SMP/dual (multi) core phase II

[feenberg at nber.org (Daniel Feenberg)]

On Sun, 6 May 2007, markw at mohawksoft.com wrote:

> SMP dual 2X (dual core) CPUs, i.e. 4 processing cores and two physical
> processors.
>
> I prefer dual SMP to dual core as the multi-core processors share
> internals of the processor and are less capable than two independent
> processors. That's at least as far as I've been able to gather from
> research without actually devising and executing tests to prove this
> conclusion in practice. Kind of like the difference between Siamese Twins
> vs Identical Twins.
>
> If someone has an educated argument against this position I would really
> like to be educated.
>
> Second, does anyone out there know if dual core processors can (as a
> matter of hardware and OS implementation) be used to run two different
> processes simultaneously rather than merely two different threads of a
> single process. In true SMP, this is no problem, but with multi-cores,
> things like the memory and I/O controllers are shared and does this
> present a problem with things like memory addressing? i.e. does the same
> VMM map have to be loaded into the memory controller, meaning that
> separate processes with different virtual memory mapping could not benefit
> from two cores.
>
> Anyone have any real experience or info?

We have tested quad core and dual core Intel core 2 duo processors doing 
statistical calculations (long float arithmetic, mostly) and found the 
quad core runs 4 independent processes at just about the same speed (each) 
as 2 dual core processors of the same clock rate run them. So no advantage 
seen to more sockets, only cores and clock rates count. None of the 
processes is multi-threaded.

Our little compute cluster ended up with 3 quad core motherboards and 1 
double socket dual core. All 4 machines have nearly the same capacity. 
This could be an artifact of floating point, we haven't tested other load 
types.

Daniel Feenberg



-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.