The ELvis remote visualization cluster provides high-end GPU rendering capabilities for performing (interactive) visualization of large datasets. These datasets can be stored on SURFsara HPC systems (e.g. simulation output from a compute job on Cartesius) or can be uploaded to the cluster as well. The user runs his/her visualization application on the ELvis nodes at SURFsara instead of on their local desktop machine.
This has a number of advantages:
- The data to be visualized can be read directly from the storage at SURFsara, without having to be transferred to the user's local machine. Only a desktop image is transfered to the user, using the VNC remote desktop protocol.
- Multiple ELvis nodes can be used in parallel to provide increased visualization capabilities, in terms of GPU power and available memory. Note that this mode of usage requires the visualization application to support this kind of parallel rendering.
- A stack of visualization software is maintained by SURFsara, freeing you, the user, from having to spend time on this issue. In principal, user-written applications based on OpenGL are also supported.
Note that the RVS cluster is meant for providing interactive GPU-accelerated visualization, not for performing batch-oriented GPU-accelerated computations.
Using the remote visualization service more or less follows the following workflow:
- The user logs in on the ELvis login node and reserves one or more nodes
- After reservation is successful (usually instantaneous, if nodes are available) a VNC server will have been started on one of the nodes
- The user connects to the VNC server
- From the remote desktop the user launches the preferred visualization application and loads his/her data. The output of the hardware-accelerated rendering on the ELvis node is transparently captured, integrated into the VNC image stream, which is sent in compressed form through the VNC connection.