Cytomic "The Drop" Resonant Filter

[andy-cytomic]

Hi Andy
When i try the real time oversampling modes in Cascade mode, 4X upwards sounds like the filter is just turned off? doesnt seem right?
Im on Live 9.0.4, thanks

Can you tell me what the little cpu meter in the top right is reading? If it's above 100% then you probably won't hear much. I recommend sticking to x2 for realtime and then render at x8 if you want higher quality. The Cascade filter has to run 4 poles all the time which doubles the cpu usage of the Sallen Key if you only want 2 poles of filtering. This is due to the structure of the Cascade filter and cannot be changed. Also make sure you are running the Cascade not the Cascade SH if you want to save on CPU.

[foetus666]

Damn Andy the new filters are awesome. Just ran a little test to see if I could get the roland vibe I love out of it and it was just brilliant. All filtering is the Drop Cascade SH



https://soundcloud.com/andrewpolzin/cytomic-the-drop

[woodsdenis]

Great stuff Andy tx.

[themossinator]

Hi Andy
When i try the real time oversampling modes in Cascade mode, 4X upwards sounds like the filter is just turned off? doesnt seem right?
Im on Live 9.0.4, thanks

Can you tell me what the little cpu meter in the top right is reading? If it's above 100% then you probably won't hear much. I recommend sticking to x2 for realtime and then render at x8 if you want higher quality. The Cascade filter has to run 4 poles all the time which doubles the cpu usage of the Sallen Key if you only want 2 poles of filtering. This is due to the structure of the Cascade filter and cannot be changed. Also make sure you are running the Cascade not the Cascade SH if you want to save on CPU.

yeah its going ballistic - CPU probably cant handle the oversample?? thats ok thanks Andy

[David Else]

I just tried it for the first time, and wow!

My initial thoughts are:

- the loud pops on changing the high pass filters are a potential ear health hazard and should be prioritized to get rid of.

- bring on the Cascade SH high pass!

- the Cascade SH with a little drive is by far the most alive and organic of the filters. for example, if there is high resonance it seems that each time it plays the loop there is a subtle difference in the distortion/whatever it is doing that makes it more 'analog'. you can really hear the extra clipping, and it sounds fantastic. comparing it to what i thought was the best software saturation, this is by far the most alive and excellent saturation I have heard, no flatness unless you push it into insane levels of distortion. brilliant!

- not certain yet, but i think running it at native 96k sounds quite a bit better than oversampling it from 44.1 to 96k. i think i am hearing that slight 'smallness' that real-time oversampling can bring. not that this is a disaster, as thankfully you have total control of oversampling and can use it how you choose try switching the plugin to be 100% dry and turning it on and off at both 44.1k and 96k in your DAW where the drop oversampling settings make it always 96k... that is what i am talking about.

i think this could be the next level of accurate analog emulation, esp the Cascade SH. things are getting more alive! seems the deeper the modelling the more alive things get, i always assumed digital could never get there, but developments like this make me think digital will be equal or better than analog soon

[camsr]

In FL studio I get over 50% cpu usage on it's meter with the SH, in REAPER it's 7%. Strange!?

[sqigls]

Still having the problem where no sound will pass through the plug until i restart my OS.

EDIT: it seems to be happening far more regularly than before.

[Crackbaby]

Confirming that latest version works without A/B etc..

[HcDoom]

Sounds great. One question, is this plug-in multithreaded or...?

SH uses 22% with no oversampling here, with 2x oversampling goes to 44% (i7 2600 @ 3.8ghz). Than I check Win for system resources and only one core is under work, other 7 do almost nothing.

[andy-cytomic]

I just tried it for the first time, and wow!

My initial thoughts are:

Thanks! And thanks for taking the time to share your thoughts

- the loud pops on changing the high pass filters are a potential ear health hazard and should be prioritized to get rid of.

Yep, fair enough, I'll spend time on this and a few other little tidy ups for the next release before tackling new models.

- bring on the Cascade SH high pass!

I hope the high pass Cascade SH is possible to release. Currently circuit simulation software can't solve things properly when there are fast transition without adaptively up sampling to around 1 Mhz when the going gets tough, which is just not possible for me. I spent two solid weeks trying everything to workaround the issues and haven't given up, but this is a tough circuit to solve properly.

- the Cascade SH with a little drive is by far the most alive and organic of the filters. for example, if there is high resonance it seems that each time it plays the loop there is a subtle difference in the distortion/whatever it is doing that makes it more 'analog'. you can really hear the extra clipping, and it sounds fantastic. comparing it to what i thought was the best software saturation, this is by far the most alive and excellent saturation I have heard, no flatness unless you push it into insane levels of distortion. brilliant!

Glad you like it! The difference each time is most likely to be caused by the sin wave of self oscillation being at a different phase compared to the audio each time through, the incoming audio causes a non-linear amount of frequency modulation of the cutoff so at the end of it the sin wave will be at a pretty random phase compared to when it started the loop so then the whole loop will sound slightly different.

- not certain yet, but i think running it at native 96k sounds quite a bit better than oversampling it from 44.1 to 96k. i think i am hearing that slight 'smallness' that real-time oversampling can bring. not that this is a disaster, as thankfully you have total control of oversampling and can use it how you choose try switching the plugin to be 100% dry and turning it on and off at both 44.1k and 96k in your DAW where the drop oversampling settings make it always 96k... that is what i am talking about.

This is also true of recording audio at 88.2/96 compared to 44.1/48, you have loads more headroom before you have to low pass filter away the top end to keep things bandlimited. Most anti-aliasing filters are half band which cut -3 dB at "nyquist" or half the sample rate, which is 22.05 khz. For me this makes audio sound less alive and sound more boxed and less spatial, this is most evident on sources with high frequency content like cymbals or a small crowd clapping their hands. I figure our instincts are pretty finely in tune with working out which direction a snapping twig has come from and this kind of spatial location works by phase differencing very high frequencies, so even if you can't hear a steady sin tone at 18 khz I think your ears can still perceive auditory queues up this high and beyond. I really look forward to doing a bunch of tests in this area to confirm my suspicions. But also keep in mind that most people will listen to compressed audio so you're in a situation like back in the vinyl days, too much top end sounds ultra harsh.

i think this could be the next level of accurate analog emulation, esp the Cascade SH. things are getting more alive! seems the deeper the modelling the more alive things get, i always assumed digital could never get there, but developments like this make me think digital will be equal or better than analog soon

Thanks! There are definitely diminishing returns as more and more is modelled, but I don't mind using hundreds and thousands of transistors to model a single transistor It's only the beginning

[andy-cytomic]

Sounds great. One question, is this plug-in multithreaded or...?

SH uses 22% with no oversampling here, with 2x oversampling goes to 44% (i7 2600 @ 3.8ghz). Than I check Win for system resources and only one core is under work, other 7 do almost nothing.

I have already made Cascade SH algorithm run as much as possible in parallel using SSE1 and SSE2. It is impossible to split the processing of a single Cascade SH filter into multiple threads that run each on a different core since each and every sample depends on the previous state, as well as what the next output is going to be (this is called implicit). Since there is a feedback loop around the entire filter for resonance it can't be split into 2 two pole sections because all 4 poles depend on each other all the time and there is no way to tease this apart to be parallel. Inside the plugin I already use SSE2 to compute both the left and right signals in parallel at double precision, and I also use SSE1 to compute 4 values at once in parallel for certain parts of the algorithm that don't require double precision. Because the Cascade SH and the other full algorithms use lots of cpu I also provide the more cpu efficient versions of the algorithms (eg Cascade), and oversampling that can switch amounts when rendering.

In addition an i7 2600 computer has 4 cores ( http://ark.intel.com/products/52213 ), each core of which is split into 2 virtual hyper-threaded cores. If an algorithm efficiently pipelines instructions by issuing large numbers of parallel SSE instructions on memory that is already in the cache (which is what I'm doing in The Drop) then there is no performance gain from this virtual splitting of physical cores. So this means that you actually have 3 cores spare not 7, not sure if that is good or bad news for you!

[andy-cytomic]

In FL studio I get over 50% cpu usage on it's meter with the SH, in REAPER it's 7%. Strange!?

That sounds right. Please check your Reaper manual for how it reports cpu usage. Because 7*8 = 56% I am guessing you are running on a machine with 4 cores, each of which is split into 2 virtual hyper-threaded cores and reaper is reporting the total cpu usage summed from all cores then divides it by 8.

[andy-cytomic]

Still having the problem where no sound will pass through the plug until i restart my OS.

EDIT: it seems to be happening far more regularly than before.

Thanks for letting me know. If you can please provide me with any information as to how to reproduce this issue I would appreciate it. Your previous email to me you stated: "The problem doesn't seem to be reproducible" but if it is happening more often perhaps you can spot a pattern now. Any information is appreciated so I can track it down including which plugins are placed before The Drop on every channel that processes any of the audio that The Drop receives: www.cytomic.com/contact

[camsr]

In FL studio I get over 50% cpu usage on it's meter with the SH, in REAPER it's 7%. Strange!?

That sounds right. Please check your Reaper manual for how it reports cpu usage. Because 7*8 = 56% I am guessing you are running on a machine with 4 cores, each of which is split into 2 virtual hyper-threaded cores and reaper is reporting the total cpu usage summed from all cores then divides it by 8.

Actually you are almost right, 4 cores 4 threads. I tested it further with processor affinity, setting FL and REAPER to exclusive single cores. The windows cpu meter shows FL using about 55% of the core, while REAPER is using about 30%. Processor is Q9450 @ 3.2g, HT disabled.

[xiphiuz]

hi,

A little testdrive with this new beta where these parameters are automated with only an cascade sh 2 pole lowpassfilter:
- cuttoff
- resonance
- drive
- lfo-mod

furthermore: its just a saw wave as a sound source and some chords plus delay and reverb

https://soundcloud.com/xiphiuz/xiphiuz- ... 14-cascade