Re: Is Intel exceptionally unsuccessful as an architecture designer?

On Sun, 29 Sep 2024 18:26:39 -0000 (UTC)
Brett <ggtgp@yahoo.com> wrote:

Michael S <already5chosen@yahoo.com> wrote:

On Sun, 29 Sep 2024 03:41:07 -0000 (UTC)
Lawrence D'Oliveiro <ldo@nz.invalid> wrote:
 

On Sun, 29 Sep 2024 02:08:28 -0000 (UTC), Brett wrote:
 

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

On Sun, 22 Sep 2024 16:58:10 -0400, Paul A. Clayton wrote:
 

My guess would be that CPU RAM will decrease in upgradability.
More tightly integrated memory facilitates higher bandwidth and
lower latency (and lower system power/energy).   

 
Yes, we know that is the path that Apple is following. That seems
to be the only way they can justify their move to ARM processors,
in terms of increasing performance. Doesn’t mean that others will
follow. I think Apple’s approach will turn out to be an
evolutionary dead-end.   

 
Intels newest server cpu moves the dram onto the socket getting
rid of DIMMs.   

 
And Intel is not exactly in the best of market health at the
moment, is it? 

 
It seems, Brett is confusing Intel's client CPUs (Lunar Lake) for
Intel's server CPUs (Sierra Forest and Granite Rapids).
Don't take everything he says at face value. As a source of
information Brett is no more reliable than yourself. 

 
Four times the dram bandwidth,
 

For the same # of IO pins LPDDR5x has only ~1.33x higher bandwidth
than DDR5 DIMMs at top standard speed. Somewhat more, if measured per
Watt rather than per pin, but even per Watt the factor is much less
than 2x.

DIMMs are DOOMED.

In the long run, I don't disagree. On client computers - the run would
not be even particularly long. On servers - it will take more than
decade.
But even on clients it's not going to be completed overnight or over
1-1.5 years.