Session

DescriptionGPUs dominate modern supercomputers and AI clusters, yet their scarcity and high energy consumption expose architectural limitations. This minisymposium explores non‑GPU accelerators that offload specialized workloads, reduce GPU pressure, and expand design options for scientific computing and machine learning systems. Topics include quantum accelerators integrated into HPC+QC workflows for applications such as computational chemistry and electronic structure, and data processing units (DPUs) that offload storage and networking to enable efficient, GPU‑aware data services. The program also highlights RISC‑V–based accelerators and full‑stack solutions from emerging vendors as cost‑efficient, open‑ISA alternatives to GPUs, supported by case studies demonstrating performance gains and energy savings. Cross‑cutting discussions will address hybrid scheduling, orchestration frameworks, and standardized APIs required to integrate quantum, DPU, and RISC‑V technologies into production HPC environments. By bringing together experts across quantum computing, data‑centric architectures, and open hardware ecosystems, the minisymposium outlines the evolving landscape of non‑GPU acceleration. Its goal is to identify where these technologies can complement, and in some domains potentially replace, traditional GPUs in future exascale‑class systems for science and machine learning.