Open SunMahatma opened 2 years ago
Achieving μs-scale Preemption for Concurrent GPU-accelerated DNN Inferences 该论文提出面向DNN推理任务的强实时高并发GPU调度技术,支持对非实时任务的百微秒级抢占,可将系统整体吞吐量提高了1.1~4.3倍。
Achieving μs-scale Preemption for Concurrent GPU-accelerated DNN Inferences 该论文提出面向DNN推理任务的强实时高并发GPU调度技术,支持对非实时任务的百微秒级抢占,可将系统整体吞吐量提高了1.1~4.3倍。