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.