Q: Do any of the CFD solvers shown here fully run on GPUs, i.e., both the matrix construction and the linear algebra solve?
A: Yes, Orchard is able to run on GPU as well. In this case what we have done is hold out one GPU per node for inferencing and using the other GPUs for running the CFD code .

Q: I’ve used Dask to scale AI related jobs successfully on Summit.  Are there plans to consider benchmarks to include that Dask?
A: We have benchmarked Dask on Andes before, the scaling start to break down at 256 nodes and above.

Q: What are the challenges you foresee in the Autonomous Workflow for Single Crystal Neutron diffraction experiment - if you had to replace the Edge Workstation with one containing an alternate (non Nvidia Accelerators)?
A: Our stream and inference stack is based on Redis, so no strong dependency on Nvidia stack.

Additional notes: Mark Coletti (ORNL): I’ve been able to run Dask on Summit with > 1000 nodes.
Feiyi Wang:	Good to hear, maybe you can share your experience and code