Keynote

8:30

Communications, Collectives, and Telemetry

9:30

• Paper Type: Long
  103
  Optimizing Allreduce Operations for Modern Heterogeneous Architectures with Multiple Processes per GPU

Break

10:00

GPUs, Accelerators, and Runtime/Programming Models

10:30

• Paper Type: Long
  110
  Exploring C++ Standard Parallelism for GPU Programming in a Particle-In-Cell Application
• 11:00
  Paper Type: Long
  109
  A Statically Scheduled Vector Accelerator with Explicit Data Movement Control
• 11:30
  Paper Type: Long
  106
  Quantitative Characterization of Host-Initiated CUDA Memory Allocators: Performance, Fragmentation, and Concurrency Trade-offs

Lunch Break

12:00

Systems, Memory, and Runtime Adaptation

1:30

• Paper Type: Short
  107
  Adapting Workloads for CXL-Based Disaggregated-Memory Systems
• 1:50
  Paper Type: Long
  108
  Towards an Adaptive Runtime System for Cloud-Native HPC
• 2:20
  Paper Type: Long
  112
  HaSP: Harnessing Storage-side Parallelism for matrix multiplication
• 2:50
  Paper Type: Short
  104
  Incidence Constraints in Hypergraph Partitioning on GPU

Break

3:10

Scientific AI, Energy/Performance, and Partitioning

3:30

• Paper Type: Long
  105
  A Benchmark Suite for Evaluating Scientific AI Workloads on GPUs
• 4:00
  Paper Type: Long
  102
  Energy–Performance Trade-offs in Federated Learning with SmartNIC-Enabled Communication on HPC Systems
• 4:30
  Paper Type: Long
  111
  ET: Bridging the Gap on Energy Telemetry for Multi-GPU Communication Collectives

End of the workshop

5:00