Please don’t dump the output of a chatbot to the forum. It’s of no value because it hasn’t been verified by a human, and it takes up a ton of space and makes it harder for us humans to have actual conversations.
If a chatbot gives you useful suggestions, try them out and write up the results yourself. If any of us want to know what ChatGPT thinks we’re perfectly capable of asking it.
6 Likes
I think I figured this out. It seems to be a cache issue. I put in some cache invalidate/clean at key moments and the noises go away.
Also, zoomed in on a scope, the nature of the distortion is that sometimes a sample would be the same value of a very recent sample. That is, the falling edge of a triangle wave would have an occasional incorrect sample whose value was greater than where the sample should be, about equal to the value from one or two samples ago. And vice-versa for the rising edge. That rings of cache issues: one CPU grabs an old value from memory as the other CPU simultaneously processes the module that generated that value.
I should be able to fix it now, though it probably will add a small amount of CPU usage proportional to the number of cables in a patch. Though… I have some optimizations in mind for that as well… We’ll see.
2 Likes
Very interesting…I am glad you were able to localize it and hopefully will be able to fix it. Thanks as always for your help!
I fixed this in v2.0.0-rc3. If you have a chance to test it out, let me know:
1 Like
Did some initial tests…still hear a bit of distortion when using 32K but it clears up when using 48K (could be aliasing artifacts). No more interaction with graphics. Much better…thanks!