DAWNBench

An End-to-End Deep Learning Benchmark and Competition

DAWNBench is a benchmark suite for end-to-end deep learning training and inference. Computation time and cost are critical resources in building deep models, yet many existing benchmarks focus solely on model accuracy. DAWNBench provides a reference set of common deep learning workloads for quantifying training time, training cost, inference latency, and inference cost across different optimization strategies, model architectures, software frameworks, clouds, and hardware.

The first iteration of DAWNBench is over, and the competition results and key takeaways have been finalized. However, we are still curious to see how well people can do on this benchmark and are now accepting rolling submissions. The original results before the April 20, 2018 deadline are archived for reference. For a more comprehensive benchmark, please consider submitting to the updated MLPerf benchmark.

Image Classification on ImageNet

Training Time

Objective: Time taken to train an image classification model to a top-5 validation accuracy of 93% or greater on ImageNet.

Rank Time to 93% Accuracy Model Hardware Framework
1

May 2019

0:02:43 ResNet-50

ModelArts Service of Huawei Cloud

source

16 nodes with InfiniBand (8*V100 with NVLink for each node) Moxing v1.13.0 + TensorFlow v1.13.1
2

Dec 2018

0:09:22 ResNet-50

ModelArts Service of Huawei Cloud

source

16 * 8 * Tesla-V100(ModelArts Service) Huawei Optimized MXNet
3

Sep 2018

0:18:06 ResNet-50

fast.ai/DIUx (Yaroslav Bulatov, Andrew Shaw, Jeremy Howard)

source

16 p3.16xlarge (AWS) PyTorch 0.4.1
4

Sep 2018

0:18:53 Resnet 50

Andrew Shaw, Yaroslav Bulatov, Jeremy Howard

source

64 * V100 (8 machines - AWS p3.16xlarge) ncluster / Pytorch 0.5.0a0+0e8088d
5

Aug 2019

0:23:11 Resnet 50

ZTE AI Platform

source

8 nodes with InfiniBand (8*P100 for each node) TensorFlow v1.12.0

Training Cost

Objective: Total cost of public cloud instances to train an image classification model to a top-5 validation accuracy of 93% or greater on ImageNet.

Rank Cost (USD) Model Hardware Framework
1

Sep 2018

$12.60 ResNet50

Google Cloud TPU

source

GCP n1-standard-2, Cloud TPU TensorFlow v1.11.0
2

Aug 2019

$19.00 Resnet 50

Chuan Li

source

Lambda GPU Cloud - 4x GTX 1080 Ti ncluster / Pytorch 1.0.0
3

Apr 2019

$20.89 ResNet50

Setu Chokshi (MS AI MVP | PropertyGuru)

source

Azure ND40s_v2 PyTorch 1.0
4

Sep 2018

$27.00 ResNet50

Google Cloud TPU

source

GCP n1-standard-2, Cloud TPU TensorFlow v1.11.0
5

Feb 2019

$42.66 Resnet 50 v1

GE Healthcare (Min Zhang)

source

8*V100 (single p3.16xlarge) tensorflow 1.11 + horovod

Inference Latency

Objective: Latency required to classify one ImageNet image using a model with a top-5 validation accuracy of 93% or greater.

Rank 1-example Latency (milliseconds) Model Hardware Framework
1

Nov 2019

0.4945 ResNet26

ModelArts Service of Huawei Cloud

source

Huawei Cloud [pi2.2xlarge.4] ModelArts-AIBOX + TensorRT
2

Oct 2019

0.5792 ResNet26d

Iluvatar CoreX, X Lab, P.S.R team: Gang Xu, Yu Song, Tao Yang, Fanwu Han

source

1 T4 / 128 GB / 40 CPU TensorRT 6.0.1.4
3

Oct 2019

0.6373 ResNet26d

Perseus AI Cloud Acceleration team in Alibaba Cloud

source

Alibaba Cloud [ecs.gn6i-c8g1.2xlarge] Pytorch+PerseusInference
4

Jul 2019

0.8200 ResNet50

PingAn GammaLab & PingAn Cloud team

source

PingAn Cloud {1 T4 / 128 GB / Xeon(G) Gold 6130} Caffe 1.0
5

Jun 2019

0.9695 ResNet50

InferenceX Team of Didi Cloud

source

Didi Cloud [1 T4 / 16 GB / 8 vCPU] ifx

Inference Cost

Objective: Average cost on public cloud instances to classify 10,000 validation images from ImageNet using of an image classification model with a top-5 validation accuracy of 93% or greater.

Rank Cost (USD) Model Framework Hardware
1

Oct 2019

$0.00 ResNet26d

Perseus AI Cloud Acceleration team in Alibaba Cloud

source

Pytorch+PerseusInference Alibaba Cloud [ecs.gn6i-c8g1.2xlarge]
2

Jun 2019

$0.00 ResNet50

InferenceX Team of Didi Cloud

source

ifx Didi Cloud [1 P4 / 16 GB / 8 vCPU]
3

May 2018

$0.01 ResNet50

Perseus AI Cloud Acceleration team in Alibaba Cloud

source

TensorFlow 1.12.2 Alibaba Cloud [ecs.gn5i-c8g1.2xlarge]
4

Dec 2018

$0.02 ResNet50

Perseus AI Cloud Acceleration team in Alibaba Cloud

source

TensorFlow 1.10.0 Alibaba Cloud [ecs.gn5i-c8g1.2xlarge]
5

Apr 2018

$0.02 ResNet50

Intel(R) Corporation

source

Intel(R) Optimized Caffe Amazon EC2 [c5.2xlarge]

Image Classification on CIFAR10

Training Time

Objective: Time taken to train an image classification model to a test accuracy of 94% or greater on CIFAR10.

Rank Time to 94% Accuracy Model Framework Hardware
1

Oct 2019

0:00:28 Kakao Brain Custom ResNet9

clint@KakaoBrain

source

PyTorch 1.1.0 Tesla V100 * 4 GPU / 488 GB / 56 CPU (Kakao Brain BrainCloud)
2

May 2019

0:00:45 BaiduNet9P

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla 8*V100-16GB/448 GB/96 CPU
3

Oct 2019

0:00:58 Kakao Brain Custom ResNet9

clint@KakaoBrain

source

PyTorch 1.1.0 Tesla V100 * 1 GPU / 488 GB / 56 CPU (Kakao Brain BrainCloud)
4

May 2019

0:01:12 BaiduNet9

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla V100*1-16GB/56 GB/12 CPU
5

Apr 2019

0:01:14 Custom ResNet 9

Ajay Uppili Arasanipalai

source

PowerAI 1.6.0 + PyTorch 1.0.1 IBM AC922 + Nvidia Tesla V100 (Nimbix np9g1)

Training Cost

Objective: Total cost for public cloud instances to train an image classification model to a test accuracy of 94% or greater on CIFAR10.

Rank Cost (USD) Model Framework Hardware
1

May 2019

$0.02 BaiduNet9

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla V100*1-16GB/56 GB/12 CPU
2

Aug 2019

$0.04 BaiduNet9

Chuan Li

source

fastai / Pytorch 1.0.0 Lambda GPU Cloud - 4x GTX 1080 Ti
3

Nov 2018

$0.06 Custom ResNet 9

David Page, myrtle.ai

source

pytorch 0.4.0 V100 (AWS p3.2xlarge)
4

May 2019

$0.11 BaiduNet9P

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla 8*V100-16GB/448 GB/96 CPU
5

Apr 2018

$0.26 Custom Wide Resnet

fast.ai + students team: Jeremy Howard, Andrew Shaw, Brett Koonce, Sylvain Gugger

source

fastai / pytorch Paperspace Volta (V100)

Inference Latency

Objective: Latency required to classify one CIFAR10 image using a model with a test accuracy of 94% or greater.

Rank 1-example Latency (milliseconds) Model Framework Hardware
1

Nov 2019

0.1345 ResNet8

ModelArts Service of Huawei Cloud

source

ModelArts-AIBOX + TensorRT Huawei Cloud [pi2.2xlarge.4]
2

Apr 2019

0.6830 BaiduNet8 using PyTorch JIT in C++

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Jiazhuo Wang, Haofeng Kou, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla V100*1/60 GB/12 CPU
3

Nov 2018

0.8280 Custom ResNet 9 using PyTorch JIT in C++

Laurent Mazare

source

PyTorch v1.0.0.dev20181116 1 P100 / 128 GB / 16 CPU
4

Oct 2019

0.8570 Kakao Brain Custom ResNet9 using PyTorch JIT in python

clint@KakaoBrain

source

PyTorch 1.1.0 Tesla V100 * 1 GPU / 488 GB / 56 CPU (Kakao Brain BrainCloud)
5

Oct 2017

9.7843 ResNet 56

Stanford DAWN

source

PyTorch v0.1.12 1 K80 / 61 GB / 4 CPU (Amazon EC2 [p2.xlarge])

Inference Cost

Objective: Average cost on public cloud instances to classify 10,000 test images from CIFAR10 using an image classification model with a test accuracy of 94% or greater.

Rank Cost (USD) Model Framework Hardware
1

Apr 2019

$0.00 BaiduNet8 using PyTorch JIT in C++

Baidu USA GAIT LEOPARD team: Baopu Li, Zhiyu Cheng, Jiazhuo Wang, Haofeng Kou, Yingze Bao

source

PyTorch v1.0.1 and PaddlePaddle Baidu Cloud Tesla V100*1/60 GB/12 CPU
2

Oct 2017

$0.02 ResNet 56

Stanford DAWN

source

PyTorch v0.1.12 1 K80 / 61 GB / 4 CPU (Amazon EC2 [p2.xlarge])
3

Oct 2017

$0.04 ResNet 164 (without bottleneck)

Stanford DAWN

source

TensorFlow v1.2 60 GB / 16 CPU (Google Cloud [n1-standard-16])
4

Oct 2017

$0.05 ResNet 164 (with bottleneck)

Stanford DAWN

source

TensorFlow v1.2 60 GB / 16 CPU (Google Cloud [n1-standard-16])
5

Oct 2017

$0.07 ResNet 164 (without bottleneck)

Stanford DAWN

source

PyTorch v0.1.12 1 K80 / 61 GB / 4 CPU (Amazon EC2 [p2.xlarge])

Question Answering on SQuAD

Training Time

Objective: Time taken to train a question answering model to a F1 score of 0.75 or greater on the SQuAD development dataset.

Rank Time to 0.75 F1 Model Framework Hardware
1

Mar 2019

0:18:46 FastFusionNet

Wu et al. (Cornell, SayMosaic, Google)

source

Pytorch v0.3.1 1 NVidia GTX-1080 Ti
2

Dec 2018

0:27:07 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 1.0.0 1 NVidia 2080 RTX (dev box)
3

Apr 2018

0:45:56 QANet

Google

source

TensorFlow v1.8 1 TPUv2
4

Dec 2018

0:50:21 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 1.0.0 1 T4 / GCP
5

Dec 2018

0:56:43 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 1.0.0 1 P4 / GCP

Training Cost

Objective: Total cost for public cloud instances to train a question answering model to a F1 score of 0.75 or greater on the SQuAD development dataset.

Rank Cost (USD) Model Framework Hardware
1

Dec 2018

$0.57 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 1.0.0 1 P4 / GCP
2

Dec 2018

$0.76 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 1.0.0 1 T4 / GCP
3

Sep 2018

$1.23 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 0.4.1 1 K80 / AWS p2.xlarge
4

Sep 2018

$3.09 DrQA

Runqi Yang, Facebook ParlAI, Brett Koonce

source

Pytorch 0.4.1 1 V100 / AWS p3.2xlarge
5

Oct 2017

$5.78 BiDAF

Stanford DAWN

source

TensorFlow v1.2 60 GB / 16 CPU (Google Cloud [n1-standard-16])

Inference Latency

Objective: Latency required to answer one SQuAD question using a model with a F1 score of at least 0.75 on the development dataset.

Rank 1-example Latency (milliseconds) Model Framework Hardware
1

Jul 2019

7.5790 PA-Occam-Bert

Ping An Technology Occam Platform

source

Tensorflow 1.13.0 1 NVidia Tesla V100
2

Feb 2019

7.9000 FastFusionNet

Wu et al. (Cornell, SayMosaic, Google)

source

Pytorch v0.3.1 1 NVidia GTX-1080 Ti
3

Oct 2017

100.0000 BiDAF

Stanford DAWN

source

TensorFlow v1.2 60 GB / 16 CPU (Google Cloud [n1-standard-16])
4

Oct 2017

590.0000 BiDAF

Stanford DAWN

source

TensorFlow v1.2 1 K80 / 30 GB / 8 CPU (Google Cloud)
5

Oct 2017

638.1000 BiDAF

Stanford DAWN

source

TensorFlow v1.2 1 P100 / 512 GB / 56 CPU (DAWN Internal Cluster)

Inference Cost

Objective: Average cost on public cloud instances to answer 10,000 questions from the SQuAD development dataset using a question answering model to a dev F1 score of 0.75% or greater.

Rank Cost (USD) Model Framework Hardware
1

Oct 2017

$0.15 BiDAF

Stanford DAWN

source

TensorFlow v1.2 60 GB / 16 CPU (Google Cloud [n1-standard-16])
2

Oct 2017

$1.58 BiDAF

Stanford DAWN

source

TensorFlow v1.2 1 K80 / 30 GB / 8 CPU (Google Cloud)
3

Oct 2017

$1.76 BiDAF

Stanford DAWN

source

TensorFlow v1.2 1 K80 / 61 GB / 4 CPU (Amazon EC2 [p2.xlarge])

Join Us

DAWNBench is part of a larger community conversation about the future of machine learning infrastructure. Sound off on the DAWNBench google group.

Disclosure: The Stanford DAWN research project is a five-year industrial affiliates program at Stanford University and is financially supported in part by founding members including Intel, Microsoft, NEC, Teradata, VMWare, and Google. For more information, including information regarding Stanford’s policies on openness in research and policies affecting industrial affiliates program membership, please see DAWN's membership page.