Sumu uses anywhere from 30-100+ percent cpu with relatively low voice allocation on my computer, super into it and would love to support it early on but how much more optimized is it going to get? That is the only thing really holding me back at the moment, all the other madrona vst's run relatively smoothly on my computer.
Hi, Sumu will get a lot more optimized. Even the next Early Access release that I'm working to get out right now should be noticeably better. If it's not usable for you, please check back later.
May i be so bold as to ask for an additional option? One which resizes the partial count. If handling, say, 16 or 32 partials, as alternative to 64, cuts CPU usage, then i'd be happy to work with that fidelity in resynthesis.
To digress, i'd also like to submit for consideration a Modulus Time output on the KEY module. This would have Time periodically reset, like a basic ramp LFO.
May i be so bold as to ask for an additional option? One which resizes the partial count.
Absolutely. But this is not going to happen soon or maybe at all. To make a long story short, lots of optimizations depend on the number of partials being fixed when the code is compiled. This would be a huge change. My goal is still significantly less CPU use with 64 partials, so hopefully that will help.
To digress, i'd also like to submit for consideration a Modulus Time output on the KEY module. This would have Time periodically reset, like a basic ramp LFO.
I'm planning to add more LFOs in general for 1.0. Reset on note-on would be a good option for these, getting you effectively the same thing. I'll put it on the list...
Can I dare to ask if there was any progress on this? 
Unfortunatelly I can't afford to buy a brand new computer for Sumu, but I'd really love Sumu. 
...I'm in similar point as OP. Even more so as I usually run at 96, so it's 2 voices tops for me.
Coming next is a bug fix update with no optimizations. The release after that will have CPU optimizations.
Would it be possible to use GPU to accelerate audio processing with Sumu? Google can find lot's of information how to do it. An example is the Shadertoy where a piece (3 min I think, WebGL used for implementation) of music can be generated using the shaders:
https://www.shadertoy.com/view/ldfSW2
(just click on the rewind button if there is no sound)
There are many papers on the subjects, it might be possible to use OpenCL for implementation (https://github.com/felidadae/dsp_opencl_gpu/tree/master is an old example, not sure if JUCE or other framework already supports GPU audio).
Here some other example from github (actual VST3 plugins):
https://github.com/GPUAudioTeam/AptivGPUAudio
They used some AMD specific API to get audio into GPU to overcome latency problems though...
GPU acceleration seems attractive but there are a lot of problems putting it into practice. I don't have the time to write a big article on it here but please do know that I've looked at the possibility.
More importantly I have a clear path ahead for optimizing Sumu on the CPU.