Please permit me to think out loud here.

I have been quite involved in audio DSP for a long time now. Only under very rare circumstances would the GPU ever be used for audio fx processing, and only when purposefully designed to do so, for instance I've used CUDA to power convolution (and not just reverb) algorithms. For most functions, the GPU would give no benefit due to the type of mathematical operations involved and, given that I've specialized in real-time processing of inputs that are then output (going from the ADC through .dll plug-ins doing the processing via the CPU then back out through the DAC and thus out as sound to the loudspeakers) I am having trouble imagining what the problem could be.

The reason being that it's unlikely ESO uses real-time sound ii]generation[/i] (in other words, it's probably all samples and wavetables, not generative audio synthesis), although the reverberation and 3D positional algorithms (probably using head-related transfer functions) are certainly real-time sound processing, which can indeed tax the CPU.

There's nowhere near enough sampled audio at any given time to tax RAM, and since I have everything on an SSD it can't possibly be a problem of reading the samples off of the storage medium, because I get the FPS slideshow, too.

That being said, the "consumer" grade sound cards are not meant to handle much real-time processing, even if you paid $800 USD for some surround sound card it's not comparable to what audio professionals use because we need fast input conversion to digital and fast output converted to analog audio signals, so the type of ADC/DACs we use work better for video games, too.

If I weren't at work, I'd set my Firewire audio interface up and test this theory.

Post Script: someone just told me that entirely deleting User Settings and then doing a repair worked for him, and that's one to add to my list.

Sound off and restarting didn't fix my issue, nor anyone else I knows. I don't really know who came up with this but it doesn't seem to work at all.