My anecdotal observation after a few months with the unlocked P-A GP license on a custom Ryzen 8-core I designed and built a year ago, is CPU cores vs performance is mainly down to the VSTs in use, and how. How’s that for a broad generalization?
You could certainly build some top-end beast on Intel or AMD with dozens of cores and still run up against available CPU running a physically modelled Atmos convolution reverb.
One thing that occurs to me is Waves Audio just the other day launched their new set of “real time” DSP servers. You might want to ferret out the machine specs on those—could make a good model to start from for your own design?
I think on any machine, going through all the considerations provided in Deskew’s kindly provided free e-book on optimizing any computer for latency and performance in digital audio applications is a key. These steps should yield better “performance” anyway (available CPU cores vs VSTs for example).
More specifically in my first couple months using GP on a custom build of my own (8-core Ryzen) I’ve found performance vs cores to boil down to VSTs in general and even patches in particular.
For instance I’ve found the P-A Oberheim has some patches that can just jam the CPU utilization right through 100. I’m not exactly sure why this happens yet; and I’m not even sure if the Oberheim VST itself is spiking that—and I haven’t ruled out something else reacting in my processing chain. The Knifonium will do the same for some patches, but the Oberheim seems to have more such patches. It’s probably related to the built-in effects in the stock patch I would guess.
Anyone want to chime in about optimizing the P-A VSTs?
Another consideration though — and maybe an important one — is that I believe the GP CPU number at the top R reports only CPU utilization related to audio processing (GP devs? Yes/no?). I don’t know the answer but it occurs to me if you’re selecting an Intel processor for this build, you might want to lean towards the ones with the most internal busses (and therefore affects your MOBO/chip set selection as well) because that might help efficiency in the case of GP. No doubt that’s an oversimplification on my part, but it may be a resource to consider.
I’ve seen Arturia Analog and KONTAKT players have settings about multi-core utilization. I haven’t tried to figure out optimization of those yet. I have yet to find any such preference in any of the P-A VSTs.
Another question that HAS to impact CPU utilization is sampling engine Vs synth VSTi performance. Anybody have observations or anecdotes about that?
Sorry if I’m randomly opening a half a dozen cans of worms here, but this seemed like the most appropriate place for it on the internet.
I am loving GP to be clear! I got two! And a huge part of GP is the support quality.