What type of study is this?

This is a Quantitative Study study.

Real-time, Work-conserving GPU Scheduling for Concurrent DNN Inference

View Full Paper

Ask AI

Bookmark

Authors

This system improves latency and throughput for DNN inference tasks, suggesting enhanced GPU utilization and scheduling efficiency.

Reef achieves low-latency performance for real-time DNN inference tasks on commodity GPUs, significantly improving utilization.
Evaluation of Reef shows less than 5% overhead in latency for real-time tasks, while increasing throughput by up to 1.53× compared to sequential scheduling.
The reset-based preemption scheme in Reef enables microsecond-scale preemption for best-effort kernels, facilitating controlled concurrent execution.
Reef outperforms Triton in end-to-end latency for real-time tasks by 1.12× to 5.20×, while maintaining similar throughput levels.

Han et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark