Hi Dom, yes Genome works with other captures frequencies and you are correct on oversampling, the neural network models (NAM, AIDA, Proteus) have to run at the own frequency and cannot be oversampled without altering their response.
Regarding oversampling and when you should use it: some blocks which can generate distortion will create aliasing. Distortion/overdrive pedals, amp sims but also compressors can generate aliasing. You may or may not hear the aliasing and you even may not find it to be a problem, but many people do want an option to get rid of it. It's s subtle effect.
My recommendation: if you are very sensitive to latency, favor latency over removing the aliasing. And use the high levels oversampling values for rendering.
Dean, Some DAWs like Reaper can actually take care of the oversampling and send for example a 768KHz signal instead of the project's 48 (oversampling x16). This would be equivalent to Genome at max OS, try that with Helix using only TSM in Genome (which would be closer to Helix amp sim in terms of modeling) and l will be interested in your results.
Sascha Frank, regarding CPU: you cannot really compare the CPU needed for electronic approximation amp modeling and neural network modeling. If you are aiming at the best audio quality, standard profiles in NAM are incredibly CPU hungry, it's not because of a bad implementation, it's just how the technology is at this point. If you look at the newest hardware units able to do a decent neural network modeling, they pack an incredible amount of power compared to say a Helix or a Kemper.
Look at the TSM amps for example, they only need a fraction of the CPU needed for NAM. Our new TSM-AI format also asks for quite a bit of CPU, since neural networks are used. But you can try several quality options in the TSM-AI (or use lighter captures in NAM called nano or feather) and instantly save on CPU.
Regarding Genome load up times on first launch I 100% agree with you guys and be assured that it's a priority for us to fix this.