From the original SynthMaker forum (2004–2014), with its members’ names and dates.
Cyforce ·
Today i tinkter a little bit, and ask me, how much less cpu eat a 1024 wavetable instead of a 2048 wavetable, and mhmmm i get a suprising result...
(picture attached) - strange that 2048 eats sometimes less cpu than 1024, any reason for this?
MichaelBenjamin ·
i would guess the only difference is in memory usage, not in cycles - the analyser measurements seem to be not that accurate on newer systems.
infuzion ·
+ we'll need the OSM to test for ourselves...
Cyforce ·
Usually i mentioned, if the wavetable size is smaller, the cpu usage is smaller too, through the less points for calculating. And also for a 1024er wavetable, the root key area of a single cylce wave with this lenght is around F0 or something, so and if the normal played note range goes from c1-c6 a 1024er wavetable would match more than a resized one to 2048, on the logical way. Because if the source single cycle is recorded on C1 for example it must be only stretch by max. 10% for targeted 1024er size, but for 2048 it must be stretched by 210% - so more space for corruption.
But if 1024 & 2048 eating nearly the same cpu count, hm whats now better? Staying at 2048?
rl ·
Cyforce wrote: But if 1024 & 2048 eating nearly the same cpu count, hm whats now better? Staying at 2048?
If memory is not an issue I'd say go with the larger table size, since you gain accuracy and reduce aliasing. With larger tables you can also reduce a couple of CPU cycles by using a cheaper interpolation algorithm (e.g. linear instead of cubic).
BTW: from my experience the CPU cycle measurement is not accurate - always varies about 10 to 30 cycles on my computer.
trogluddite ·
Will probably be obvious once you tell me, but what is QOI, DOI, RLI on your readout? I guess 'I' is for interpolation, but I just can't get the rest (too long at work today maybe...)
If you are using the regular Wave Table/Wave Table Read - then I would guess it works like the stock saw/triangle osc's. Their tables contain 256 resampled versions of the waveform (more than enough to have one for every key on the keyboard) - so there's never a massive amount of 'stretching' going on, it's mostly done by switching between different table entries.
So I would agree with rl, use more entries for better accuracy/less interpolation at low frequencies where there are many more harmonics below the SR/2 frequency limit.
Cyforce ·
The QUI/DUI was only shortcuts...
Quad/4th Ordered Interpol Double/2th Ordered Interpol Linear ...
trogluddite ·
Ah, cheers, I thought it must be something like that.
Wrote something here? To have a post removed or your name hidden, write to [email protected].
SynthMaker