M3 Pro chip a bad choice for Logic Pro?

[Mark Horoocks]

I don't mean to harp on about this; and as an intel Mac user, anything is probably faster than my current set up, lol! Here is another reputable video reviewer who seems amazed at Logic Pro's performance on the M3. Now, I do not profess to be a 'techie', but both videos posted on here are not clickbait and are well made, well considered reviews of M3 performance using Logic Pro, IMHO. I just wish there was some info from Apple as to why this happens on the new M3. If they're deliberately disabling cores not to function in Logic Pro, then users may well simply move to another DAW. If there's a good reason for it, then why can't they make it clear? The Logic part is at 9:20. Someone asked for a time stamp on the previous video, but there were segment stamps listed on the timeline, as there are here too. Thanks for any feedback as it's good to hear what more knowlegeable members think/feel about this.

 

[synth_hero]

7 hours ago, Mark Horoocks said:

Someone asked for a time stamp on the previous video, but there were segment stamps listed on the timeline, as there are here too.

Delving into YT videos without any specific references can be a bit of a rabbit hole and time-wasting: hence the request for a timestamp. Referring to the video from the OP, the reviewer suggests that that he used a "Model DI guitar on a track". While the tests seemed rigorous, there was no specifics on the type of model DI used. Then he ran it through the Archetype Nolly plugin on all the tracks. This is not a realistic scenario. 

Check out this video from a Mac Pro Tools user who presents a more realistic example.

At 1:40 he describes his test session used for his comparisons. he uses compression, EQ, lo-fi, reverb and delay plugins for each track and runs 120 tracks in each test. 

Point taken: the M2Max Macs are better than the M3 Max Macs for certain DAWs: Logic Pro X, Ableton Live 11 and Pro Tools Ultimate. The reason for this being that the DAWs do not use the efficiency cores which are present in a proportionally greater number on the M3Pro chipset . . . 

Edited December 8, 2023 by synth_hero
correcting inaccuracies

[jameszhan]

 

Hi all! This is James from the video testing 7 DAWs on Apple Silicon Macs. Happy to see the friendly discussion here 🙂 

A few points I wanted to emphasize:

• YES I did set Logic to use all the cores in all 3 Macs.
• I have seen many people saying that DAWs will receive updates that optimizes them for the M3 chips. Not sure if we can count on that as the 4 DAWs that can't fully utilize e-cores exhibited the same issue even on the first gen chip. I'm not saying it can't happen, but we don't know if or when that will happen so this is something I believe people should keep in mind if they're buying a Mac for music production right now.
• My tests are only testing a multitrack audio processing scenario, and synths might use the CPU differently.

On 12/3/2023 at 1:15 PM, des99 said:

The problem with YouTube videos is someone makes a video about something they don't understand (instead of actually investigating it), then other people treat those videos as canonical, and now think the software has a problem, and start blaming Apple for their terrible inefficiency and poor coding, rather than checking out the behaviour, reading the manuals, and determining for themselves whether there is an *actual* issue.

I would appreciate if you could at least give their video a watch before claiming someone was spreading misinformation in their video! I spent a lot of time doing rigorous testing to verify the results before making the video. The last thing I want is spreading misinformation 🙂

So far, I've received many comments from people verifying that they did notice a performance boost when switching from a DAW that doesn't fully use e-cores to one that does.

Furthermore, a dev from Ableton did say that the team is working on getting Live to fully utilize the e-cores in Apple Silicon chips (which implies that Live currently does not):

Quote

M1,2,3 apple silicon chips today have an architecture composed of performance and efficiency cores. This was known in the past as a big.LITTLE architecture and did not exist on personal computers, only mobile phones. Nowadays it's both on Mac and PC with Intel's latest offerings as well.

This creates a challenge on how a DAW best distributes work to various cores. It's definitely being worked on by our engine team.

(Source)

In general, I think it's absolutely sensible to question information you encounter on the internet (YT included)! In fact, I encourage people to test their DAWs themselves instead of just 100% taking my word for it. I purposely included the audio settings I used for each DAW, listed the DAW versions, and chose a plugin (Archetype Nolly) that has a free 14-day trial so people can try to reproduce the same tests at home.

I always do the best I can to do comparisons that are accurate and make sense in real life, but everyone's usage is different and I can only focus on selected scenarios (in the case of this video, multitrack audio processing).

Just to clarify, I'm not a programmer so I don't know what goes on in the backend of each DAW and the OS, but what I did do was showcasing the real-life implications of different DAWs' varying ability to use the cores in Apple Silicon chips.

Hope this clears some things up and I'm happy to chat more 🙂

[des99]

29 minutes ago, jameszhan said:

would appreciate if you could at least give their video a watch before claiming someone was spreading misinformation in their video! I spent a lot of time doing rigorous testing to verify the results before making the video. The last thing I want is spreading misinformation

Hi James - thanks for the post.

To be clear - I wasn't saying/claiming your video was spreading misinformation - as I made clear more than once, I haven't watched your video. I *have* seen plenty of videos that get some fundamentals like this wrong, hence the comment about some of my general experiences with YT, including specifically around Logic, in our discussion around the issues.

As you'll see from this thread, I haven't made any specific comment referencing your video at all, so saying I was claiming this video was spreading misinformation is reading more into my comments than I put in there.

The only thing I said in reference to your video was this:

On 12/3/2023 at 6:07 PM, des99 said:

I haven't watched a bunch of videos, but depending on what was said, perhaps the video chap doesn't know how this works, left Logic on the default, and saw a decreased performance over earlier machines that had more p-cores. I can't really comment on that, as I haven't watched the video.

I think I've made it clear I wasn't claiming you were spreading misinformation, or that you were wrong. And someone later pointed out that you did set Logic to use all cores in the video in response to that comment, and did a better job at representing the content of your video for the purposes of discussion.

More than that I can't say, as I neither have an M3 or more comments to make, other than already knowing the M3 Pro has changed it's relationship in the MBP range compared to the M1/M2 lineup, for product marketing reasons - it's now no longer an "almost Max" for Logic, but is quite some way off a Max now.

[jameszhan]

11 hours ago, synth_hero said:

Referring to the video from the OP, the reviewer suggests that that he used a "Model DI guitar on a track". While the tests seemed rigorous, there was no specifics on the type of model DI used.  

Hi, James from the video testing 7 DAWs here! What I said was "mono guitar DI," meaning:

• It's an audio recording, not MIDI
• It's mono, no stereo

11 hours ago, synth_hero said:

Then he ran it through the Archetype Nolly plugin on all the tracks. This is not a realistic scenario.  

It actually does reflect real-life usage, but it just does not seem so on the surface, so I understand the criticism. I actually explained in the video, but I will put the reasons below in case others are reading.

In the two years I’ve been testing Apple Silicon macs for audio processing, I’ve learned that testing DAW performance on these chips really comes down to multicore utilization. Obviously, no one will duplicate the same track so many times in a real mixing session, but you probably have many tracks, and those tracks will probably have all kinds of plugins on them. In a typical mixing session like that, you are basically relying on the multicore performance your DAW can get out of the chip, and that’s what I’m simulating here with my test setup. It doesn’t really matter that I’m duplicating an amp sim many times; what matters here is that we can see how each DAW uses the cores in the CPU, and how much of a difference the M1 Pro, M2 Pro and M3 Pro chips have compared to each other.

The first time I reported on some DAWs not fully using e-cores in Apple Silicon chips was actually last year with an M2 Pro Mac, and in that video, I did a test that was more realistic, with different plugins on each track—kind of like Colt's test in the video you linked. The results were the same as when I just used the same plugin on each track, in that DAWs that could fully utilize e-cores performed better than those that couldn't.

So, for the M3 Pro video, in order to 1) test more DAWs, 2) keep the video relatively short and digestible, 3) allow viewers to recreate the tests themselves, I opted to stick with the simpler test setup that still reflect the performance differences in real-life.

Interestingly, my M3 Pro video got picked up by way more people than the M2 Pro one, even though I already revealed the same information in the M2 Pro video last year.

I think my M2 Pro video, albeit covering more types of tests, was simply too complex and the information was not very digestible/accessible. In this year's M3 Pro video, I zeroed-in on 1 issue and kept the testing simple, resulting in a more accessible video. 

Hope this helps!

 

Edited December 8, 2023 by jameszhan
clarify who I am

[synth_hero]

Hello James,

So you do understand why I am skeptical about your tests. Thanks for your explanation. It does clear things up. 

All the best.

S.

[Dewdman42]

I wish you had also included Digital Performer in the comparison test.  Some others have independently tested and stated that it's using all the e-cores also.

Now this is where it gets interesting though.  There have been some discussions on the vi-control forum where the lead developer at VSL has chimed in about e-core and p-core usage and the multi-threading complexities that have come up.  Apparently at recent ADC conference in London, they were able to sync up with devs from Steinberg, Apple, etc..to discuss the issue.  It would appear that some multi-threading strategies from years past may have some problems now with the mixed bag core situation that is now part of ARM architecture on the mac, and some PC's are starting to come out in that architecture also.  And this has led, for example, with people using Cubase to have problems with audio dropouts with Synchron player, but not have the problem when using LogicPro, or perhaps using cubase+vepro.  It appears that internal threading methods have to be reengineered in order to properly choose when to use e-cores vs p-cores.  

What I'm trying to say is that it's entirely possible that Apple is excluding the E-cores on purpose and for a good reason.  There is a lot more to this story that will be developing over coming months as VSL, Steinberg, Apple, MOTU and others figure out how to work out core and thread usage with systems that have two or more types of cores being used.  

Apparently, from what I read, AU has some built in functionality also for properly managing some OS level thread grouping features and VST has nothing like that in it currently..  so that is also a good mark for Apple, even if on the face of it from this test kind of looks like Steinberg is using more cores and getting more tracks.  This is a developing situation and I think a lot is going to change over the next couple years related to how threading and core usage is handled.

See from this post forward in this thread: https://vi-control.net/community/threads/vsl-synchron-player-problems-on-m1-ultra.144765/page-3#post-5438477. this post is from the lead dev at VSL.

[des99]

25 minutes ago, Dewdman42 said:

It would appear that some multi-threading strategies from years past may have some problems now with the mixed bag core situation that is now part of ARM architecture on the mac

Yes, this already bit some plugin developers - I helped report this to Korg and they had to rework their software engines in their recent plugins to remove the old multicore support that worked really badly on Apple silicon. Roland have similar problems in their softsynth engines. Multicore on u-he plugins works great on Intel, but terribly on Apple silicon. There are definitely optimisation differences that devs are learning with Apple silicon, after decades of Intel optimisations.

And like you say, there may be technical issues about the core useage too, indeed.

[Dewdman42]

I think basically the various DAW developers mentioned are pow wowing together at ADC to brain storm about best approach.  When competitors are literally sitting down to talk to each other about how to move forward with the tech, you know there are are some issues that need to be solved regarding this.  It may or may not need to include changes in the OS, in the plugin API's, in the way DAW's are written and for the sake of multi-threaded plugins...they do would need to figure all this out.  It's quite possible the OS is trying to make things automatic and that poses a problem for DAW's which need to prioritize p-cores for the realtime work but can use e-cores for buffered tasks.  For example, DAW's could easily use e-cores for non-live tracks...with a larger buffer, but then fore p-cores for the live track.  E-cores should be prioritized for battery performance, you don't want the p-core to be used unless it's really needed.  So the OS and DAW is probably trying to figure that out and it seems to me like LogicPro is basically saying, put all the plugins on P-cores.  Could some plugins go to e-cores?  I say yes, but then you'd have to do fancier stuff, maybe kind of like the non-real time stuff that Dp is able to do to pre-render some tracks...so some plugins can be pre-rendered and why not use e-cores for that, or have larger buffers for non-live tracks, etc..  all possible, but then it just opens up a new can of worms with new optimization strategies to make use of all the cores as much as possible and I think they will be working through this for quite some time frankly.

I think it's entirely possible that though this test shows Cubase and Reaper (and apparently DP) to be using all the cores....there is a good chance that with certain kinds of plugin use they would get drop out city while the others that are constrained to P-cores will be more predictable, at least for now.  

[mparker]

Thanks so much folks for helping clarify the situation. It’s not so frustrating when the situation is clear and I’m not sure why the Logic ‘expert’ I spoke to at Apple couldn’t have been as helpful. He seemed to imply that I was plotting the downfall of the company by simply asking why Logic behaves in this way, 😂! Even if the situation hasn’t been resolved yet, it gives me hope that if I fork out 3 grand on a new M3 chip computer, I won’t be throwing money down the drain. Thanks James for chiming in! I think your YouTube videos are top-notch! 😎