• Audio

  • How can I work out if my puredata will run well on Bela before purchasing?

Hi there, totally newbie here. I've built a groovebox in Pd which I was planning to create a piece of hardware for, using arduino and rPi. However, I've just discovered Bela and it seems to be more suitable for what i'm trying to do.
My only worry is that it's already a pretty big Pd patch and needs plenty of room to grow over time. I don't know whether Bela will have enough power to run it. The patch already uses about 35% of the CPU of my M1 Macbook air. The hardware will have around 60 analog inputs and 10 digital inputs, which I would use the multiplexer for. I don't know if that would increase the CPU usage on Bela?

Is there any way I can work this out before splashing the cash on a Bela?

Thank you for any guidance you're able to provide.

    damu_89 The patch already uses about 35% of the CPU of my M1 Macbook air.

    That's a whole lot of CPU. I'd say it's highly unlikely it will run on Bela. That's really unfortunate, because this sounds like a great project! This would probably be even more power than available on a Pi, really, unless you take advantage of the multiple cores there and try and split the load among them using the [pd~] object!

    On Bela there is a way to reduce the CPU usage of a Pd patch using the Heavy compiler. This supports only a subset of Pd objects (https://github.com/CarloCattano/hvcc/blob/master/docs/14.Supported_vanilla_objects.md) and can be even 10x as fast as regular vanilla Pd, depending on the patch. I am not sure that will bring the CPU usage into an acceptable range, but if you do not use externals or some fancier vanilla objects not included in the supported list, it may be worth a try.

      Thanks very much for your quick reply. Routing all the ins and outs with Pi and Arduino is quite a bit more complicated but it sounds like my next step should be checking it runs Ok on Pi.

      The heavy compiler sounds really helpful, thank you. I will have a look at that too.

      I've just looked at my Pd patch and pretty quickly cut some things to bring CPU usage down to 18% from 30%.

      Is there anyway to know a ball park figure to aim for whereby this patch would run on a Bela having been designed on M1 Macbook air? If I knew that then I could try to cut down the CPU load of the patch a bit more.

      It's a problem related to the relative speed in Ghz of the processors but I don't know a simple way of comparing what the processors are capable of. Is it the clock speed of the Bela processor I need to know?

      Thank youuuuu

      hmmm it's complicated to do a 1-to-1 mapping with a given ratio, because the differences in CPU speed will be highly application-dependent. Does your patch consume the same amount of CPU at all times? Can you make it so that it uses CPU from the moment it starts, without need for any user interaction? In that case, I could just try and run it on Bela and tell you where you stand with regard to CPU budget.

      That would be amazingly helpful if you wouldn't mind? The CPU usage is pretty variable but it tops out at 30% on my computer. I'll send you the cut down version. It should just work straight away. Could you please let me know when you've grabbed it? I'd prefer not to share the patch widely so i'll pull it down once you have it.

      Thank you so much. Even if it doesn't work well, this is super helpful. I really appreciate it.

      just send me an email to giulio at bela.io . I will try it out tomorrow.

      Current CPU usage on my Bela is about 150%. Not an orders of magnitude off, do you think you can do a bit better? Also, what external library is mux~ from?

      Thanks, that's great. Yeah, I think I can cut it down a bit more. mux~ is from the zexy library. I guess the synth voices weren't producing audio for you without it? I checked this last night and the CPU usage was the same regardless.

      Sounds like I need to get CPU usage on my laptop down to about 8-10% for this to work. So I think i'll go ahead and purchase a Bela.

      My other question is this. I've seen people use beaglebone black with the Bela cape. Could I do that to get a bit more CPU power or will it still be the same?

        Might it also help the cpu load if I compiled some parts of the patch into externals and loaded them that wayt?

          damu_89 I guess the synth voices weren't producing audio for you without it? I

          I don't know, I was not actually listening to the output (because it's above 100% CPU!). I was just doing the number crunching in a tight loop and measured that it took about 14.5 seconds to generate 10 seconds of audio, from which I inferred the ~ 150% CPU load. I tried again with zexy and the result doesn't change much.

          damu_89 My other question is this. I've seen people use beaglebone black with the Bela cape. Could I do that to get a bit more CPU power or will it still be the same?

          The Bela cape only works in combination with the BeagleBone Black(BBB) and it doesn't have any processing capabilities onboard. Before the end of the summer it should also work with a BeagleBone AI(BBAI), which is more powerful. Pd performance on BBAI at full CPU should be about 8x what you'd get on BBB. I am actually surprised that the Mac vs Bela CPU ratio is only 10 to 1 as this would seem to indicate ... I guess that's not too bad!

          damu_89 Might it also help the cpu load if I compiled some parts of the patch into externals and loaded them that wayt?

          That could potentially help if done properly, but also it could be a lot of work (it involves rewriting your code in C!) 🙂

                I'm sure there's loads of stuff I can do to make the patch more efficient. It's not a sods of Pd I've had to use too much before. BBAI sounds amazing. Does that mean that I could make a v2 of this groove box later in in the summer that's 8x as powerful?
                I could get some friends to help me write some of the patch in C. I think that's the direction this project will take eventually. Thanks so much for your help. I'm really excited about this technology.

                  damu_89 . Does that mean that I could make a v2 of this groove box later in in the summer that's 8x as powerful

                  well ... hopefully!

                  damu_89 I could get some friends to help me write some of the patch in C

                  or you could learn C++ yourself with our course https://www.youtube.com/playlist?list=PLCrgFeG6pwQmdbB6l3ehC8oBBZbatVoz3 !

                  If you are looking into performance optimisation but don't want to go the full C route, using the Heavy compiler for Pd patches could be the way to go (see links above) ...

                      This course looks fantastic. Exactly what i've been looking for. Thank you!

                      I'll definitely do this after I finish my current college course in a few weeks.