Re: Is Intel exceptionally unsuccessful as an architecture designer?

On Sun, 22 Sep 2024 02:12:31 -0000 (UTC)
Lawrence D'Oliveiro <ldo@nz.invalid> wrote:

On Sun, 22 Sep 2024 02:57:21 +0300, Michael S wrote:
 

On Sat, 21 Sep 2024 23:34:40 -0000 (UTC) 

 

Lawrence D'Oliveiro <ldo@nz.invalid> wrote:
 

On Sat, 21 Sep 2024 13:29:31 -0700, Chris M. Thomasson wrote:
 

CPU's that are hyper close to the memory is good wrt locality.
However, the programming for it might turn some programmers off.
NUMA like for sure. 

 
I’m pretty sure those multi-million-node Linux supers that fill
the top of the Top500 list have a NUMA-style memory-addressing
model. 

 
You are wrong.
Last ccNuma was pushed out of top100 more than a decade ago.
All top machines today are MPP or clusters.
Not that a diffference between the two is well-defined. 

 
If the difference is not “well-defined”, then how can I be “wrong”?

The difference between MPP and cluster is not well-defined.
The difference between ccNUMA and MPP-or-cluster is crystal clear.