v1o wrote:You know despite the advancements made by groundbreaking synths like the Legend (and also Monark, Repro-5 etc) there are still people not convinced that plugins can sound like real discrete analogue hardware.
I was recently reading posts by Andrew Simper on GS. He was saying there still isn’t enough computational power to do the job, too many corners are cut by current methods commonly used in plugins. He was talking about having VST plugins that work more like 3D raytrace renderers which strive for greater realism by taking hours (even days) to do the calculations to generate images. So he was talking about having your plugins that work overnight to render your audio in the highest possible quality.
While I have the highest respect for Andrew, I cannot fully agree with him here.
We have now been working on a couple of vintage synth emulations (with only The Legend released yet). Our problem has rarely been CPU usage, but the sheer amount of details that need to be modeled. Those details are not always computationally expensive, just very time consuming to implement. As an example, getting the transients right is very important in a vintage synth emulation. This requires a lot of work to make sure it works properly in all possible parameter settings (especially the "snappy" ones), but does not really require much CPU at the end of the day.
Of course there is stuff like the filter core or the VCA that can eat quite some CPU, but a lot of stuff that sounds good / is important in a vintage synth boils down to something trivial in terms of the algorithms required. The analogy with the photo-realism is ihmo better if you consider the time we spend on an emulation, rather than the computer. Analog synth modeling could be compared to creating a photo-realistic image by setting every pixel, a nightmare but entirely possible. If a synth emulation is not 100% there yet, some pixels are simply lacking