Thanks for all your efforts on this, @MikeOliphant and @gianfranco ! I’m going to try this now… Do you have any recommendations on a model to try? Do I just grab the A2 models from Tone3000 and make sure to select Lite on the MOD?
1 Like
Nevermind - grabbed some A2 models from Tone 3000 and it sounds great! I know what I’m doing tonight! 
3 Likes
That would even be greater @gianfranco…if, without getting out of the MOD WebGUI, we could quickly search and try different models…I guess that would be a huge plus to save time on the cumbersome process of searching on Tone3000 for models, getting more or less triggered by the small marketing around the model, downloading it, and uploading it to the MOD and try it. I guess that would turn the first 5 steps into 2
Hi all,
Today I was practicing with the Dwarf using a pedalboard that includes the new NAM A2 plugin, which I compiled and uploaded to the device myself using the mod-plugin-builder environment.
Just to be clear, this is my own custom build of the plugin, but the source code is exactly the same as the original patch, with no modifications.
Everything works correctly on the first load. However, if I switch to another pedalboard and then return to the original one containing NAM, the second load produces highly distorted audio, almost like a ringing or modulation-type distortion.
I’m mentioning this because it may be worth testing whether the same behavior occurs with the officially built version of the plugin.
Yay, happy happy happy. Looks like the new version from the store works on my ModDuoX.
Ive tried the same capture across these types with 128 frames buffer:
- I’ve got ~20% load on A2 lite with (0.018 ESR).
- old A1 Nano was using about ~19% (0.0697 ESR)
- A1 Lite was using ~72% (0.0197 ESR)
If ESR is calculated the same way for A2 and A1, it should be a much better precision for the same CPU power, running a bit better quality than A1 Lite with a bit more CPU usage than A1 Nano.
That’s a very good improvement.
Also seems like I even can run the full model at ~84% on A2 full, without any headroom for any other plugins though. I hope A2 Full support would not be removed at least for MDX, there might be an edge case for that
What really confuses me is that TWO running NAMS in parallell, with different A2 lite models still shows as ~20% Cpu usage, almost same as single, adding like 2%, maybe. I do not understand how to intepret that. Is it real or cpu meter lies?
(Not completely a theoretical question, I would like to run two NAMs, if possible, one for melody side of my chapman stick, one for bass side, so it would be great to know what to expect.)
2 Likes
Yes. Due to having multiple cores, running in parallel does not necessary raises the total CPU usage.
2 Likes
Hi, I also got this noise effect sometimes. Even on the bootup. Changing the NAM model, solves the noise. I thought it was because it was a too hungry CPU pedalboard: 2 NAM instances + IR reverb + other effects… Even before NAM A2, but with a custom build plugin, I got this error sometimes.
Somehow I think it is related on how the threads are a assigned to each CPU core. The most stable pedalboard I got was using 2x Feather NAM A1 pedalboard. Each one individually showing 63% CPU, but together with more plugins always stable at 75% and never noise. Somehow in mod, some magic happens 
adding plugins, first the CPU shown in Mod-ui is going higher with jumps, but as soon as you add the magic number of plugins (some of them CV without Audio), CPU goes down and stabilizes…
As a curiosity, my same pedalboard with 2 NAM instances, runs smoother with 2 NAM A1 feather than with 2 NAM A1 nano or 2 NAM A2 lite. The last needs even the portal plugins !?
I hope not — I’d like to use that patch at a gig tomorrow night.
I also experienced this sometimes. But I haven’t used nam a lot too cpu intensive, instead nam A2 with quality <= 0.49 is great even if not all the captures sounds good.
yes sure. I also prefer the quality and quantity of A2 lite !!
@AndreaDelSignore I would recommend to assign the presets of the NAM plugin to some button/s. Then you can change the NAM model back and forward if the noise happens. This only happened to me after bootup or changing pedalboards (which I normally don’t do).
Edit: another supposition is sometimes the plugin loads the full A2 not the lite. So changing the NAM updates it to lite
2 Likes
for the build I published yesterday, I actually used MOD’s fork because I wanted to change the ttl files to make the Quality become an enumeration list.
From the original code I only I did two changes:
- make Quality be a two-member only enumeration list with Lite=0 and Full=1
- make the plugin default for Quality = 0
And, following @MikeOliphant suggestion, I enabled BUILD_INTERNAL_STATIC_LSTM in the .mk file
1 Like
Now THIS is thinking ahead 
3 Likes
Really good job guys. Its great. Just removed my Dwarf from my sell list. Will enjoy it for quite some time now.
2 Likes
It is also worth mentioning (for reference) that the CPU meter shows you an average calculation of the CPU usage
I guess that’s also the reasoning behind this “magic” @fer
Don’t think so, average does not explain why more plugins or even heavy plugins are more stable (fewer xruns) than less, and show less average cpu. I think is more related how the threads are assigned to cpu cores, probably related to thermal stuff…
1 Like
checking with top -H, I saw two jack threads with 95% cpu when there is that noise (I use 2 NAM plugins). Once I change the preset of the plugin, these threads went to 27% cpu and noise disappear. So looks sometimes the plugin is initialized with A2 Full, although the pedalboard, snapshot and plugin presets are defined to A2 Lite. A kludge would be to remove the A2 full from the NAM file… maybe I try.
@MikeOliphant do you think it can be something related to the plugin? Or how mod handles the parameters?
That might explain why I do not observe this, MDX just had enough room to absorb that load on Full ( at least with no other plugins than NAM). Probably.
The plugin defaults to quality scale 1 (full). I think @gianfranco is overriding that in the LV2 config. The override must be being lost (or not saved properly) in presets.
I would suggest building the plugin with the -DDEFAULT_QUALITY_SCALE="0" cmake option.
1 Like
Yes, It already has DDEFAULT_QUALITY_SCALE=“0.49”, but I see no diference
@fer Ok, so the default value is going to be overridden by the default value in the .ttl. I think you need to modify the “lv2:default” value for the “quality_scale” port in “resources/neural_amp_modeler.ttl.in”
2 Likes