i'm on Sonoma M3 Pro.

- in Logic 11 it causes an error, it doesn't finish opening and makes a loud unpleasant noise.

- in Live 12 it works VST3 and AU flawlessly.

- in Cubase 14, my main DAW, it only works in the native version...
BUT in Rosetta mode unfortunately it doesn't work, which is the way i use it so that plugins like those from Roland Cloud, ROLI, Splice, Output, etc. can work.

(hopefully they'll soon update those Roland plugins (and others) so they can be used natively on Apple M chips.)

 

I did your test with 8 notes "Ascension" patch for 4 bars in a loop, track armed on my MacBook Air M1 2020 in Reaper 7.16, VST3 version of the plugin at 44.1 kHz, 256 samples buffer size, internal speakers:

Average 8.6%, peak at about 11% when loop restarts.

I can play 4 of these tracks simultaneously showing about 60 to 80% activity on the 4 power cores of the M1 and nearly no activity on the efficiency cores.

When I try to add a 5th track, audio is crackling and unusable.

So it seems the plugin is single threaded and needs a power core to run although Reaper is able to use efficiency cores.

It's very CPU heavy, but the Ascension patch plays 7 sounds layered, multiplied by 32 notes sums up to 224 voicees.

Without having the tracks armed, I can get 7 tracks to play, that's 392 voices.

 

There is no Rosetta support as stated in the system requirements on Yamahas website:

[OS] : macOS 14.x, macOS 13.x
[Plugin format] : VST 3, AU v2
[CPU minimum] : Intel® Core™ i5 (mid 2017 or later) or Apple silicon (*2), AVX2 support is required for Intel processors (*3)
[Memory] : 8 GB RAM or more
[Storage] : 10 GB free space
[Display resolution] : 1000 × 640 display resolution at least (1440 × 870 recommended)
(*2) Not supported in Rosetta2
(*3) AVX2 support is required for Intel processors (Web site)

 

Yes Reaper is 7.27. Yes is also is playing a more than 4 / 8 bar loop.
The only difference is I have a quantum 2626 as my Audio interface. One is with it armed one is not armed for recording.

What CPU / Audio interface are you using?
Also what patch are you using? Mine is Natural.Women

 

Last edited: Nov 16, 2024

Really no expert in this field, but probably having so much bigger cache results in something...

 

Some more numbers, I tested it on my "Gaming" PC, Ryzen 7 7800X3D (8 cores), 64 GB RAM 6000 MHz, Win 11, Reaper 7.22, 44.1 kHz, 256 samples buffer.

One track averages at around 0.16% in Reaper.

I can get 16 tracks armed to play continuously and 28 tracks unarmed, that's 1568 voices.

But this is far away from a real world scenario, because it peaks at 95% CPU when all notes start playing at once and goes down to only 5% CPU while the notes play.

 

Reaper caches everything that is playing. Unless you disable prerendering there is no way to know how much CPU it takes. The 95% is the building of the cache and the 5% the paying back the cache.

 

In Reaper, it's called "Anticipative FX processing", turning it off results in same performance as having all tracks armed for recording. That's why I have given the numbers "16 tracks armed" and "28 tracks unarmed".

 

the system I tested on is similar to yours and it's using 50% in vsthost

 

How is it that Halion 7 runs about 3%-6% on basic instruments and 35%-45% percent on more complicated synthesis (FM, heavy reverb, high polyphony, etc.), but a basic ethnic/acoustic patch on the Montage M (without reverb) devours 75-100% of the CPU's bandwidth within a matter of seconds, depressing no more than several notes in a basic sequence?

Sure my specs of an Intel i5 iMac with 16GB of RAM and a standard HD might not befit what the Montage M demands, but holy hell even a infamously CPU-hungry plugin such as Roland's own Zenology goes up to 60, maybe 70% tops on a more involved and sophisticated preset but hardly ever clips or distorts the audio output.

The fact that even the cracking teams that've made it accessible have stated their machines couldn't handle it is a testament to the absurd and needless expenditure of CPU bandwidth. We ain't talking about to a T, virtual-analogue, precisely-modeled emulation here.

 

That's the thing too, at least with OsTIrus, it's not only a modeled virtual-analogue synth, but one that aims to utilize the DSP as accurately as the Virus TI used it.

With the Montage M, it may have plenty of powerful synthesis procedures, but it really doesn't justify its disgusting gluttony for the CPU's bandwidth given that it's heavily sample based rather than a soft-DSP generating the sound digitally. Correct me if I'm wrong.

 

The Montage has three large custom Yamaha DSP chips - one for 128 voice AWM engine for it's own samples, another for a 128 voice engine for user samples, and a third for the FM engine (also 128 voices). It then has a hefty main processor with multiple cores and GPUs, and as well as running the firmware and interface etc, that also runs the AN-X VA engine.

The plugin is doing all that work in software alone - each part has a deep synth engine, multiple FX, there are global FX - it's a pretty hefty machine, and the requirements are going to vary according to what resources each performance uses. The fact that software can do the work of those custom DSPs is pretty impressive alone. It is what it is - if you don't have the hardware to run it, run something else that's lighter. Complaining about it isn't suddenly going to make it require half as much resources...

I don't think anyone though is going to get 400 voices of polyphony out of one instance, as it's going to be limited to what one core can do on a general purpose CPU - but you can use multiple instances which will be spread over your cores of course.

On my laptop, it's mostly fine, unless you have a really heavy performance with 8 parts or so, which stretches my core limits to breaking point. Typically it's 40-80% of a core, here, again, depending on polyphony and performance complexity.

 

In the instructions, so that proper comparison can be made:
"- Run 1 instance of Montage Esp Vst and load the patch named "Ascension".
- Play an 8 note, 4 bars length chord in loop, with record armed track and report your CPU consumption
(RT CPU in Reaper, be sure to have the track record armed)."

Running a different patch will give results that can not be compared with any other patch.
Use only the Ascension patch.
The track MUST be armed for recording and the result to compare is the RT CPU.
Just use one 8-note chord, sustained for the full 4 bars as I did in my picture.
Just to be sure of a good result, do not load any other plugins.
In this, my first track is just an empty track

 

Last edited: Nov 21, 2024 at 2:53 AM

I just found it sounding "dated" ... so i removed it.
Not realistic enough to replace Kontakt.
Synthesis sounding average compared to my others synths.

OUT

PS : on the other side, KORG ModWave totally impressed me

 

Last edited: Nov 21, 2024 at 8:07 AM

I have both Mac and Windows version , and both sound very good . Having Zenology from Roland , Triton , Prophecy and MicroKorg from Korg , this complete my synths need . Not to talk about the 4000 arpeggiators . In terms of using this , I use only 3 instances max , My music being simple. So no problems here with the CPU .

 

No it's not "lousy programmed", it's doing in software the work of three custom large-scale DSP chips purposely designed to run the AWM and FM engines, and the TI ARM processor running the OS and AN-X engine, and that work is significant.

Those four chips together are *significantly* more powerful than the decades old 56300 DSP chip, and even then, much as I love the usual suspects stuff, that DSP56300 engine uses the best part of a core on Apple silicon too (set latency to 0 to see the real CPU usage), which is *more* than the Yamaha uses for an equivalent polyphony.

 

The AWM engine is sample-based. The FM-X engine is modelled. The AN-X VA engine is modelled.

 

there's no reason to model chips when you have the original code and can just port it
I'm sure its just a port that is not optimized at all yet, but maybe it might be in the future

I enjoy getting to play with the new fm engine so I can finally compare it to the fs1r I've kept forever