fm module

[gentleclockdivider]

About the fm module , currently the modulation signal flow goes from left to right

To replicate certain dx 7 algoritms we need to enable the global feedback parameter for all operators , the problem is that this increases the cpu considerably .
Take dx7 algoritm .nr4 , (see screenshot )
There is one feedback loop , but inside the feedback loop there are three operators ( 4,5,6 )
Enabling individual operetor feedback on MSF. will not replicate the algortim , we need enable global feedbackto replicate this particular algoritm .
Because the output of operator 4 goes back to 6 , which in turn goes into 5 etc..

Another dx 7 algoritm which has more then 1 operator in the fedback loop is algo nr.6 .

Please consider another way for operator routings ( matrix style ) because currently it is imho unusable , especially because of the global feedabck ( cpu ) limitations

[jmg8]

When you say "global feedback parameter" are you referring to the "Enable feedback" button?
If so, the CPU is really not that bad and this is the only way to achieve certain algorithms.
All the DX7 algorithms are already there in the presets, only a few require the enable feedback to be on. Due to different versions of the ordering of the algorithms, they might not be in the same order as you think. For example here you refer to algorithm 4, in the presets this is algorithm 32.
They are all there and working correctly.
Due to the order of the operators in DX7, I had to reverse the numbers to make it work.
DX7 OP1 = MSF OP6
DX7 OP2 = MSF OP5
DX7 OP3 = MSF OP4
DX7 OP4 = MSF OP3
DX7 OP5 = MSF OP2
DX7 OP6 = MSF OP1

I also added there all the algorithms from FM8 and some complex new stuff.

[dmbaer]

Is there even a definitive list of six-op algorithms and associated numbers? I tried to track this down not long ago, and found a least two lists that did not correspond to each other.

I know what your thinking ... an error on the internet? How could this be?

[jmg8]

Hehe. Yeah I did the same.
For the MSF FM matrix presets I copied the numbers/order from Arturia DX7 V.

[gentleclockdivider]

When you say "global feedback parameter" are you referring to the "Enable feedback" button?
If so, the CPU is really not that bad and this is the only way to achieve certain algorithms.
All the DX7 algorithms are already there in the presets, only a few require the enable feedback to be on. Due to different versions of the ordering of the algorithms, they might not be in the same order as you think. For example here you refer to algorithm 4, in the presets this is algorithm 32.
They are all there and working correctly.
Due to the order of the operators in DX7, I had to reverse the numbers to make it work.
DX7 OP1 = MSF OP6
DX7 OP2 = MSF OP5
DX7 OP3 = MSF OP4
DX7 OP4 = MSF OP3
DX7 OP5 = MSF OP2
DX7 OP6 = MSF OP1

I also added there all the algorithms from FM8 and some complex new stuff.

Still I think the cpu overhead is not worth it , it triples the cpu usage just to have acces to these algoritms ( which require global feedback )

[jmg8]

Compare it to other DX7 or complex FM matrix capable synths. Is MSF higher CPU?
I have a computer that is by no means fast or modern (for testing and stuff) and the CPU here is super low with the feedback enabled.
Plus it's only a few of the many many algorithms that use the feedback, so I feel, it's not a big deal at all.

[gentleclockdivider]

Enabling 6 operators while global feedback is on( no oversampling) , and playing a chord increases the cpu load 300%
I call that pretty huge .
Considering that synths like dexed and fm 7/8 have absolutely no problem with that .
And we`re talking phasemodulation here , not the most cpu intensive algoritms by a long shot .

[jmg8]

Increase by 300%
So what is it to start with?
Checking the difference in CPU between having feedback on or off, then comparing it to other synths makes no sense. The other synths do not allow the feedback to be switched off.
So you should be looking at this situation in a positive light. MSF has a super low CPU mode, where you can disable feedback to save CPU.

Have you compared the total CPU usage for these synths?

[gentleclockdivider]

Of course I can compare with other synths that rely on phase modulation ,why should global feedback increase cpu usage ?
We're talking about simple phase modulation .
Maybe the developer can give some insight


With 300% I mean 3 times as much with no global feedback
A 6 voice chord with 6 operators , no global feedback =6%
With global feedback enabled I get 18 % cpu
Msf is pretty cpu sufficient in general , but this global feedback is not desired .

[jmg8]

Again, what you are looking at is not the correct way to compare to other synths.
Enable feedback then compare the CPU to other FM synths. How does the CPU compare?

Saying 300% or 3 times as much is not so useful.
For some it could be 1% without and 3% with, which would be extremely low CPU.

So please enable the feedback with 6 OPs and play your 6 note chord and see how MSF stands up against Dexed and FM8. Is it higher CPU?

[Niowiad]

To be fair he said 6% which becomes 18%

The thing is, I have a terrible 2012 AMD cpu and I'm not getting more than 4% cpu (128spls buffer in REAPER) with matrix feedback enabled at 4x upsampling for the FM module with 6 operators all doing something.
Keeping 1x upsampling under the same settings doesn't go above 2-3%.

[jmg8]

So it is really low CPU, same as I get here.

What I wanted to know is how it compares to other FM synths.

[Niowiad]

I removed Dexed time ago so I can't tell about it.
In my case Blue2, FM8 and Sytrus require lower cpu (at least 30% less it seems) on a similar task although it can be hard to exactly replicate the exact conditions and behaviour and upsampling on different synths.
Still not much of a problem with MSF if it's still on those acceptable values and especially as I'd like to upgrade to a much much better last-gen cpu rather soon.

[jmg8]

Cool, thanks for testing.
I don't know if FM8 has upsampling? Not sure, so I reckon to test against it with MSF in no upsampling mode. Also FM8 might disable the OPs or feedback system if not used, so it might be best to actually set values in the matrix to know for sure they are contributing to the CPU usage.

[MeldaProduction]

Sometimes I feel like going the extra mile (such as the dedicated optimization for nonfeedback situation) is pointless. Makes me think that it's better to create mediocre stuff... no features, no advanced stuff, people wouldn't complaint ...