… models, comparing the performance differences between training and inference, and … •
We demonstrate the need for a broader variety of deep learning workloads by looking at the …

… of these workloads, and more importantly, the training performance … , we characterize deep
learning training workloads from … of various workloads using different training architectures, to …

… learning model often includes a great number of parameters. In this paper, an efficient
deep learning … is proposed to predict the cloud workload for industry informatics. In the proposed …

… In this paper we present a detailed workload characterization … cluster utilization for DNN
training workloads on multi-tenant … , and (3) failures during training. Based on our experience …

Load imbalance pervasively exists in distributed deep learning training systems, either
caused by the inherent imbalance in learned tasks or by the system itself. Traditional …

… -arrival workloads. In detail, we analyze the historical workloads in the workload repository
… , and then adopt FFT to extract the periodic workload submission. To generate future-arrival …

… However, many multi-model workloads are unable to … optimize multi-model deep learning
workloads. HiveMind optimizes a “… tuning and multi-model inference workloads by up to 10× on …

… Second, many of these DNN training workloads are … of these workloads, we conduct a case
study on a Tesla V100 GPU [4] using six representative DNN training data center workloads […

… increase the diversity of job workloads to enable more general … to benefit the community
of deep learning systems. Based on the … Then we describe the DL workloads running in this …

… on allocating heterogeneous resources for DNN training workloads, we believe that Gavel
can be used for nonDNN workloads as well. Other workloads that are amenable to GPU …