GPU decoding only at 2% utilization

[Selur]

Yes, since if you use Vapoursynth, without enabling gpu hardware decoding or when your source can't be decoded by the gpu decoder chip the hardware decoding can't be used.
Using NVEnc© as encoder + enabling "NVEnc->Hardware->Only use encoder" will be the fastest. Then no Vapoursynth&Avisynth can be used, since everything is done through NVEnc. (having decoding, filtering and encoding directly on graphic board is the fastest)
When Using NVEnc© and using Synth, enabling "NVEnc->Misc->Misc->DirectSynthLoading" (and depending on the input 'MT'), might also speed things up. This way, the Vapoursynth/Avisynth scripts will be directly loaded by NVEnc©.

For decoding, when using NoSynth, enabling "Config->Input->Decoding->Use gpu for decoding" might speed things up, assuming the input can be decoded by the decoder chip of your gpu.

Using NVEnc as encoder and Vapoursynth, and your source can be decoded through the decoder chip, depending on the used SourceFilter, decoding can get faster if that source filter supports gpu decoding and that is enabled. "Filtering->Vapoursynth->Misc->Source", 'Use DGDecNV when available', 'Prefer BestSource' with 'gpu' and' configuring 'Libav hardware decoding mode' can help there.

Also, using gpu based filters might help.
When using Vapoursynth or Avisynth, depending on the filters, it's normal that the gpu isn't full utilized since:
a. cpu based filters might bottleneck the gpu filters
b. copying data between cpu and gpu will slow stuff down.

Cu Selur