Currently CAE workstations use 3D accelleration to edit 3D models for mechanical engineering. The models are the basis for highly sophisticated mathematical analysis that can only be handled by huge HPC servers.

We are facing 2 issues. On the one hand the tools are quite expensive and we want to consolidate them. On the other hand huge quantities of data have to travel over "low" bandwith (32 Mbps).

So the plan is to do the modelling work in the datacenter where the number crunchers are located. This should be done on terminal servers.

We have evaluated No Machine remote desktop capabilities and one is basically able to work with the presented display.

However the CPU is heavily utilized to move the 3D objects and there may be flickers if rotated or animated.

Now I ask myself if it would be possible to use 3D accelleration (e.g. OpenGL) for terminal sessions just like the application runs on a workstation.

Also I wonder if servers could be equipped with 3D graphics cards and if they would get utilized for remote sessions.

And what happens if there are several users on the machine. Is one graphics adapter able to display multiple desktops or do you need for example 10 GPUs for 10 concurrent users?

The application runs on UNIX, Linux and Windows and currently the users work on a WinXP workstation to model their systems, send the data to a remote UNIX server with lots of CPUs. There the users are able to login with X11 forwarding and they startup their program and load the transferred material. On the server they start the HPC transformation and send the result back to their regular workstation.

I think that it might be possible to forward the actual display to the clients with full 3D accelleration...

Does anyone have some experience with this?