Determine GPU-sharing and resource saturation limits

Determine whether several independent Ceridwen fits can share a single GPU effectively and quantify how completely one Ceridwen fit saturates the GPU’s memory and computational resources.

Background

The reported comparison assumes that one Ceridwen fit runs on each GPU. Consequently, the paper does not establish whether multiple independent galaxy fits can be executed concurrently on one device or whether such concurrency would improve throughput. The extent to which a single fit exhausts GPU memory and compute capacity is likewise unresolved and is relevant to deploying Ceridwen for large survey samples.

References

The comparison further assumes one fit per GPU; whether several independent fits can share one device, and how far a single fit saturates its memory and compute, we have not measured.

— CERIDWEN: Fast and Flexible GPU-Accelerated Stellar Population Inference  (2609.30145 - Stoffers et al., 24 Sep 2026) in Section 5, Computational performance, Section 5.1

Determining which workloads can be co-located, and when resources should be repartitioned or shared, remains an open problem.

— Towards Efficient HPC Systems for Agents: Challenges and Opportunities  (2609.38723 - Zheng et al., 30 Sep 2026) in Section Compute, subsection “Opportunities: Trial-Aware, Hardware-Aware, and Event-Driven Scheduling”