Re: Success report: installing pure64 on a dell poweredge 1850
On Thursday 09 September 2004 20:42, Karl Hegbloom wrote:
> On Thu, 2004-09-09 at 11:25 +0200, Frederik Schueler wrote:
> > This indeed changes quiet a few things.
> Yes, but the 1.8 GHz Opteron is apparently still faster than the 2.8 GHz
> Nocona. Rather than a 'my cpu is faster than yours' "pissing contest",
> I'd like to see a WHY one cpu is faster than the other discussion.
> It must be mainly due to the thing about how the Nocona isn't really
> fully 64 bits wide all the way through the chip... IIRC...? Is that
> correct? What other factors are there in this sort of comparison? (I'm
> a sophmore CS student.)
- Pipeline length (amd is shorter)
- Number of parallel execution units (amd generally has more, I think)
- Memory latency (amd has lower latency)
- Memory bandwidth
- Cache size and structure and latency
- Which operations are provided natively, ie. how different x86 instructions
are broken down into micro-ops.
- How good gcc is at optimizing for that CPU.
- System topology. Intel uses a shared Front-Side Bus with memory and IO hubs.
AMD uses point-point HyperTransport links and on-cpu memory controller.
I'm sure there are others. Which factors are important depend on the code