AS macs: Memory bandwith vs. single core Apps (pro audio). What Mac makes sense ?

[streetfunk]

Heya,

i just took this screenshot : ( its about the numbers chart / Memory Bandwith)

from this thread: ( which has no relation to this topic)

https://forums.macrumors.com/threads/what-is-the-likely-ram-upgrade-path-for-apple-silicon.2396304/

My Topic:
To notes, i am allready on a M2pro, coming from a M1. I ask in regards to my future planning.
I work alots with a professional audio app which runs on itself explicitly on only one core. ( its a pluginhost, used for realtime-play music making purposes. So, Latency plays a big role, i run things with low(er) audio buffer settings)
my Query:

In regards to the different Memory Bandwith of these different AS Macs.
Specifically vs. "single Core only" Audio Apps, where is the limit ? up to which Mac can i gain a win vs, its increased memory/Data Bandwith ?
From which model on are these numbers "for me" pointless ones ?
I´d suspect: up to M2pro do i gain ! but above ?


Please notes, fact is: i was able to run my M1Mini into its knees, bevore i hit the cPU limit ( again: one core usage based)
The only thought that makes sense to me, vs. what i was doing, is, that i was running it into its knees based on the Data Transfer rate.


unimportant, just optional JFYI:
This are very specific audio tasks. Not much people who would ever try such stuff out. But i do.
We could just register it under: "realtimeplay-sounddesign".
It is sound manipulation in realtime, based on manipulating the allready generated audio stream.
Now, "manipulatiing" it in series, on several stages, one behind the other => *at the same time*.
Thats how i was able to run my M1 into its knees, while the CPU load (on just one core) was still fine.
I´ve not donne the same patches on my M2pro yet. So, I´ve not run my M2pro into its knees yet.

 

[leman]

Well, are you talking about bandwidth or are you talking about latency? If your application is single-core only, the max RAM bandwidth accessible by an M2 CPU is 100GB/s (lower in reality). M2 Max is clocked higher than the base M2, so the CPU itself might be a bit faster. As to latency, it goes higher the "bigger" the chip (i.e. M2 will have lower memory latency than M2 Max). Note that we are still talking about nanoseconds here, nothing you will notice.

 

[streetfunk]

or are you talking about latency?

No, latency is clear to me.
I meantioned it, in case the "realtime-aspect" would make a difference vs. my point in question.

If your application is single-core only, the max RAM bandwidth accessible by an M2 CPU is 100GB/s (lower in reality).

That one is the point !
Thats very interesting what you are saying !

So, i can have my takeouts from here. My query alllready solved !

As to latency, it goes higher the "bigger" the chip (i.e. M2 will have lower memory latency than M2 Max). Note that we are still talking about nanoseconds here, nothing you will notice.

this is interesting too.
Yeah, i understand that the higher the clock, the better for me.
But vs. the "latency itself" is it in this -applesilicon case- negligible then.

I thank you very much @leman !
it gives a nice, fresh, insight.

 

[streetfunk]

ok, i think i get the picture now.
and this fresh post in the above mentioned* thread is explaing the same picture.
* edit: its from this thread: https://forums.macrumors.com/threads/m3-chip-generation-speculation-megathread.2393667/page-6
screenshot:

@Basic75

gotcha.

 

[SjoukeW]

The only good way to know if extra memory bandwidth can be used is look for benchmark of the software you want to use.
Also the available bandwidth of the chip might be high, but that doesn't mean that 1 cpu core can use it all.
Benchmarks can tell you.