Apologies your thread structure is a bit hard to read but ill try to give some information.
You should be able to run High Fidelity with most graphics card and os that supports the OpenGL 4.1+. the RSP3 simply has a Broadcom VideoCore IV as its GPU, but generally only the CPU is used When hosting a server. Same would apply for the TX2.
Now generally speaking to render in minimum requirement High Fidelity is based on VR minimum requirements, currently is, which generally is accepted to be anything that can run rendering of medium-intensive software at 90 fps. However, you can use slower GPU if you do not need to reach that 90 fps target. Infact, it still runs on a Intel Iris HD and technically a HD4400.
When looking at this way, you could get the High Fidelity Interface client running on a TX2, but not at full 90fps.
They are also working on an Android version which, natively runs on an (ARMv7-8). Currently however you would have to build your own, as there is no ready made build available, and most RSPI experiments are just Server Hosting, not rendering.
Now unlike OpenSim or Second Life, none of the physics are done on the server side as of the moment. So
Most of the entities in question will be static from the perspective of the server, where as dynamic objects are entities updated by the clients connecting.
This is much cheaper to keep track of than have multiple dynamic objects and should be much cheaper on the ram from the server’s perspective.
The 32x32x32 km also is a theoretical limit. You technically could go beyond that but you will start having more noticable floating point issues when trying to align things. It would make more sense to split up areas of a domain if you have such a large area and have a very high entity density in that 32km cube. This is however what they had in the road map a few years back, but connecting multiple domains together is something that hasnt been discussed much since.
You are also most likely keep track of much less objects, as there is no scale limit to the size of models. Most of the ram will be used to Providing assets, doing short caches of voice comms among other.