Which daw has the best CPU load management?

[kvotchin]

On windows ASIO drivers matter a lot, they make or break the performance of the system, if you can grab an RME.

Indeed. My usual preferred setup is Reaper w/ an RME interface, and I couldn’t be happier with the responsiveness, super-low latency, and so on.

[syntonica]

Some guy did tests on an M2 Ultra Mac Studio. Logic did 2x what other DAWs did.

https://m.youtube.com/watch?v=3CBIdw0L7mM&t=2728s

I skimmed through all four of this guy's videos and it seems he has no clue what he's doing. (Using Kontact notwithstanding. )

Regardless, if you really need over 1000 sampler tracks, the Studio M2 Ultra with 192Gb RAM seems to be the way to go.

[PAK]

even if it is of the latest generation and complies with the demands of DAW developers, it seems more complicated to find stability compared to the latest Mac M2?

Although the traditional reason given, when people say that sort of thing, is usually something about a “bigger PC ecosystem” (more hardware driver combos etc) there’s more going on at this point.

First, there’s some degree of a fresh start with Apple Silicon. Though it’s a theoretical advantage (which sacrificed longer-term compatibility to make the transition) all I can say is Cubase stability has never been so good, to the point I would recommend any PC Cubase user (who can) considers switching to AS at this point. The difference, for me, has been night and day.

Secondly, Microsoft just can’t seem to get power management right - especially when it comes to real-time audio needs. Almost every meaningful Windows/BIOS audio tweak you’ll see is about altering or turning their “management” of these things off. The issues, with different core types, just seems to be an extension of their difficulties in these areas. Maybe Windows on ARM will allow for a similar fresh start. But, for now, it’s a definite difference and something which impacts desktops as much as laptops for audio.

Thirdly, there is the hardware. The new Macs are systems on a chip, with an expectation that everything else is added via Thunderbolt and follows standards based protocols (IE no drivers required). Along with that, where stuff does still require drivers, Apple’s moving everything out of the kernel side, which prevents drivers from crashing the OS. (RME have made the transition and it doesn’t seem to have impacted their performance in any noticeable way). So the better stability is beyond theoretical at this point - there are actual reasons for it. This also means hardware should, at least in theory, experience less issues with future Mac OS updates.

Of course, the major downside to Apple’s approach (and it’s a big one) is that your main computer specs are limited to what you buy at the time of purchase. Audio users also need to consider avoiding AU if they can, as it’s historically caused issues for 3rd party plugins with OS updates. VST (and, I expect, CLAP) has faired significantly better in this department.

For PC’s all you can really do is not cheap out on the power supply (power stability helps everything), stick to better known component brand names (Samsung etc) and hope the motherboard maker did the same (An easy way to tell if a mobo is cheaping out is to check the Intel boards use Intel networking and Intel USB rather than 3rd party chips.. ) The 3rd party motherboard solutions will often tend to give more issues.

[PAK]

[I skimmed through all four of this guy's videos and it seems he has no clue what he's doing. (Using Kontact notwithstanding. )

Regardless, if you really need over 1000 sampler tracks, the Studio M2 Ultra with 192Gb RAM seems to be the way to go.

There's enough variables that it's pretty difficult to do any sort of meaningful testing with this stuff.

For instance, it's pretty common you'll see people saying they can do "x" instances of Diva. The reality is you can still bring down an AS Ultra chip with just 8 tracks (instances) of Diva if you know the right (wrong? ) settings to use. Which isn't meant as criticism of Diva btw - which is excellent in terms of the amount of control it gives. It's more of a statement on how most people aren't aware of the variables involved (EG Diva will use a lot more CPU power if the resonance is enabled Vs Off etc.. )

[syntonica]

[I skimmed through all four of this guy's videos and it seems he has no clue what he's doing. (Using Kontact notwithstanding. )

Regardless, if you really need over 1000 sampler tracks, the Studio M2 Ultra with 192Gb RAM seems to be the way to go.

There's enough variables that it's pretty difficult to do any sort of meaningful testing with this stuff.

For instance, it's pretty common you'll see people saying they can do "x" instances of Diva. The reality is you can still bring down an AS Ultra chip with just 8 tracks (instances) of Diva if you know the right (wrong? ) settings to use. Which isn't meant as criticism of Diva btw - which is excellent in terms of the amount of control it gives. It's more of a statement on how most people aren't aware of the variables involved (EG Diva will use a lot more CPU power if the resonance is enabled Vs Off etc.. )

I was trying to be tongue-in-cheek here. Not sure how I did.

I was concerned about things like I think he said he had left all the Cubase tracks armed. I'm not familiar enough with Cubase to see in the video, but I do know that leaving even one track armed can affect playback negatively. Plus, using the in-application CPU meters is not useful at all except for testing within that application. Different DAWS report CPU usage differently. You need to go with a system-reported value, from Stats, Menu Meters, or just the Activity Monitor.

Also, Logic may be smart enough to tell that the same samples are being loaded repeatedly and optimizing reuse rather than loading duplicates. I would need to see RAM usage to confirm, but filling the RAM and causing repeated swaps will pretty much kill the vibe there.

Not sure why he kept on about his RME interface, either. Does it use it's own drivers? I can't imagine that they'd be any more efficient than using CorepAudio and why you'd want to bypass it as it's very efficient and flexible.

Anyway, too much hinky stuff going on in those tests. If anybody wants to send me a Mac Studio Ultra with 192GB/1TB, I'd be happy to do some proper testing.

[PAK]

I was concerned about things like I think he said he had left all the Cubase tracks armed. I'm not familiar enough with Cubase to see in the video, but I do know that leaving even one track armed can affect playback negatively.

Armed gives an indicator of real-time performance. Unarmed will buffer using ASIO-Guard, provided you’re using an ASIO interface. In theory the extra buffering can help push higher track counts, but also adds yet more variables since you can adjust the buffer. So, depending on the context, it can make sense to test with all tracks armed.

You make some very good points about things like how hosts might manage duplicated ram use, and how a lack of memory will tank performance, and I don’t know any specifics of what Logic does there. I do know memory bandwidth seems to help Kontakt, so you’ll get better performance with an Ultra due to its 800GB/s memory bandwidth, beyond just the extra core count.

Not sure why he kept on about his RME interface, either. Does it use it's own drivers? I can't imagine that they'd be any more efficient than using CorepAudio and why you'd want to bypass it as it's very efficient and flexible.

You’d want ASIO for ASIO guard specifically in Cubase. Otherwise it has little reason to exist on Macs. It actually runs on top of Core Audio anyway (Steinberg provides a transition layer to Cubase).

Beyond that, ASIO cards may not to expose all of their I/O for general Mac OS usage, meaning you might still benefit from ASIO support to easily see the I/O (unless you want to resort to Audio Midi Setup and so-called “aggregate devices” etc). So there are sometimes other reasons.

Re Performance: At low buffer settings the onboard, with core audio, is still approx 1ms higher than RME ASIO using Thunderbolt. So, beyond I/O, there’s still a reason you'd want to use RME ASIO. If using USB (whether 2 or 3) it's generally around 2ms higher than Thunderbolt with lower buffers. So that reasoning wouldn't apply, and onboard with Core Audio should typically be expected to outperform USB sound devices for latency..

[jens]

...
E-cores are basically half speed cores compared to P-cores with HT.
- and no daws so far really support placing the right threads on P-cores only AFAIK.
- what I saw in ProcessLasso was same thread handling like E-cores were as main P-cores.
- just a year ago Cubase had warning not supporting E-cores, not sure if that is handled now

This E-cores thingie is the biggest BS ever introduced into CPU´s and just betraying to bumb the number of cores without the technology to make them real cores...

Intel should hire you immediately because you obviously know this CPU-architecture stuff a lot better than these clueless rookies who design their completely flawed chips.

I bet you could come up with a much better and a whole lot more efficient design in a day or two.

That's what I love about KVR: there's some real experts sharing their wisdom here.

[machinesworking]

Most of the YouTube videos showing near 100 tracks with heavy VSTs have the buffer size set to 1024.
Suitable for mixing, not for production.
If you are recording with a midi keyboard, you have to drop the buffer size to less than 256.
In that case may be 30 tracks are possible.
In short, the CPU have not developed that much, but better now.
Regards.

140 tracks of Diva in DP11 and Reaper at a buffer of 128, 48khz. Mac Studio 24 cores 16P AND 8E.
There is definitely a difference with new CPUs.

[PAK]

140 tracks of Diva in DP11 and Reaper at a buffer of 128, 48khz. Mac Studio 24 cores 16P AND 8E. There is definitely a difference with new CPUs.

If you’re willing to test something.. as it seems you're on an M2 Ultra (M1 Ultra here.. ). Go to Browser / Diva Factory Third Party and load “PAK Cycle Parc”. Change Mode to Poly, Voices to 16, and Stack to 4, Accuracy to Divine. Important: Do not enable Diva’s multicore setting. Latency doesn’t matter much as long as it’s useable realtime (sub 10ms).

Can you play a 4 note chord without glitches? The M1 Ultra has to be limited to 12 notes without multi-core. I think the M2 won’t quite make it to 16 notes either, but it’ll be close The M3 Ultra will likely be able to finally fit a 16 note Divine Diva inside of a single performance core.

Of course, if you’re curious (whether it manages 16 or you need to bump down to 12) you can then duplicate the tracks, and see how many voices of polyphony you can manage to get with Diva set to Divine mode before it overloads the CPU ..Curious if the M2 ultra can make it past 8 tracks

BTW For anyone wondering.. You can bump Diva down to “Great quality” with no real impact to that particular sound, and it should allow about 2.5x the voice count (IE almost 200 notes of polyphony, and likely 220+ on an M2 Ultra).

[Trancit]

...
E-cores are basically half speed cores compared to P-cores with HT.
- and no daws so far really support placing the right threads on P-cores only AFAIK.
- what I saw in ProcessLasso was same thread handling like E-cores were as main P-cores.
- just a year ago Cubase had warning not supporting E-cores, not sure if that is handled now

This E-cores thingie is the biggest BS ever introduced into CPU´s and just betraying to bumb the number of cores without the technology to make them real cores...

Intel should hire you immediately because you obviously know this CPU-architecture stuff a lot better than these clueless rookies who design their completely flawed chips.

I bet you could come up with a much better and a whole lot more efficient design in a day or two.

That's what I love about KVR: there's some real experts sharing their wisdom here.

And it took just 2 weeks to come up with this genius comment???
Respect!!!

You are a real high-flyer... now I remember again why you are on my foe list...

[machinesworking]

Along with that, where stuff does still require drivers, Apple’s moving everything out of the kernel side, which prevents drivers from crashing the OS. (RME have made the transition and it doesn’t seem to have impacted their performance in any noticeable way). So the better stability is beyond theoretical at this point - there are actual reasons for it. This also means hardware should, at least in theory, experience less issues with future Mac OS updates.

So this is what Core Audio was about more than 15 years ago, any stability Mac OS experiences over Windows in this department is nothing new to Apple Silicon. Core Audio literally limits a developers ability to talk directly to the kernal, but it's more than good enough. There is a theoretical advantage to PCs in terms of getting latency to near zero, but of course at the cost of wasting way too many CPU cycles on low latency.

Audio users also need to consider avoiding AU if they can, as it’s historically caused issues for 3rd party plugins with OS updates. VST (and, I expect, CLAP) has faired significantly better in this department.

Yeah, that's not true at all, in any way shape or form. AU was conceived to limit issues with cross platform plugins being ported with remnants of Windows code in them, it in the past would sometimes carry a slightly higher CPU load, but that was over a decade ago and any tests I've done since then are a dead match. I've used both over the years and the only thing I can say without a doubt is that even to this day VST3 is the most likely to be unstable, on Apple Silicon or Intel.

Again, over the years VST has never been more stable than AU, you're safe using either, and personally I instal all three because both of my main DAWs (DP and Live), handle all three formats, and all three can have issues in a particular DAW, so it's really unlikely that the instability is in all three formats.

[jens]

...
E-cores are basically half speed cores compared to P-cores with HT.
- and no daws so far really support placing the right threads on P-cores only AFAIK.
- what I saw in ProcessLasso was same thread handling like E-cores were as main P-cores.
- just a year ago Cubase had warning not supporting E-cores, not sure if that is handled now

This E-cores thingie is the biggest BS ever introduced into CPU´s and just betraying to bumb the number of cores without the technology to make them real cores...

Intel should hire you immediately because you obviously know this CPU-architecture stuff a lot better than these clueless rookies who design their completely flawed chips.

I bet you could come up with a much better and a whole lot more efficient design in a day or two.

That's what I love about KVR: there's some real experts sharing their wisdom here.

And it took just 2 weeks to come up with this genius comment???
Respect!!!

You are a real high-flyer... now I remember again why you are on my foe list...

Nah, it took me two weeks to have a moment dull enough so that I would stoop down to re-visiting this thread and actually read crappy posts such as yours (sometimes when I can't sleep I find stupid stuff kind of entertaining which I normally wouldn't waste my time on).

[machinesworking]

140 tracks of Diva in DP11 and Reaper at a buffer of 128, 48khz. Mac Studio 24 cores 16P AND 8E. There is definitely a difference with new CPUs.

If you’re willing to test something.. as it seems you're on an M2 Ultra (M1 Ultra here.. ). Go to Browser / Diva Factory Third Party and load “PAK Cycle Parc”. Change Mode to Poly, Voices to 16, and Stack to 4, Accuracy to Divine. Important: Do not enable Diva’s multicore setting. Latency doesn’t matter much as long as it’s useable realtime (sub 10ms).

Can you play a 4 note chord without glitches? The M1 Ultra has to be limited to 12 notes without multi-core. I think the M2 won’t quite make it to 16 notes either, but it’ll be close The M3 Ultra will likely be able to finally fit a 16 note Divine Diva inside of a single performance core.

Of course, if you’re curious (whether it manages 16 or you need to bump down to 12) you can then duplicate the tracks, and see how many voices of polyphony you can manage to get with Diva set to Divine mode before it overloads the CPU ..Curious if the M2 ultra can make it past 8 tracks

BTW For anyone wondering.. You can bump Diva down to “Great quality” with no real impact to that particular sound, and it should allow about 2.5x the voice count (IE almost 200 notes of polyphony, and likely 220+ on an M2 Ultra).

Here's the issue with that particular stress test. You can always overload a single CPU in any DAW, doing multiple held notes in divine isn't going to allow for multiple plugins on one core. Using absolutely the biggest CPU plugins doesn't allow a multi core system to allocate plugins to different cores, the only DAW I know of where the developers themselves say that it can manage CPU from a single track on multiple cores is Bitwig, it's probably why it does better than Live in stress tests.

I don't know about you but 90% or more of my tracks are not polyphonic chords on heavy CPU plugins, but I will use a hundred plugins of various CPU piggery. So a hundred or more heavy CPU, but not the absolute heaviest pads makes more sense to me to determine which DAW manages cores the best, not as much how much the computer hardware holds up.

For fun I can do it though, I'll get beck to you on that.

[Freaqpeak]

Since i bought my AMD 3950x no project has gone over 80%.. running 60 tracks of synth vsti's and fx and 40 tracks of audio tracks + fx. Mostly Hive and Zebra. Some diva and vital as well.
128 sample latency with RME + Cubase 13..
Cubase uses all 16 cores, np..

I agree tho that Reaper is prob most efficient. Too bad I hate the workflow and lack of features..
Bitwig uses more cpu but I dont use as many tracks, mostly bitwigs own synths and fx ..never ever got crackles in bitwig either.

[PAK]

So this is what Core Audio was about more than 15 years ago

The push to kill KEXT's is more recent, and I reckon they're going to end support completely with the end of Intel support (likely in about 2 years time?.. We'll see )

Yeah, that's not true at all, in any way shape or form.

You’re honestly arguing AU hasn’t historically had more issues?! Perhaps you have examples of lots of AU plugins, coded prior to 2009, which continued to function on current hosts right up until 32 bit support was pulled? I’m genuinely curious to know what they are if you do?.. The number of VST 2 plugins which could achieve this are too numerous to mention, but included well known older plugins like Pro-53, FM7, Steinberg Hypersonic 2 etc.

Me-thinks you’re confusing this ability with the numerous (well documented) issues with the transition to VST 3. That isn't quite the same thing as Apple completely breaking your ability to load your plugins. Not even close.

Apple (naturally) use their own software to push their own standards. If they update Logic and AU in some way which breaks things for 3rd parties then the attitude is “too bad”. If users want to be subject to that whim then that’s ok, just as long as you know what using AU means. I would rather a standard like CLAP succeeded over either of them btw ;)