On Tue, Jun 06, 2006 at 10:12:33AM +0200, Francesco Pietra wrote:
> After the installation of amd64 etch with generic kernel, I chose a kernel for
> multi dual opteron.
> uname -a reports:
> Linux debian 2.6.15-1 amd64-k8-smp # SMP
> In view of compilation, I I wonder about the following:
> Dual (or more) Opteron designs are not SMP (Symmetric Multi Processor), but
> NUMA (Non-Uniform Memory Architecture). With SMP the two CPUs share the same
> bus to memory, but with NUMA each CPU hassome local memory connected to a
> local memory bus. The other CPU can still get to that memory, but it's a bit
> slower. If you would only putin 1x 2GB, you will severely slow down the other
> CPU cause it has to go through the other CPU to do memory accesses. See
> The cpu design may have a bearing for the compilation of programs for
> multiprocessor systems, like mpqc (in the compilation notes, mpqc refers
> to "symmetric multiprocessor", see
They _are_ SMP. Every processor is identical and of the same speed.
The fact memory access speed is not symetric doesn't mean the CPUs
> Do smp specified on kernel and NUMA for the opteron characteristics fit
> together? I can imagine that SMP specified on the kernel chosen simply refers
> to the presence of more than one dual-opteron. Anyway, it is a bit confusing
> because it might be seen in conflict with the NUMA characteristics of
SMP simply refers to more than one cpu thread/core/whatever executes
code. Intel's hyper threading, dual core, multiple processors, all
> If it is a silly question, I beg pardon.
Not silly. It is getting somewhat confusing.
- From: Francesco Pietra <email@example.com>