The M1 Ultra answered my main curiosity about Apple silicon

[mattkime]

Does the M1 family of chips scale?

Answer: Oh hell yes.

It will be interesting to see more information about these machines. Due to the ubiquity of intel machines, I'm sure they'll live on in some form but ARM designs are the future.

It will be interesting to see if there's more adoption of ARM on the windows side. Its happening in the cloud computing space - if I remember correctly, you typically get 20% more computing power for your dollar with ARM processors in the cloud. There are varying levels of support for this across applications but its going to become an obvious choice for a lot of cases.

[Lux Interior]

The base Pro will have 8 Max (4 Ultra) with 512 GB Unified Memory and start a 1 arm + 1 leg.

[mikebw]

Lux Interior wrote:
The base Pro will have 8 Max (4 Ultra) with 512 GB Unified Memory and start a 1 arm + 1 leg.

It sounds nice, but do we have any idea if they can double up the M1 Ultra? The M1 Max works b/c of the special interconnect they designed in before the fact. Any other means would be a lot slower and then (I think) pushes back to the older dual processor paradigm, meaning re-writing of code for individual apps in order to take advantage of it.

I was thinking we might see an M2 for the next Pro. Then the M2 tech trickles down into the other products over time instead of this awkward waiting game of building up from the bottom that we have been doing with the M1.

[mattkime]

mikebw wrote:
It sounds nice, but do we have any idea if they can double up the M1 Ultra? The M1 Max works b/c of the special interconnect they designed in before the fact. Any other means would be a lot slower and then (I think) pushes back to the older dual processor paradigm, meaning re-writing of code for individual apps in order to take advantage of it.

I'm cure they _can_ - its just a question of whether its worthwhile.

>meaning re-writing of code for individual apps in order to take advantage of it.

As far as I know, this is still necessary

[kj]

Since Qualcomm bought a company started by ex-apple silicon engineers, and at least one apple silicon engineer was hired by intel (he may work on things other than ARM, idk), it seems likely. I think Qualcomm could make chips that rival Apple's, if they decided to, and I think they have, so I'm sure we'll see more ARM on the windows side. And obviously the chips for Android phones will continue to get better.

[Lux Interior]

mattkime wrote:
[quote=mikebw]
It sounds nice, but do we have any idea if they can double up the M1 Ultra? The M1 Max works b/c of the special interconnect they designed in before the fact. Any other means would be a lot slower and then (I think) pushes back to the older dual processor paradigm, meaning re-writing of code for individual apps in order to take advantage of it.

I'm cure they _can_ - its just a question of whether its worthwhile.

>meaning re-writing of code for individual apps in order to take advantage of it.

As far as I know, this is still necessary
My post was just joculation (jocular speculation), but the M1 chips already have many cores. Why would you need to re-write for more cores? It it's native on the M1, it should work on any M1.

That is, the code should already be written for multiple cores. Adding more cores might need optimization, but not re-coding.

This isn't the same as going from one Intel CPU, to dual CPUs.

[mikebw]

Lux Interior wrote:
[quote=mattkime]
[quote=mikebw]
It sounds nice, but do we have any idea if they can double up the M1 Ultra? The M1 Max works b/c of the special interconnect they designed in before the fact. Any other means would be a lot slower and then (I think) pushes back to the older dual processor paradigm, meaning re-writing of code for individual apps in order to take advantage of it.

I'm cure they _can_ - its just a question of whether its worthwhile.

>meaning re-writing of code for individual apps in order to take advantage of it.

As far as I know, this is still necessary
My post was just joculation (jocular speculation), but the M1 chips already have many cores. Why would you need to re-write for more cores? It it's native on the M1, it should work on any M1.

That is, the code should already be written for multiple cores. Adding more cores might need optimization, but not re-coding.

This isn't the same as going from one Intel CPU, to dual CPUs.
Agree, all harmless speculation. Cores are cores, yes.

My reply was going off of what was said during the presentation yesterday about why the M1 Max was easy to make into M1 Ultra, and how that approach is better than running two together via a more traditional interconnect which, they said would be slower, hotter and would be more work for developers.
https://www.apple.com/apple-events/march-2022/ 26:00.

What we don't know is if they have another trick up the M1 sleeve and can in fact stick two Ultras together, or not. I hope the answer is yes.

[gabester]

mikebw wrote:
Agree, all harmless speculation. Cores are cores, yes.

These days it's no longer about cores but chiplets. The M1 Ultra being two M1 Max stuck together is an example of chiplet-based (or more likely a "chiplet-like") design.

More info about an emergent industry standard:
https://arstechnica.com/gadgets/2022/03/...-chiplets/

Essentially, chiplet design will let many differentiated cores be interconnected into a unit.

[mattkime]

gabester wrote:
More info about an emergent industry standard:
https://arstechnica.com/gadgets/2022/03/...-chiplets/

Essentially, chiplet design will let many differentiated cores be interconnected into a unit.

I read that, or at least tried to. It was unclear to me what differentiated a chiplet from...not a chiplet.

[rich in distress]

It’s like FarmVille, but with silicon.