HPC-vQPU: A Service-Export Architecture for Virtual QPUs on Batch-Scheduled HPC Systems
2026-05-29T08:52:29Z•fae6e2c7daa23b233b5214d448d92c6baaa73f4394d7ac8a05dd37eb27415941
AMTCARMCUDAGNNGPU-simulationHPCKafkaKnativeKubernetesMPIPIMTensor-CoresUPMEMZEROGNNbatch-schedulingdevice-snapshotsfault-tolerancemetadata-driven-executionquantum-simulationreplicationscheduler-isolationserverlesssilent-data-corruptionvirtual-QPUzero-copy-alltoall
What happened
This arXiv feed (cs.dc) collects recent papers on HPC and distributed-system techniques for accelerators, large-scale training, and resilient execution. Notable works include: HPC-vQPU, an outbound-only service-export architecture to host virtual QPUs on batch-scheduled supercomputers using atomic, calibration-aware device snapshots to keep scheduled simulator jobs hermetic; ZEROGNN, which moves metadata/control fully on-device to eliminate host-mediated GPU orchestration overhead in sampling-based GNN training; a replication-based scheme for detecting and correcting Silent Data Corruption (SD
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_dc
- Record identifier
- fae6e2c7daa23b233b5214d448d92c6baaa73f4394d7ac8a05dd37eb27415941
- Enrichment time
- 2026-05-29T08:52:29Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.