the GHz.

one more tech hint. i see you have ddr1 ram _at_ 400 MHz, which is not a very
fast chip. A faster memory chip on the dual opteron would partially account
for the lack of gain

cheers
:)
luca

On Thursday 02 November 2006 17:35, Pedro Colin wrote:
> Hi Luca,
>
> > I mean, although your procs are dual core, you have actually 4 processing
> > units (and that's the factor ~2 that you get).
>
> I am not an expert on these computer things but I thought dual core
> essentially meant 2 procs. In fact, a factor of 2 with lower speed procs
> (2.0 versus 2.4 Ghz) means that the performance of my quad is more than
> simply 4 procs.
>
> > do the cpus have dual channel memory access?
> > just a hint..
>
> 16 GB DDR1-400 ECC REG LP(64x8)(1GB x 16)
>
> That is 16 GB de RAM.
>
> Thanks for replying.
>
> Pedro
>
> > cheers,
> > luca
> >
Received on 2006-11-02 18:07:30

This archive was generated by hypermail 2.3.0 : 2022-09-01 14:03:42 CEST