fmr wrote:Korg Odyssey is NOT a good example. Everybody agrees that the code needs to be optimized and that it should be lighter than it is.
Everybody? How did they reach their conclusion? Did they see the source code? Are they well versed in the latest coding techniques? Did they take latency and sample rate into account, host used, whether they're using USB or PCI(e) audio, and did they use features like ASIO guard? Most of all, have they actually made polyphony count comparisons with other software of equivalent quality? EG Diva, Roland Cloud VA polys etc..
Odyssey itself is single core per instance. It also uses polyphonic unison on presets. This means it can use a lot of polyphony without people understanding that they're doing it. If a synth has poly unison, and high CPU complaints, my first thought is whether the user knows this
My remarks were based on worst case examples. More typically Odyssey does 16 notes at about ~65% CPU on a first gen i7 (EG bump the first preset up to 16 note unison.) Now try a preset, like 47 Playcode, and you should see the usage rise up higher.
I got was around 45% CPU (and this was with way more than 12 notes - probably double than that, or even more, if you consider the long releases).
Well, it wouldn't have been since it has a 16 note polyphony voice limit
(Unless they changed that recently?)
And, as said, CPU usage varies by presets. I picked 47 as an example, but iirc there's likely heavier ones. Many other assumptions there too (EG sub 4ms output latency, ASIO guard off etc etc.)
Overall, I don't think Odyssey compares too badly to other synths on its level. EG I think it's likely you'll see higher per note usage from Diva (certainly with Divine, probably with great.. though its usage depends on the resonance setting used.. )
BTW The Odyssey iOS version has an 8 voice limit per instance, at least on the devices I have, though I've never tested how many instances I could use.. (Can't remember what hosting limits Gadget has)
But you may have lot's of other examples if you have an iPad.
It'd depend pretty heavily on which iPad. Ideally you'd want to compare to an iPad Pro, and there's likely to be a new version of that out in September, with a new generation of CPU. There will then be another generation of ARM CPU's after that, before whatever Apple propose to put out in 2020. At least you can somewhat predict this, due to the iPhone release cycles. So even those results, from a September 2018 iPad Pro, will still be two generations off whatever they're proposing to use on an ARM laptop, assuming the 2020 date even holds up. This also assumes they go with the same design and thermal contraints of the iPad. They may well use something which goes beyond that, or is customized in some other way. BTW the fact that they're including Intel, in the iOS app stuff, rather than skipping straight to ARM, also tells you they're hedging their future bets to some extent..
But, it'd be interesting to see the results people get, if there were a thread which made direct comparisons.. and maybe a good time to do it if there's new iPad Pros in September(ish). Bit off topic for this one though, at least if/until Roland Cloud arrives on iOS