Ableton Forum

Load a Sampler with new filter models(OSR, MS2...), play multiple notes simultaneously,
CPU usage multiplied unexpectedly.

I got a response from Ableton Support as follows.

" I can confirm that this issue is inherent to all instruments that use the new filter models, Unfortunately I didn't get an answer from the people who programmed these devices. 
I'll put you on hold and come back to you as soon as I can tell you more.

I can't promise you any solution to this however, as this might be rather expensive to fix.

Sorry for the inconvenience.

All the best,

Tech Support "

What warp modes are you using? Release settings? There are a lot of additional factors that can add CPU. Do the filters work better if you are using unwarped material?

Want better sounding filter models? It doesn't come for free. Either freeze or use the original filters.

Yep, I think the problem is with the support staff's answer rather than the software. The basic filters are just that, quick and resource cheap.
the Cytomic filters are by necessity more resource intensive as they are based on circuit modelling to reproduce the non-linearities of the original vintage filters. That's more resource intensive.

I'm surprised the support staff didn't have that answer on an internal knowledgebase

doghouse wrote:Want better sounding filter models? It doesn't come for free. Either freeze or use the original filters.

new filter algorithms are efficient and use a fixed amount of CPU.

for example,

- Sampler 16 voices, Filter off : 2% CPU

- Sampler 16 voices, Filter on (Clean) : 5% CPU

- Sampler 16 voices, Filter on (SMP) : 40% CPU

- Sampler 16 voices, Filter off + Auto Filter (SMP) : 5% CPU

bakku wrote:

doghouse wrote:Want better sounding filter models? It doesn't come for free. Either freeze or use the original filters.

new filter algorithms are efficient and use a fixed amount of CPU.

for example,

- Sampler 16 voices, Filter off : 2% CPU

- Sampler 16 voices, Filter on (Clean) : 5% CPU

- Sampler 16 voices, Filter on (SMP) : 40% CPU :evil:

- Sampler 16 voices, Filter off + Auto Filter (SMP) : 5% CPU

Where does the CPU read come from?

What about modulation?

Stromkraft wrote:

bakku wrote:

doghouse wrote:Want better sounding filter models? It doesn't come for free. Either freeze or use the original filters.

new filter algorithms are efficient and use a fixed amount of CPU.

for example,

- Sampler 16 voices, Filter off : 2% CPU

- Sampler 16 voices, Filter on (Clean) : 5% CPU

- Sampler 16 voices, Filter on (SMP) : 40% CPU

- Sampler 16 voices, Filter off + Auto Filter (SMP) : 5% CPU

Where does the CPU read come from?

What about modulation?

no offense, It's easy to reproduce, you can check it yourself.

I'm using
Mac Pro 2012, OS 10.11.6, Live 9.7.2

bakku wrote:

Stromkraft wrote:
no offense, It's easy to reproduce, you can check it yourself.

I'm using
Mac Pro 2012, OS 10.11.6, Live 9.7.2

I don't have your machine. You are making no sense. I'm not disputing the new filters use CPU. But your data is pretty much pointless for comparison as you don't say if it's the Live load engine or Activity Monitor. Nor do you say how many cores you have or give a CPU speed.

But i get it. Leaving you there.

bakku wrote: new filter algorithms are efficient and use a fixed amount of CPU.

for example,

1 - Sampler 16 voices, Filter off : 2% CPU

2 - Sampler 16 voices, Filter on (Clean) : 5% CPU

3 - Sampler 16 voices, Filter on (SMP) : 40% CPU

4 - Sampler 16 voices, Filter off + Auto Filter (SMP) : 5% CPU

You mean you are holding 16 notes in this scenario?
I think points 1-3 are comparable, but not point 4.
Points 2 & 3 would mean 16 filters (one per voice), point 4 is only 1 filter (behind samplers output).

bakku wrote:

doghouse wrote:Want better sounding filter models? It doesn't come for free. Either freeze or use the original filters.

new filter algorithms are efficient and use a fixed amount of CPU.

for example,

- Sampler 16 voices, Filter off : 2% CPU

- Sampler 16 voices, Filter on (Clean) : 5% CPU

- Sampler 16 voices, Filter on (SMP) : 40% CPU

- Sampler 16 voices, Filter off + Auto Filter (SMP) : 5% CPU

I wrote all the filter DSP, so if you are confused about anything in the following please let me know.

Decent analog modelled filters require lots of computation, which translates directly to CPU usage. From your measurements we can calculate the total cpu per filter is roughly: (40-2)/16 = 2.4% cpu per filter (per cpu core), which is very reasonable for the quality on offer. In the case of having a single filter after all 16 voices are summed then again you have roughly the same cpu: 5-2= 3% cpu (per cpu core).

If you don't want to expend that much cpu then please use the clean version which has no drive, and cannot self oscillate, and so take much less cpu. Another option to lower cpu, as you have already shown, is to use a "string synth" type approach and use one auto filter after all the voices are summed down to a stereo signal. If you want the tone and drive of the analog modelled filters but still want them to take less % of your cpu then the only option left is to buy a faster computer, which is still much cheaper than purchasing a hardware sampler with analog filters.

I hope that helps!