Performance on M1 Macs
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
So I'm about to start this test again, but one of the reasons that I got interested in Bitwig is it's outstanding performance with U-He Diva, I got something like 24 instances on great mode on an 09 12 core Mac Pro.
Recently I bought a M1 MacBook Air, love it, installed Bitwig and.. damn? I'm getting less than 20 instances??? Compare this to DP11 where I'm getting around 43! I think I got something like 13 instances in DP10 on the Mac Pro.
Anyone else using M1 Macs have any clue as to what's going on here? I'm going to run more tests tonight but my initial thoughts are Bitwig is not at all optimized for M1, it might be "native" but it's running poorly. I don't expect it to get what DP can, just at least more than it does on a ten year old chipset. Hopefully I'm proven wrong. FYI CPU performance meters in Activity Monitor show them pegging almost in DP at 43 instances, and only using half of the 8 cores, I don't have a way to determine it but I would assume it's not using the Efficiency cores art all and at 50% or so on the Performance cores before spiking the audio.
Recently I bought a M1 MacBook Air, love it, installed Bitwig and.. damn? I'm getting less than 20 instances??? Compare this to DP11 where I'm getting around 43! I think I got something like 13 instances in DP10 on the Mac Pro.
Anyone else using M1 Macs have any clue as to what's going on here? I'm going to run more tests tonight but my initial thoughts are Bitwig is not at all optimized for M1, it might be "native" but it's running poorly. I don't expect it to get what DP can, just at least more than it does on a ten year old chipset. Hopefully I'm proven wrong. FYI CPU performance meters in Activity Monitor show them pegging almost in DP at 43 instances, and only using half of the 8 cores, I don't have a way to determine it but I would assume it's not using the Efficiency cores art all and at 50% or so on the Performance cores before spiking the audio.
- Banned
- 11467 posts since 4 Jan, 2017 from Warsaw, Poland
I'm kind of dissapointed that we don't really know how well M1 performs with Bitwig... For example, to which Intel or AMD processor it compares to in terms of real-world DAW performance?
I found a great deal on MBP '13 M1 and was considering getting it, because I don't really need tons of performance and paying at least 3x the price for 2x CPU performance boost in new Pro/Max lineup doesn't sound appealing (and I don't care about more ports, better screen, better camera, better speakers, etc.). But what if it's much worse than my current laptop, i.e. i7-8750h? Those would suggest it beats it by 40-70% in both single- and multi-core workflows, but does it really?
https://www.cpubenchmark.net/compare/Ap ... 4104vs3237
https://nanoreview.net/en/cpu-compare/i ... s-apple-m1
I found a great deal on MBP '13 M1 and was considering getting it, because I don't really need tons of performance and paying at least 3x the price for 2x CPU performance boost in new Pro/Max lineup doesn't sound appealing (and I don't care about more ports, better screen, better camera, better speakers, etc.). But what if it's much worse than my current laptop, i.e. i7-8750h? Those would suggest it beats it by 40-70% in both single- and multi-core workflows, but does it really?
https://www.cpubenchmark.net/compare/Ap ... 4104vs3237
https://nanoreview.net/en/cpu-compare/i ... s-apple-m1
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
I'm not sure that's accurate? more like there aren't enough people like me who bought an M1 and are using Bitwig natively yet to determine? The performance with DP11.02 suggests that the M1 far outstrips Xeons and i7 chips if the DAW can handle it. I don't expect Bitwig to get 43 instances of Diva or anything, that's as much due to the buffering/pre-rendering DAWs like Cubase, Logic and DP do to etch out more tracks, but it should IMO come in at the typical 60-70% ratio to those DAWs, so somewhere around 30. Like I said the CPU meters in Mac OS Activity Monitor showed Bitwig not utilizing the cores before something is spiking, so it could be a Diva M1 VST + Bitwig anomaly.
I would say 90% of the time this is accurate, but there are oddities due to the newness of the chips. A USB-C SSD enclosure I have is incompatible with the Air, it works fine on the Xeon Mac Pro, but it comes in at a whopping 45mbs read speed on the Air. Moving the SSD to a USB3 enclosure I have gets it back up to USB3 speeds of around 350Mbs.I found a great deal on MBP '13 M1 and was considering getting it, because I don't really need tons of performance and paying at least 3x the price for 2x CPU performance boost in new Pro/Max lineup doesn't sound appealing (and I don't care about more ports, better screen, better camera, better speakers, etc.). But what if it's much worse than my current laptop, i.e. i7-8750h? Those would suggest it beats it by 40-70% in both single- and multi-core workflows, but does it really?
https://www.cpubenchmark.net/compare/Ap ... 4104vs3237
https://nanoreview.net/en/cpu-compare/i ... s-apple-m1
The solid performance from DP11 makes me think if anything it's a bug in Bitwig with Diva that's having this issue. I'll test more tomorrow but it's 2AM here and I have to work in the AM.
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
OK because I'm an idiot and hate sleep apparently, I ran the test again.
I'm getting 20 Diva instances in VST2 and VST3 natively. This is not bad, I just foolishly figured it would beat the 12 core Xeon here. What I see in Activity Monitor points to it using the Performance Cores and stacking 5 per core before crackling. Compare this to the Mac Pro with 12 cores getting only two instances per core, but eking out 24 by sheer core count.
What's interesting and weird is other DAWs for the most part performed quite worse with the 12 core, so DP11 for instance just smokes on the M1 so far. Again this is comparing a laptop to a workstation, the 12 core 3.33ghz Xeons are still good chips. DP11 has a "live mode" which is supposed to make it more like Bitwig and Live, tomorrow I'll see how it adds up without the rendering and buffering that it does to boost track count.
BTW the other reason I got the Air is it's never twice as powerful. The number quoted was 1.7 times and that looks accurate when you see the Geekbench scores.
I'm getting 20 Diva instances in VST2 and VST3 natively. This is not bad, I just foolishly figured it would beat the 12 core Xeon here. What I see in Activity Monitor points to it using the Performance Cores and stacking 5 per core before crackling. Compare this to the Mac Pro with 12 cores getting only two instances per core, but eking out 24 by sheer core count.
What's interesting and weird is other DAWs for the most part performed quite worse with the 12 core, so DP11 for instance just smokes on the M1 so far. Again this is comparing a laptop to a workstation, the 12 core 3.33ghz Xeons are still good chips. DP11 has a "live mode" which is supposed to make it more like Bitwig and Live, tomorrow I'll see how it adds up without the rendering and buffering that it does to boost track count.
BTW the other reason I got the Air is it's never twice as powerful. The number quoted was 1.7 times and that looks accurate when you see the Geekbench scores.
- Banned
- 11467 posts since 4 Jan, 2017 from Warsaw, Poland
Sure, number of users is one thing, but unwillingness to do proper tests is also a factor. I asked one of the members here, when v4 was released, to test the same project in Bitwig running via Rosetta2 and natively to see how big is the difference (if any), but they refused. Same with Mattias from YT, who made couple of videos about M1, but no actual comparison.machinesworking wrote: ↑Tue Oct 26, 2021 1:12 amI'm not sure that's accurate? more like there aren't enough people like me who bought an M1 and are using Bitwig natively yet to determine?
-
- KVRer
- 2 posts since 10 Dec, 2020
I ran a quick test on my M1 MacBook Air (16gb RAM version):
Bitwig 4.0.6, latest Diva, Built-in speakers, 48k, 24bit, buffer size 256, plugins hosted inside Bitwig, default start-up preset on each Diva, the same 3-voice chord on each track.
The result: 34 instances before any overloads occur. As you can see from the screenshot, the performance CPU cores are pretty much saturated, so it would seem that Bitwig is quite well optimized.
Bitwig 4.0.6, latest Diva, Built-in speakers, 48k, 24bit, buffer size 256, plugins hosted inside Bitwig, default start-up preset on each Diva, the same 3-voice chord on each track.
The result: 34 instances before any overloads occur. As you can see from the screenshot, the performance CPU cores are pretty much saturated, so it would seem that Bitwig is quite well optimized.
You do not have the required permissions to view the files attached to this post.
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
I used a single quarter note run with no breaks, might be the difference, but I also had what looks like a square wave pattern instead of the flat peak you're getting with the performance meter. Also maybe due to the difference in MIDI patterns.Extrup wrote: ↑Tue Oct 26, 2021 3:23 am I ran a quick test on my M1 MacBook Air (16gb RAM version):
Bitwig 4.0.6, latest Diva, Built-in speakers, 48k, 24bit, buffer size 256, plugins hosted inside Bitwig, default start-up preset on each Diva, the same 3-voice chord on each track.
The result: 34 instances before any overloads occur. As you can see from the screenshot, the performance CPU cores are pretty much saturated, so it would seem that Bitwig is quite well optimized.
Bitwig 34 Divas.png
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
Running the same test at 256 48 with a dotted half note, three note chord, two bar measure, can run 35 instances. This is the issue with non standardized tests. The quarter note run is allowing voices to pile up, and even with the chord if I let the chord overlap release wise it can't get anywhere near 34 instances.
Also, it might be noted that the disparity between DP11 and Bitwig here might be due to a few different things, Bitwig is relying on VST, VST3 (tested on both BTW, no difference in performance), and does not do the whole pre - buffering thing DP etc. do, but as you can see from Extrups example jpg above Bitwig isn't really using the Efficiency cores at all, and DP at 43 instances of Diva near pegged all 8 cores.
It's just really interesting to me how this plays out, the old Xeons here fit two instances per core perfectly in this same test, and DP choked much much sooner than Bitwig under the 12 core Xeon chips. The opposite is mostly true with the M1, though it's fairer to say that the M1 with it's larger CPU per per core, is capable of smartly allocating resources to things like pre-rendered AU's whereas raw CPU power wise, 12 cores will beat 4 Performance cores. Later tonight I'll do this same test in DP11 with the Diva VST, AU is the default in DP11 because AU is the only format capable of running Rosetta plug ins in a native M1 host like DP and Bitwig etc.
Also, it might be noted that the disparity between DP11 and Bitwig here might be due to a few different things, Bitwig is relying on VST, VST3 (tested on both BTW, no difference in performance), and does not do the whole pre - buffering thing DP etc. do, but as you can see from Extrups example jpg above Bitwig isn't really using the Efficiency cores at all, and DP at 43 instances of Diva near pegged all 8 cores.
It's just really interesting to me how this plays out, the old Xeons here fit two instances per core perfectly in this same test, and DP choked much much sooner than Bitwig under the 12 core Xeon chips. The opposite is mostly true with the M1, though it's fairer to say that the M1 with it's larger CPU per per core, is capable of smartly allocating resources to things like pre-rendered AU's whereas raw CPU power wise, 12 cores will beat 4 Performance cores. Later tonight I'll do this same test in DP11 with the Diva VST, AU is the default in DP11 because AU is the only format capable of running Rosetta plug ins in a native M1 host like DP and Bitwig etc.
-
- KVRer
- 2 posts since 10 Dec, 2020
Exactly - that's what I found too. It's all about the amount of active voices. Guess it's time for a standardized Bitwig-Diva performance benchmarkThis is the issue with non standardized tests. The quarter note run is allowing voices to pile up, and even with the chord if I let the chord overlap release wise it can't get anywhere near 34 instances.
Yep, anything that shifts the processing load from the realtime domain into a pre-buffered domain like for example Cubase's ASIO Guard or apparently DP as well (I was unaware) does, will give more performance - but only until all instruments are shifted back into the realtime domain as it would happen if you record enable them. My guess, since Bitwig is also designed for live performance, this is avoided on purpose so every instrument is retains instant playability? Someone correct me if I'm wrong please.buffering thing DP etc.
This may be a good thing. I found that DAWs that use the efficiency cores (I think all DAWs will once you run them in Rosetta mode) can get dropouts when the load suddenly changes, even if the maximum load could actually be handled by the performance cores. For example, if you have your session's instrumentation suddenly expand after being small and not very resource-consuming for a while, chances are, you will get crackles because the thread needing more processing is handled by the insufficient efficiency core.Bitwig isn't really using the Efficiency cores at all
- KVRist
- 62 posts since 6 Sep, 2021
Been taking Bitwig through its paces on my new MacBook Pro 14" M1 Pro and Bitwig is absolutely flying through it. All the native devices and presets load up instantly, as do M1 Native plugins such as Valhalla, Overloud TH-U, and Serum. Coming from a MacBook Pro 16" i9, the difference is very apparent. Large projects (80-100 tracks) that would glitch out at a 512 buffer on the Intel MacBook now run at about half the DSP meter on a 256 buffer size.
One annoying issue though is that non native plugins take a few seconds to load, presumably to activate the Rosetta wrapper. So when opening an existing project with a lot of non native plugins, the project takes a fair bit longer to load than on my Intel machine. I experience the same issue with Ableton Live 11. With both DAW's though, once the project is loaded, the performance is extremely smooth.
One annoying issue though is that non native plugins take a few seconds to load, presumably to activate the Rosetta wrapper. So when opening an existing project with a lot of non native plugins, the project takes a fair bit longer to load than on my Intel machine. I experience the same issue with Ableton Live 11. With both DAW's though, once the project is loaded, the performance is extremely smooth.
multi-platinum music producer / songwriter
https://www.takaperry.com
https://www.takaperry.com
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
A standardized test for any DAW makes sense to me. Though one thing you realize when doing stress tests, the way a plug loads into a single core matters, a lot. I tested multiple DAWs using Diva and Reaktor Blocks and Bitwig, Diva, and Xeon Mac Pros are a winning combination. Bitwig along with Live was near last in line with Blocks though, for whatever reason Logic, DP and Reaper would skunk the performance DAWs with Block, badly. So the problem is you need to own multiple plug ins from different manufacturers to get a real picture, which you hint at.
No this is right, I don't own Cubase, but I have Logic, Reaper and DP11 here and all of them use some sort of pre-rendering into RAM or buffering to alleviate CPU strain, which means all of them are more touchy when doing things like adding instruments while the sequence is running etc. DP11 just added a "Live Performance Mode" which mostly does what you would expect, but it's still not as "uninterrupted audio" as Bitwig or Live. It's a good idea though, that Live and Bitwig should think about in reverse. Seems more like something Bitwig would do.Yep, anything that shifts the processing load from the realtime domain into a pre-buffered domain like for example Cubase's ASIO Guard or apparently DP as well (I was unaware) does, will give more performance - but only until all instruments are shifted back into the realtime domain as it would happen if you record enable them. My guess, since Bitwig is also designed for live performance, this is avoided on purpose so every instrument is retains instant playability? Someone correct me if I'm wrong please.buffering thing DP etc.
I dunno that seems a bit fishy to me? I'll take all the CPU I can get, I don't think Mac OS is incapable of allocating resources when cores reach their limit and I'm willing to bet the Performance Cores fill up first, you're only going to use Efficiency cores once you're already taxing your system. At some point you run out of resources but both our tests prove that Bitwig craps out before touching the Efficiency cores, I would like that CPU, it's essentially another core on the Air.This may be a good thing. I found that DAWs that use the efficiency cores (I think all DAWs will once you run them in Rosetta mode) can get dropouts when the load suddenly changes, even if the maximum load could actually be handled by the performance cores. For example, if you have your session's instrumentation suddenly expand after being small and not very resource-consuming for a while, chances are, you will get crackles because the thread needing more processing is handled by the insufficient efficiency core.Bitwig isn't really using the Efficiency cores at all
- Banned
- 11467 posts since 4 Jan, 2017 from Warsaw, Poland
If I understand you correctly (that Bitwig should introduce some sort of pre-buffering) then Claes was pretty adamant that it's not something they're looking into, as it goes against the core priniciples of Bitwig of modularity, audio-rate modulation, live performance features, etc. There will always be a performance "penalty" for Bitwig, but I prefer to think of it as a trade-off that allows me to work in it the way I mostly can't in any other DAW.machinesworking wrote: ↑Tue Oct 26, 2021 10:09 pm...DP11 just added a "Live Performance Mode" which mostly does what you would expect, but it's still not as "uninterrupted audio" as Bitwig or Live. It's a good idea though, that Live and Bitwig should think about in reverse. Seems more like something Bitwig would do.
- Banned
- 11467 posts since 4 Jan, 2017 from Warsaw, Poland
Thanks for that!takaperry wrote: ↑Tue Oct 26, 2021 9:08 pm Been taking Bitwig through its paces on my new MacBook Pro 14" M1 Pro and Bitwig is absolutely flying through it. All the native devices and presets load up instantly, as do M1 Native plugins such as Valhalla, Overloud TH-U, and Serum. Coming from a MacBook Pro 16" i9, the difference is very apparent. Large projects (80-100 tracks) that would glitch out at a 512 buffer on the Intel MacBook now run at about half the DSP meter on a 256 buffer size.
Do you happen to have any experience of Bitwig running on new 14'' M1 Pro vs. the "old" 13'' M1?
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
Yeah I get it kind of, but I think they're missing out. Running DP in Live Performance mode drops me down to 24 instances in my single note run and 20 in Extrups version with a single chord. Compare that to less than 20 instances in Bitwig on that same quarter note run, and 43 instances in DP11 with Live Performance Mode turned off. Using DP11 is more than the cpu performance gain from an M1 Air to the M1 Max. It's the same experience I had with the Live and Logic days, for songs where I don't have an idea of what I want to do, Bitwig is far more fun, for songs where I have distinct ideas etc. it's just easier to use a DAW like DP that has a lot more juice.antic604 wrote: ↑Tue Oct 26, 2021 10:22 pm If I understand you correctly (that Bitwig should introduce some sort of pre-buffering) then Claes was pretty adamant that it's not something they're looking into, as it goes against the core priniciples of Bitwig of modularity, audio-rate modulation, live performance features, etc. There will always be a performance "penalty" for Bitwig, but I prefer to think of it as a trade-off that allows me to work in it the way I mostly can't in any other DAW.
-
machinesworking machinesworking https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=8505
- KVRAF
- Topic Starter
- 6850 posts since 15 Aug, 2003 from seattle
back to this, DP11 in Live Performance Mode is exhibiting the same use of only Performance cores. So it's likely that MOTU coded their Pregen as they call their buffering, to use the Efficiency cores when everything is pegged.Extrup wrote: ↑Tue Oct 26, 2021 2:41 pmThis may be a good thing. I found that DAWs that use the efficiency cores (I think all DAWs will once you run them in Rosetta mode) can get dropouts when the load suddenly changes, even if the maximum load could actually be handled by the performance cores. For example, if you have your session's instrumentation suddenly expand after being small and not very resource-consuming for a while, chances are, you will get crackles because the thread needing more processing is handled by the insufficient efficiency core.Bitwig isn't really using the Efficiency cores at all