There are a couple of things you can do to reduce the memory footprint of a model in HOOPS Communicator, which is probably the reason you see this issue, in particular on lower-end graphics cards:
Use the streaming mode instead of scs loading. This mode will generally leave more data on the server and thereby reduces memory pressure on the client.
Adjust the streamCutoffScale during startup. Choosing a higher value will mean data only above a certain size will be requested from the server
Set a memoryLimit on startup. If the viewers memory exceeds this value, it will unload existing geometry before new geometry will be streamed.
use the onDemand streaming mode to be able to choose which part of the model will be streamed.
Use SSR (server-side rendering) to render the model completely server-side, in which case the client-side memory footprint will be massively reduced.
When converting the model, adjust the –tessellation_quality command line option for converter to a low value. This should reduce the models overall triangle count though it depends on the input model if changing this parameter will have any effect.
Set the –generate_LODs command line option, in which case converter will create multiple detail levels for each brep body in a model. You then have control over which LOD level to load during viewing.