I've always found Cinebench translates a lot better to DAW performance.

On my new 16” MBP the M4 Max single core performance is much better with Logic 11 than Cubase Pro 14 which has a similar performance to the sc performance of my now 2 year old 13900K.

So I conclude the sc performance of the M4 Max chip is fantastic but more in the context of “for a laptop” rather than in any class leading computer sense.

Generally, it appears to me, sc performance has reached a near theoretical max now silicon is at 3nm (transistors are mere atoms apart)

They can’t go thinner, they can’t go faster clocks (it’ll melt) they need to find a way to link cores without bridging latency so that the need to rely on single core performance is a thing of the past.

I think not. You need to read more enlightening articles like Moore's law may have expired, but chips keep getting faster, or The future of high-speed computing may be larger CPUs with optics and also realise software optimisation might be where it's at, as have been suggested already, provided consumers ask for that extra power. Are they? I certainly hope so.

Also, single core performance is not the same thing as single thread performance. In DAWs generally one track is one thread, but as all tracks/threads need to run in the same time frame, with more threads than cores, each single core is taxed with more than one thread and all those run sequentially.

Effectively this means until you have the same or more cores than threads, more cores will yield better performance, provided CPU overhead and thread coordination doesn't taper this off. I can't assess where this might happen as there are too many parameters I don't know. But up to a point I think you can verify this trend today, depending a bit on the DAW. But it's more fun to just make some music, isn't it?

Yeah, they need a new paradigm, coz it’s not a single CPU etched onto a bit of melted sand

I’m sure they’ll figure it all out - the tech guys always do.

Meanwhile - it’s buffer juggling as normal in my studio spaces be it Mac or PC.

Ya the single core bottle neck is quite frustrating. They need to find a way to spread out single tracks across cores.

Question… if you have a track that has a ton of plugins, feeding a bus that has a bunch of plugins, feeding a master bus with more…. Are all those in-line / single core?

Or do the chips break out tracks, and then their buses, across cores?

Why? How is that even a bottleneck? There are obvious solutions that cost more money and that is more cores or bigger audio buffers, or better drivers, or lowering quality for playback only (HQ for exports).
I think the Mx series, or any system on a chip, might be in a better shape for this. However, I wouldn't hold my breath. I think there would need to be a completely new real time programming paradigm.

I don't see why any track waiting for data from other tracks needs to be in the same thread, and if it isn't, on the same core. The important bit is being able to process within one sample (I think)**, but this might still involve any latency needed by processing.

I also know different DAWs do these things differently and that the OS decides most of it, often without the DAW knowing very much about this, or even anything at all. A DAW may request (and get) resources from the OS, but the implementation details may not be that interesting for the DAW. This is separation of concerns.

** It's probably more likely to be within the duration limits of the current audio buffer.

Cubase overrides the OS core threading/allocation and does it’s own,whereas Reaper lets the OS do it.

This has pluses and minuses for both programs.


M

I don't get it when people complain that single core performance is not enough of cutting edge tech consumer electronics CPU's.

You can offload CPU to a UAD DSP accelerator or use a summing box with remote controlled hardware on the master bus.

Depends on what you want on the master bus. Most of what I’d like isn’t available as a DSP option.

Oh yeah. I trade a universally usable CPU for some driver-needing black box.

Those options aren’t always the solution for single core bottleneck. All depends on what you are doing

Doesn't have to be a DSP or analog stuff. Also can be a 2nd computer with digital IO. I have my streaming computer adat piped to my main rig. There is also software called audiogridder that allows networking for adding additional CPU via a plugin.

So if you think in solutions instead of problems there always remain options.

Proof? Even indicators this is what is actually happening. I think if investigated results will show Cubase might ask more intelligently yielding something more useful, but I don't believe macOS gives up control allowing overrides in the way you imply here. I might be wrong, but this would be a game changer in many ways.

That said, if there is anyone able to pull this off in a successful way it's those that've been at this for a while and have sought maximum performance since decades.

After all the confusion out there, this is exactly what I was hoping to read.

Did you change your opinion over the last months?

Can you run Sessions with serial chains of busses? For example Drum bus - > pre master -> Master - and process each of these generously? With oversampling stuff?

That would be wild!

Yes, you can, but it still has its limits of course. I haven’t done any serious true a/b’ing but what I can say with certainty is, sessions that were crapping our regularly, to the point of not being able to play back, on my M1, even at 512 or 1024 buffer, are running perfectly at 128 /256.

I’m trying to keep 128 buffer. It’s mostly working. But when I do get biiiig sessions with lots of bus stuff plus ozone, I will sometimes get some overages, especially if trying to record audio. 256 is perfect.

So, it’s noticeably a big improvement from M1. But it’s not like 3x better or anything. Prob closer to the gains on geekbench. Up to double. And allows lower buffer.

Thanks for sharing this experience of yours in Logic Pro. From your account it's somewhat hard to follow exactly what you mean with "single core experience" Did you actually test just one single core? That would involve using a single track, or possibly matching the track count to the p-core count, but I think this might not be fully suitable for the M4, though I haven't tried. I know of no other way to make sure only one core does meaningful work to measure.

Judging from what you shared here it sounds to me like you got a a more powerful CPU in the M4. You don't share how many p-cores your M1 have, so it's not possible to gauge what part more p-cores in the M4 might play for your experience and not just the individual increased power each p-core may utilise.

In your initial post you state "since bus processing is all done on a single core". But the reality is a buss is just a track like any other, and each and every track in most DAWs, if not all, is processed in one core (that work may move around with the M4 doesn't this AFAIK).

A statement like "single core experience" kind of implies the buss would have this single core to itself. I hope you don't mean this, as it most certainly doesn't, as any source track would add a load. But accepting this is true, that most cores will have to process multiple DAW tracks of any kind, pretty much invalidates speaking too much about single core performance.

It's not that it's not important, because it is, it's more that performance of any core doesn't operate in vacuum, rather that for the task of multiple track audio processing it's cooperatively working with all involved processing units in the machine. The increased power of the M4 means each core can do more. That's great!

The only unique difficulty I see in a buss would be more heavy processing on it, that would probably only be true when comparing with audio tracks and not software instruments whose CPU demands possibly would be hard to match, depending on the actual instruments and the playing. Those are the only sources bringing my M1 on its knees anyway (Ableton Live).

Just to be clear, in this kind of performance discussion, I perceive a disconnect in many posts between the very understandable and acute interest in core performance enhancements and the examples brought forward claimed to be proving said enhancement. I'm not saying that anecdotes from real life projects isn't interesting, because of course they are, but it can be hard to draw valid conclusions from those descriptions.

These examples typically involve many tracks. I believe that how the totality of these is actually processed does arguably affect also the heavily loaded buss track, or what is of interest. In the context of the discussions this often appears assumed to be of less importance. I rather believe it's very important to take this "other tracks" load into account as well, and avoid those if possible, if one wants correct results.

But as I just wrote, if one wants to test what the enhanced single core performance actually means in processing some stacked task, one could more safely test just one track, because this track will by necessity, or out of DAW design reasons, be processed on one core at a time.

Single Core Test suggestion

As for tests, why not play a stereo audio file through the main buss processing times some number and attempt to identify the core (group) doing the actual work? Add up units until drop-outs, dial it back and keep it running for some time.

**EDIT** I just invalidated the imagined test below for stacked software instruments, when I attempted it in Ableton Live stacking 20 instances of the (donation) free DiscoDSP OB-Xd synth in one single track playing a 5-note chord sequence using the preset "Vintage 80s Strings" for an M1 test. The work was spread out on p-cores identically as to what happens with the same instances on their own track.

Invalidated on Ableton Live at least

Or stack software instruments on just a single track, assuming this can be done. DAW bench uses Kontakt, but I think some common free alternative should be found, so anyone could conduct tests anyone else can repeat.

„Since bus processing is done by a single core…“

Yes, it is true for those speaking about this.
And it is also true, that different cores can do your busses.

From all my tests and observations with ALL major DAWs (cannot speak for Logic) this is what happens exactly:

If you put audio fx in series, and this includes routing a bus a into bus b, into bus c, etc. - this entire chain can only be rendered by one core/cpu thread. Includes the master channel as well.

It’s even written in the Ableton Live manual.

If you got 3 separate busses all going into the master individually, then this does not apply.
Then the chain for one core would be bus a > master. Another separate core would proceed bus b. Another one could proceed bus c.

You can easily test this by stacking a heavy ovesdampled plugin on 1 single track until your cpu crackles.

Then take half of those plugins away and paste them on the master instead.

You will reach crackling at the exact same number of plugins in the total signal flow chain (from track to master).

Now create a bus in between and split again the fx sitting on the master into two more halves, pasting the new half on that bus in between track and master. Same performance results.

The limit for one core (and therefore for the entire DAW) is how much serial audio flow can be performed by a single cpu core.

Now to prove the opposite, let’s say the total number of instances of that heavy fx plugin that run serial would be 6.
if you now clean the master and busses. Start new over with single tracks and put 4 of those instances on a track (which would not make your daw crackle, right) you should then be able to duplicate that track 5 times easily, making it a total of 20 units of the hungry fx running smoothly in your project, which could not be handled if more than 6 were sitting in a serial signal flow chain.

If you follow me till here, this is how this whole single Core multi core thing works.

Anybody please correct me if I am wrong.

Thanks for the suggestion to do tests. I made some new ones, verifying much of what you suggested, though on M1 I saw 2-p-cores engaged with the same load (Abbey Road Plate) on (A) a single track, and (B) that track into one buss and (C) that buss into a 2nd buss. I'll redo these later on M4.

You would seem to be in agreement with my main point that a buss will not have a single core to itself. In typical projects all cores process multiple tracks, or possibly series of them as one, as you suggest.

This points to the fact that other cores matter too, for additional tracks getting processed on these instead and through that hopefully lessen the load of heavily burdened tracks. But I know of no way to predict how cores process audio in real projects where the number of tracks exceeds the number of p-cores. I hope that some kind of DAW requests for resources get some kind of reasonable response from the OS, so there is some distribution. I haven't seen a pattern in this. Doesn't necessarily mean there isn't one, of course.

I might add that the main reason I'm shopping for an M4, is precisely the increased power of each p-core in the M4 series as well as a potentially greater number of p-cores.

Yes I fully agree with what you said.

What happens, if the DAW detects that there is a signal flow somewhere in the project, as heavy as a single core could handle. Will it make sure, that core dealing with the heavy serial chain will got nothing else to do other than just that chain, or will it also do some additional tracks?

The latter would be bad.

I am also aiming for the M4 for the higher power of each p-core. Hope that everything works out the way we plan.