Split CNN Inference partitions layers between DPU and GPU with a GNN predictor, reporting up to 3.37x latency reduction over single-device runs and 96.27% GNN accuracy on tested models.
In2024 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
DPU or GPU for Accelerating Neural Networks Inference -- Why not both? Split CNN Inference
Split CNN Inference partitions layers between DPU and GPU with a GNN predictor, reporting up to 3.37x latency reduction over single-device runs and 96.27% GNN accuracy on tested models.