Hasty Briefsbeta

Bilingual

End-to-End Hardware-Driven Graph Preprocessing for Enhanced GNN Performance

2 days ago
  • #GNN
  • #Hardware Acceleration
  • #FPGA
  • AutoGNN is an FPGA-based accelerator designed to enhance GNN performance by addressing preprocessing bottlenecks.
  • It leverages FPGA's reconfigurability and specialized components like adder trees, UPEs, and SCRs for efficient graph preprocessing.
  • AutoGNN dynamically profiles graph inputs and reprograms itself to handle varying workloads optimally.
  • Implemented on a 7nm enterprise FPGA, AutoGNN achieves up to 9.0× speedup compared to conventional preprocessing systems and 2.1× over GPU-accelerated systems.