It is important to note that the H100 provides no visualization capabilities. It cannot be used for visualization programs or as a render farm. Rather, the H100 excels at computational tasks involving CUDA, OpenACC, training, and inference. If a user needs to run a visualization program such as Paraview, it should be submitted to the post partition instead. When properly utilized, a single gpu partition node has the computational abilities of dozens of CPU-only Joule 3.0 systems. However, users must be mindful of the 80 GB memory limit of each H100 card and properly structure the jobs to fit within this memory limitation.
