Poor performance explained!

[audiolex1]

Originally Posted by midnightrambler

Hey, getting a mix is hard, but that's my job. Building a wall is hard. I'm not having a go but I don't think "because it's hard" is a great excuse. Try harder.

Well look. This is new tech and it comes with the territory. It is a lot easier to get a mix together.
I mean would you rather have a buggy version than a stable one that doesn't use the E cores?

I mean I remember mixing on HDX and struggling on a 7 card system.

Here is a video showing DAWs using E cores, so there are 2 choices. Get more H cores or switch.

https://www.youtube.com/watch?v=FSqX4bt9to4

[midnightrambler]

I have no skin in this game. Well only that I'm still using a 5,1 so I'm going to have to upgrade at some point. I don't really care enough about this to get fizzled about about it, but I don't expect someone as normally as reasonable as Darryl to come on patronising everyone about how "hard" something is. I think everyone knows this stuff is *hard*, but we also - fairly reasonably imo - expect programmers to know how to handle their shît.

[K Roche]

Originally Posted by strangedaysuk

So I run Apple silicon and noticed that only 4 cores are being used.

I understand this is an excuse used for reliability. However in the video below Reaper and Cubase seem to have no issue with it. And to top it of the offer almost double the performance. Even Logics performance is terrible.

Makes me consider Reaper for now till the developers properly get to grips with Apple Silicon. I now know why the M1 felt so much slower that Intel despite good benchmarks, the issue is in the programming.

https://youtu.be/FSqX4bt9to4?si=J4_D4hH-3PhqZSUs

Humm you do realize the specific single words we use often actually reveal more about what we really feel and our perspective , than the sum total of the meaning in a sentence or paragraph.

I tried Reaper and was very impressed how fast it downloaded and how fast launched compared to PT .

Then spent the next 60 days of free trial trying to decode the convoluted nomenclature and go through the gyrations to fix the very cumbersome work flow. No doubt if I had started with Reaper I might think it was spiffy but so many years with PT makes it's work flow problematic for me. BUT hey YMMV go for it ..

[smurfyou]

Since I last posted in this thread I've done several mixes on my M2 Studio without a hiccup. And an offline bounce of an hour show at 11X speed

For a real-world data point.

And I was so skeptical and paranoid I kept my Intel machine running to the side for the first two weeks I used the Studio. "Just in case"... But didn't touch it except to copy some files.

[alexthemusicbloke]

I think I would rather have a machine which performs consistently than being able to reach the maximum number of plugins under the right circumstances.

The efficiency cores are there for a reason, you use them to run secondary tasks to keep the important processes running smoothly. Avid have gone with the right approach IMO.


Reaper... it's a superb piece of software and I'm a big fan, but it is absolutely not a reliable mixing tool in the same way PT is. Great it can run so many plugins, but I can't get to the end of a complex mix without encountering a mountain of bugs, instabilities and unpredictable behaviour once the session gets big.

[strangedaysuk]

Originally Posted by alexthemusicbloke

I think I would rather have a machine which performs consistently than being able to reach the maximum number of plugins under the right circumstances.

The efficiency cores are there for a reason, you use them to run secondary tasks to keep the important processes running smoothly. Avid have gone with the right approach IMO.


Reaper... it's a superb piece of software and I'm a big fan, but it is absolutely not a reliable mixing tool in the same way PT is. Great it can run so many plugins, but I can't get to the end of a complex mix without encountering a mountain of bugs, instabilities and unpredictable behaviour once the session gets big.

Ahh sorry had to jump in here.


Is this really true? I mean yes having some spare CPU is great but my old i9 MacBook 2019 is fast than these processors. I was running 100+ tracks with no issues. Admittedly I get better battery life and it’s significantly quieter, but it’s actually running far less- the i9 was 8 core hyperthreading. It would struggle when pushed but I knew its limits. I wanted to keep up to date so bought into the hype. I am skeptical that it’s much better is some ways. I think more on this may come out as it starting to. If you buy a top of the line max chip you are probably fine. I just wonder if you need to limit it to only performance cores. I know a Cubase user who has a big grin on his face right now who is a Mac user and he is also a very big engineer!


I prefer Pro Tools and Logic and even Live for some stuff. I just wish I could adjust my own settings, I can always turn it back if I hit a wall.

[Darryl Ramm]

Originally Posted by strangedaysuk

Ahh sorry had to jump in here.


Is this really true? I mean yes having some spare CPU is great but my old i9 MacBook 2019 is fast than these processors. I was running 100+ tracks with no issues. Admittedly I get better battery life and it’s significantly quieter, but it’s actually running far less- the i9 was 8 core hyperthreading. It would struggle when pushed but I knew its limits. I wanted to keep up to date so bought into the hype. I am skeptical that it’s much better is some ways. I think more on this may come out as it starting to. If you buy a top of the line max chip you are probably fine. I just wonder if you need to limit it to only performance cores. I know a Cubase user who has a big grin on his face right now who is a Mac user and he is also a very big engineer!


I prefer Pro Tools and Logic and even Live for some stuff. I just wish I could adjust my own settings, I can always turn it back if I hit a wall.

So now you do have a performance problem? What are you measuring to claim this? What are the specs of the Intel and Apple silicon system? What CPU errors are you getting when? What troubleshooting have you one?

And CPU meters... I hope you are not looking at CPU meters on these computers and using that to imply throughput or excess capacity. They mean so very little.

i9s 8 cores are usually not slouches, well as long as they are cooled well. But hyperthreading is a smoke and mirrors illusion. What appears to be extra cores in CPU meters can't deliver the amount of work that the CPU count/CPU meters make it look like they can. Doubling the apparent core count with hyperthreading might give 0, 10, 20, 30 percent increase in capacity, not the 100% that might be expected. And for some workloads turning on hyperthtreading will increase CPU errors and decrease throughput, highly workload dependent and less likely now that plugins are better ported to AAX. But that's analogous to what likely happened with E-cores and why Avid disabled them. I expect they ran the current code, (with no special scheduling awareness of E cores) on both E and P cores and the performance under load was not good because of CPU errors, and they disabled the E cores and performance got better and they left it that way, hopefully until somebody can get time to do the engineering needed to get this to work better.

And the flakiness/sensitivity here to hyperthreading and E-core related problem are likely much worse when tracking and using small (where for some workloads "small" might be or anything smaller than 1024 or 2048 samples) h/w buffers than with playback. The benchmark in that video was all playback, never exercised small h/w buffer performance, won't even go near the user configurable h/w buffer in Pro Tools since it's all playback/high latency domain. And it's using one specific plugin and only one plugin in each signal path. This may not be a great workload model for many real life sessions. Pro Tools and some other DAWS will struggle more with scheduling when there are longer chains of plugins and the scheduler tries to keep those plugins running on the same core for CPU cache locality and other reasons. This and other behavior was better optimized starting in Pro Tools 2018 (.? can't remember the month release) but the effects of longer plugin chains can still be seen. And at least with Pro Tools that benchmark give an overly nice impression of increasing CPU load behavior up until things don't work. Real world CPU error fighting can be a lot messier.

I would hope any developer doing this would leave at least an undocumented way to enable E cores, and Avid really don't want folks trying it. But it's something to go peeking around at if you are curious.

[strangedaysuk]

Originally Posted by Darryl Ramm

So now you do have a performance problem? What are you measuring to claim this? What are the specs of the Intel and Apple silicon system? What CPU errors are you getting when? What troubleshooting have you one?

And CPU meters... I hope you are not looking at CPU meters on these computers and using that to imply throughput or excess capacity. They mean so very little.

i9s 8 cores are usually not slouches, well as long as they are cooled well. But hyperthreading is a smoke and mirrors illusion. What appears to be extra cores in CPU meters can't deliver the amount of work that the CPU count/CPU meters make it look like they can. Doubling the apparent core count with hyperthreading might give 0, 10, 20, 30 percent increase in capacity, not the 100% that might be expected. And for some workloads turning on hyperthtreading will increase CPU errors and decrease throughput, highly workload dependent and less likely now that plugins are better ported to AAX. But that's analogous to what likely happened with E-cores and why Avid disabled them. I expect they ran the current code, (with no special scheduling awareness of E cores) on both E and P cores and the performance under load was not good because of CPU errors, and they disabled the E cores and performance got better and they left it that way, hopefully until somebody can get time to do the engineering needed to get this to work better.

And the flakiness/sensitivity here to hyperthreading and E-core related problem are likely much worse when tracking and using small (where for some workloads "small" might be or anything smaller than 1024 or 2048 samples) h/w buffers than with playback. The benchmark in that video was all playback, never exercised small h/w buffer performance, won't even go near the user configurable h/w buffer in Pro Tools since it's all playback/high latency domain. And it's using one specific plugin and only one plugin in each signal path. This may not be a great workload model for many real life sessions. Pro Tools and some other DAWS will struggle more with scheduling when there are longer chains of plugins and the scheduler tries to keep those plugins running on the same core for CPU cache locality and other reasons. This and other behavior was better optimized starting in Pro Tools 2018 (.? can't remember the month release) but the effects of longer plugin chains can still be seen. And at least with Pro Tools that benchmark give an overly nice impression of increasing CPU load behavior up until things don't work. Real world CPU error fighting can be a lot messier.

I would hope any developer doing this would leave at least an undocumented way to enable E cores, and Avid really don't want folks trying it. But it's something to go peeking around at if you are curious.

Last section I agree with. It’s easy to measure, Hyperthreading can enable gains, I know this as an IT professional the difference to the servers I work on makes a significant difference which we measured recently when finding some of the older servers had it disabled for software that we no longer run, so we can turn it on.


The M chips are just not as powerful as they are made out to be. Better battery life, yes (although running half the power in some cases it’s no surprise), The M2 pro whilst made to appear 12 core is in fact 8 core. It’s just a no brainer that when you look at this information, and whilst there are some real benefits to Apple silicon, for pure performance, it’s just not quite what was sold to us. I’m planning a desktop computer and can save loads by just using an Intel/AMD and going that route. I haven’t committed just yet as I’ve been moving away from using Logic and going back to Pro Tools / Ableton Live, 2 Daws I’ve used a lot in the past, and when you take portability out the equation for the amount you save, a OC starts to make a lot of sense. However I will be testing this first in a similar spec to my MacBook to see how this feels - this is because I have access to said device. Might be interesting or a flop of an idea. I don’t intend on just going with the views, that would be silly, I’ll test it and see. If it’s similar then my next big upgrade would be to get something PC based but more power (yes double the ram I’d use in a MacBook as it’s integrated in that but it’s still cheaper and upgradable).

[huzzam]

Originally Posted by strangedaysuk

The M chips are just not as powerful as they are made out to be.

Honestly the fault lies with Apple's marketing (i.e. calling the M1 an 8-core processor, without even differentiating P vs E cores). But don't write off the whole architecture... Any of the Max chips are monsters by any account, and each generation gets a good boost. The Pro chips have gotten weaker with each generation, as they swap P for E cores. The plain Mx chips have made small gains with each generation, but yes, they're still behind an i9-9900k for Pro Tools performance. (I mention that because I've tested M1 vs that chip.)

[strangedaysuk]

Originally Posted by huzzam

Honestly the fault lies with Apple's marketing (i.e. calling the M1 an 8-core processor, without even differentiating P vs E cores). But don't write off the whole architecture... Any of the Max chips are monsters by any account, and each generation gets a good boost. The Pro chips have gotten weaker with each generation, as they swap P for E cores. The plain Mx chips have made small gains with each generation, but yes, they're still behind an i9-9900k for Pro Tools performance. (I mention that because I've tested M1 vs that chip.)

Yes agree - I think thats the realisation. I get it - great design in one sense. The marketing though - I never had an issue with a system that reached a point where it became unreliable, I just backed of by 10%. Now I cannot even make that judgement?