A quick thing you could try:

• create a box (with box geometry), that represents the subset that you want to view.
• Place the camera inside the box
• Call camera.show_object(box) to setup the camera for this region.
• Make sure that the box is rendered first by doing box.material.render_queue = 1001 (essentially marking it as background).

This should at least constrain the interaction to a subregion of interest. With a bit of luck, the performance will also be better, because most of the mesh is hidden out of view, so the fragment shader will not have to run for most triangles.

I agree with Korijn that extracting a portion of the mesh is out of scope for PyGfx. Meshes can be tricky to work with, but its also not rocket science; a mesh is just vertices (i.e. an array of positions) and faces (each face is 3 indices into the array of vertices, which together form a triangle). So you could do:

• Build a reversed face map, that maps vertex indices to face indices.
• Determine what vertices are inside the region of interests, based on their position.
• Use the reversed map to determine what faces are in the selection.
• Create a new mesh using these faces.

Multiple features come together to support this kind of problem (real time rendering of huge geometry).

• Camera/view frustum culling Frustrum culling #363
• Occlusion culling
• Out of core rendering
• LOD rendering

Just dropping this for now so you have some leads to read up on.

Regarding mesh subset selection functionality: that's not really a rendering problem! So I would say it is out of scope for pygfx (the interactive flow in the UI). Rendering only part of a mesh is possible already by filtering the indices you are rendering, and/or the draw range.