I’m using the RC3 and RC3-384 and can’t create a good quality screenshot with progressive passes. It always save the first pass to the screenshot (even with --enable-streaming). For now I’m increasing the samples per pixel but that seems to be slower and take some times.
Also it looks like paraview render the same image about 3 times to take a screenshot, at least 2 times. The first before showing the screenshot options, the second to save it, and the third to refresh the screen after saving the file.
Another problem (not sure if it’s the expected behavior) is when I tried to use the pvserver and pvdataserver/pvrenderserver to offload this OptiX render to my server, it still uses the GPU on the client machine. Is it possible to use the OptiX on the server or it is limited to client?
Both my client and server have the VisRTX support and it shows when connecting to the server the gpu as recognized.
Please file one bug report for the screenshot not cooperating with --enable_streaming and another for the multiple renders per screenshot. Those seem fixable.
Regarding client server behavior, it can sort of work but there are a couple of caveats. First make sure your remote rendering threshold is sufficiently low for the server to try to render of course. Second, with and MPI parallel job where the data is split among the nodes there are rendering artifacts. This is because rays miss and hit the background as they go off the local memory space. Third, kind of a special case of the second really, is that the server side has an implicit black background which makes most things will look darker compared to when rendered on the client side. The background issue seems pretty fixable. The full distributed memory project is harder and an ongoing research project at TACC.