Re: Is Intel exceptionally unsuccessful as an architecture designer?

On Mon, 23 Sep 2024 22:05:27 +0000
mitchalsup@aol.com (MitchAlsup1) wrote:

On Mon, 23 Sep 2024 21:34:03 +0000, Michael S wrote:
 

On Mon, 23 Sep 2024 21:10:00 +0000
mitchalsup@aol.com (MitchAlsup1) wrote:
 

On Mon, 23 Sep 2024 15:06:50 +0000, Scott Lurndal wrote:
 

"Paul A. Clayton" <paaronclayton@gmail.com> writes: 

On 9/17/24 8:44 PM, Lawrence D'Oliveiro wrote: 

On Tue, 17 Sep 2024 23:45:50 +0000, MitchAlsup1 wrote:
 

"the CPUs are simply I/O managers to the Inference Engines and
GPUs." 

>
That particular Wheel of Reincarnation will never turn that way.
>
Why? It comes down to RAM. Those addon processors will never
have access to the sheer quantity of RAM that is available to
the CPU. And motherboard-based CPU RAM is upgradeable, as well,
whereas addon cards tend not to offer this option. 

>
My guess would be that CPU RAM will decrease in upgradability. 

>
LDO's statement "will never have access to the sheer quantity of
RAM that is available to the CPU" is flat out wrong.
>
Marvell already offers a CXL add-on processor card that supports
up to 4TB of DRAM with 16 high-end ARM64 V series cores. 

>
At somewhere near 3× the latency to DRAM.
 

>
Where did you find this figure? 

 
I calculated it based on how messages get passed up and down PCIe
linkages and that that plug in memory has to be enough wire distance
to need a PCIe switch between CPU die and Plug. Then add on typical
memory controller and DRAM controller, and that is what you have.
 

I have read both product brief and press release and didn't see any
latency numbers mentioned, not even an order of magnitude.
>
I suppose, in order to get real datasheet one would have to sign
NDA.
>
Somehow I don't see how anything running over PCIe-like link can be
as fast as you suggest. 

 
I was not suggesting it is fast, I was suggesting it is slow.
 

And I think that 3x, i.e. ~150 ns latency of cache miss measured at CPU
core, is more than fast enough to be competitive. But I don't think that
it is technically possible.

If the size works for your application--great !
If the latency does not work for you--less great.