Collective Knowledge Aggregator proof-of-concept
Crowd results Raw CK browser Graphs Reports Datasets Models Home
Select CK-powered unified experimental workflow:
[ Participated users, platforms, OS, CPU, GPU, GPGPU, NN, NPU ] [ How to participate ] [ Motivation (PPT) (PDF) ] [ Papers 1 , 2 , 3] [ Android app ] [ Collective training set ] [ Unified AI ]
[ Community-driven AI R&D powered by CK ], [ CK-Caffe2 / CK-Caffe ], [ CK-TensorFlow ], [ Wikipedia, paper 1, Paper 2, YouTube CK intro ], [ CGO'17 test of time award for our interdisiplinary R&D ]
Type: DNN engine: Model:
Platform: CPU:
OS: GPGPU:

# Platform OS CPU GPGPU Type DNN engine Model Choices (env) FWBW
min time

(exp time)
stat. repetitions		 FW BW Per layer HW costs All usage costs (preparation, training, inference, errors, etc) Model size Model accuracy on ImageNet Model topology and parameters Power consumption (W)
min / max		 Acoustic noise (dB)
min / max		 Memory usage (MB) Bug detected? User Replay
1 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, alexnet Batch size=1

 9 ms.

(9 ± 0 ms.)
3 repetitions 		3 ± 0 ms. 6 ± 0 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 8 MB nvidia-caffe-crowdbench

[ See raw files ]

(3697284a99975f40)
2 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 bvlc, alexnet Batch size=1

 9 ms.

(10 ± 0 ms.)
3 repetitions 		2 ± 0 ms. 7 ± 0 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 8 MB nvidia-caffe-crowdbench

[ See raw files ]

(7584f9d25e4ada99)
3 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 10 ms.

(10 ± 0 ms.)
3 repetitions 		3 ± 0 ms. 7 ± 0 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB nvidia-caffe-crowdbench

[ See raw files ]

(4205bb9546ddf9ce)
4 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 22 ms.

(23 ± 2 ms.)
3 repetitions 		8 ± 0 ms. 15 ± 1 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB nvidia-caffe-crowdbench

[ See raw files ]

(988b91a0cfd09ce5)
5 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, alexnet Batch size=1

 23 ms.

(23 ± 0 ms.)
3 repetitions 		5 ± 0 ms. 18 ± 0 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 8 MB nvidia@home

[ See raw files ]

(4cf39158fcf5787c)
6 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 deepscale, squeezenet, 1.1 Batch size=1

 25 ms.

(25 ± 0 ms.)
3 repetitions 		6 ± 0 ms. 18 ± 0 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB nvidia@home

[ See raw files ]

(7442d8d089624d60)
7 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, alexnet Batch size=10

 27 ms.

(27 ± 0 ms.)
3 repetitions 		10 ± 0 ms. 16 ± 0 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 83 MB nvidia-caffe-crowdbench

[ See raw files ]

(a17a5c174085b2d3)
8 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 bvlc, googlenet Batch size=10

 43 ms.

(45 ± 3 ms.)
3 repetitions 		14 ± 0 ms. 30 ± 2 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 551 MB nvidia-caffe-crowdbench

[ See raw files ]

(28a0958b347b3291)
9 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 deepscale, squeezenet, 1.1 Batch size=10

 43 ms.

(47 ± 3 ms.)
3 repetitions 		16 ± 1 ms. 30 ± 2 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 323 MB nvidia@home

[ See raw files ]

(b0f365425a897d97)
10 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, alexnet Batch size=10

 49 ms.

(51 ± 3 ms.)
3 repetitions 		20 ± 0 ms. 32 ± 3 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 83 MB nvidia@home

[ See raw files ]

(50b2cef8499ebf06)
11 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, googlenet Batch size=1

 49 ms.

(49 ± 72 ms.)
3 repetitions 		14 ± 1 ms. 106 ± 73 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB nvidia@home

[ See raw files ]

(7089174ae382c6e3)
12 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 52 ms.

(55 ± 3 ms.)
3 repetitions 		31 ± 2 ms. 24 ± 2 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(a46eea1eb57bc3ef)
13 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,clblast)

Version trunk-880111e

 bvlc, googlenet Batch size=1

 54 ms.

(54 ± 658 ms.)
3 repetitions 		14 ± 1 ms. 697 ± 657 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB nvidia@home

[ See raw files ]

(867b72b90e0e957b)
14 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.0 Batch size=1

 74 ms.

(78 ± 6 ms.)
3 repetitions 		44 ± 2 ms. 36 ± 4 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101)  MB 0.576801 / 0.803903 deploy.prototxt - - 53 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(523a18200e54386a)
15 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 92 ms.
3 repetitions 		32 ± 4 ms. 69 ± 9 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB tegra_crowdtuning

[ See raw files ]

(131410692af0b1ee)
16 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, googlenet Batch size=10

 93 ms.

(98 ± 8 ms.)
3 repetitions 		30 ± 1 ms. 70 ± 7 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 551 MB nvidia@home

[ See raw files ]

(fb1d4838908eb5ac)
17 System manufacturer System Version (System Product Name) Windows 7 Intel(R) Core(TM) i7-4930K CPU @ 3.40GHz Ellesmere opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version 5a1422a93a842bd6

 deepscale, squeezenet, 1.1 Batch size=10

 98 ms.

(100 ± 1 ms.)
3 repetitions 		47 ± 0 ms. 49 ± 0 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 323 MB dhawt11@gmail.com

[ See raw files ]

(7d34662b8e625b52)
18 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 99 ms.
3 repetitions 		36 ± 6 ms. 68 ± 5 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB -

[ See raw files ]

(1ec382e4451ba6c0)
19 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=10

 104 ms.

(104 ± 0 ms.)
3 repetitions 		45 ± 0 ms. 58 ± 0 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 551 MB nvidia-caffe-crowdbench

[ See raw files ]

(9f4499ab2d749cbf)
20 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 deepscale, squeezenet, 1.0 Batch size=1

 106 ms.
3 repetitions 		37 ± 9 ms. 93 ± 18 ms. Stats per layer



  MB 0.576801 / 0.803903 deploy.prototxt - - 53 MB -

[ See raw files ]

(148f898e4808b2db)
21 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 109 ms.
3 repetitions 		71 ± 0 ms. 36 ± 0 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB tegra_crowdtuning

[ See raw files ]

(7870446f8bf48c4a)
22 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,clblast)

Version trunk-880111e

 bvlc, alexnet Batch size=50

 114 ms.

(115 ± 1 ms.)
3 repetitions 		45 ± 0 ms. 70 ± 1 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 416 MB nvidia@home

[ See raw files ]

(8789bcb037c4c298)
23 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,clblast)

Version trunk-880111e

 deepscale, squeezenet, 1.1 Batch size=50

 125 ms.

(129 ± 6 ms.)
3 repetitions 		63 ± 4 ms. 67 ± 2 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 1615 MB nvidia@home

[ See raw files ]

(f37952b93ae44c49)
24 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 deepscale, squeezenet, 1.1 Batch size=50

 125 ms.

(128 ± 4 ms.)
3 repetitions 		60 ± 0 ms. 69 ± 3 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 1615 MB nvidia@home

[ See raw files ]

(fc810e65c46b6f08)
25 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, alexnet Batch size=50

 147 ms.

(149 ± 1 ms.)
3 repetitions 		78 ± 1 ms. 71 ± 1 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 416 MB nvidia@home

[ See raw files ]

(0ce1bf99d64f4797)
26 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 resnet, resnet152 Batch size=1

 170 ms.

(172 ± 2 ms.)
3 repetitions 		31 ± 0 ms. 136 ± 2 ms. Stats per layer



 236 MB deploy.prototxt - - 506 MB nvidia-caffe-crowdbench

[ See raw files ]

(45179b0f820db414)
27 Shuttle Inc V20 (SP35) Ubuntu 14.04.5 LTS Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz GeForce GT 610 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 215 ms.

(215 ± 0 ms.)
3 repetitions 		89 ± 0 ms. 120 ± 0 ms. Stats per layer

 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB Yes 51d443519307ce92

[ See raw files ]

(89e30f1198ba64c3)
28 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, alexnet Batch size=1

 221 ms.
3 repetitions 		36 ± 14 ms. 268 ± 84 ms. Stats per layer



 244 MB 0.568279 / 0.799501 deploy.prototxt - - 8 MB -

[ See raw files ]

(346f48affa76e2a0)
29 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 bvlc, googlenet Batch size=100

 226 ms.

(227 ± 3 ms.)
3 repetitions 		90 ± 0 ms. 137 ± 3 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 5515 MB nvidia@home

[ See raw files ]

(82cc588cde84f75a)
30 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, alexnet Batch size=1

 236 ms.

(243 ± 9 ms.)
3 repetitions 		131 ± 8 ms. 111 ± 4 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 244 MB 0.568279 / 0.799501 deploy.prototxt - - 8 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(c003680ff534026a)
31 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 265 ms.
3 repetitions 		77 ± 16 ms. 223 ± 25 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB -

[ See raw files ]

(485b2e386bee26ac)
32 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,clblast)

Version trunk-880111e

 bvlc, googlenet Batch size=50

 273 ms.

(280 ± 11 ms.)
3 repetitions 		105 ± 3 ms. 175 ± 9 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 2757 MB nvidia@home

[ See raw files ]

(a1b5945ee29344ea)
33 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, googlenet Batch size=50

 273 ms.

(274 ± 2 ms.)
3 repetitions 		110 ± 1 ms. 164 ± 2 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 2757 MB nvidia@home

[ See raw files ]

(cebbf8e5a6b6a8bf)
34 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 resnet, resnet152 Batch size=1

 275 ms.

(278 ± 3 ms.)
3 repetitions 		84 ± 0 ms. 189 ± 3 ms. Stats per layer



 236 MB deploy.prototxt - - 506 MB nvidia@home

[ See raw files ]

(cde624da78a1b6c4)
35 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 297 ms.

(331 ± 44 ms.)
3 repetitions 		162 ± 25 ms. 177 ± 18 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(6bc6463983e1d7cb)
36 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 310 ms.
3 repetitions 		93 ± 14 ms. 226 ± 0 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB tegra_crowdtuning

[ See raw files ]

(3aa42d083d1f1dc3)
37 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz cpu BVLC Caffe framework (cpu)

Version trunk-90e8abc

 bvlc, googlenet Batch size=1

 324 ms.

(367 ± 81 ms.)
3 repetitions 		234 ± 52 ms. 166 ± 29 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 54 MB 0.689299 / 0.891441 deploy.prototxt 4.0 / 35.0 0 / 40 55 MB win-caffe

[ See raw files ]

(84b327181cecb404)
38 Hardkernel (Hardkernel odroid-xu3 board based on EXYNOS5422) Ubuntu 16.04.2 LTS ODROID-XU3 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 329 ms.

(368 ± 42 ms.)
3 repetitions 		233 ± 36 ms. 134 ± 20 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB my-odroid-xu3

[ See raw files ]

(06c46f6a91f56a87)
39 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,clblast)

Version trunk-880111e

 resnet, resnet152 Batch size=1

 337 ms.

(387 ± 38 ms.)
3 repetitions 		157 ± 15 ms. 212 ± 22 ms. Stats per layer



 236 MB deploy.prototxt - - 506 MB nvidia@home

[ See raw files ]

(efe4b70847bb7890)
40 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz Intel(R) HD Graphics 4400 opencl BVLC Caffe framework (opencl,viennacl)

Version master-880111e

 deepscale, squeezenet, 1.0 Batch size=1

 338 ms.

(345 ± 7 ms.)
3 repetitions 		92 ± 2 ms. 250 ± 5 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201)  MB 0.576801 / 0.803903 deploy.prototxt 4.0 / 35.0 0 / 40 53 MB ck@home

[ See raw files ]

(751a622e55953450)
41 Shuttle Inc V20 (SP35) Ubuntu 14.04.5 LTS Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz GeForce GT 610 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 bvlc, googlenet Batch size=1

 413 ms.

(413 ± 923 ms.)
3 repetitions 		163 ± 0 ms. 1168 ± 923 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB 51d443519307ce92

[ See raw files ]

(00d416c88caedea4)
42 Acer V1.06 (Aspire ES1-520) Windows 10 AMD E1-2500 APU with Radeon(TM) HD Graphics Kalindi opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e3

 deepscale, squeezenet, 1.1 Batch size=1

 417 ms.

(426 ± 11 ms.)
3 repetitions 		139 ± 8 ms. 277 ± 3 ms. Stats per layer



 5 MB 0.58388 / 0.810123 deploy.prototxt - - 32 MB 444fb83dc0f39419

[ See raw files ]

(2e28a94f8d9782a7)
43 Raspberry Pi (Raspberry Pi 3 Model B) Raspbian GNU/Linux 8 (jessie) BCM2709 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=1

 449 ms.
3 repetitions 		272 ± 1 ms. 183 ± 10 ms. Stats per layer



 50 EUR - (UK) (20151101) 5 MB 0.58388 / 0.810123 deploy.prototxt 1.2 / 4.5 0 / 0 32 MB raspberry pi volunteer

[ See raw files ]

(d7eb1a83c7409184)
44 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz Intel(R) HD Graphics 4400 opencl BVLC Caffe framework (opencl,viennacl)

Version master-880111e

 bvlc, alexnet Batch size=1

 463 ms.

(464 ± 2 ms.)
3 repetitions 		70 ± 0 ms. 393 ± 2 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 244 MB 0.568279 / 0.799501 deploy.prototxt 4.0 / 35.0 0 / 40 8 MB ck@home

[ See raw files ]

(b86d40a3849ac9a5)
45 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz cpu BVLC Caffe framework (cpu)

Version trunk-90e8abc

 bvlc, alexnet Batch size=1

 495 ms.

(514 ± 26 ms.)
3 repetitions 		236 ± 4 ms. 281 ± 27 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 244 MB 0.568279 / 0.799501 deploy.prototxt 4.0 / 35.0 0 / 40 8 MB win-caffe

[ See raw files ]

(6beaebb5a5674f4b)
46 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 resnet, resnet50 Batch size=1

 499 ms.

(573 ± 56 ms.)
3 repetitions 		181 ± 12 ms. 381 ± 53 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 98 MB 0.0 / 0.0 deploy.prototxt - - 231 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(aa5fa2525c74ee59)
47 LENOVO ThinkPad X240 (20ALCTO1WW) Ubuntu 17.04 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab1

 bvlc, googlenet Batch size=1

 573 ms.

(574 ± 2 ms.)
3 repetitions 		328 ± 3 ms. 243 ± 2 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 54 MB 0.689299 / 0.891441 deploy.prototxt 4.0 / 35.0 0 / 40 55 MB -

[ See raw files ]

(4f2d8d5cf2713849)
48 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz Intel(R) HD Graphics 4400 opencl BVLC Caffe framework (opencl,viennacl)

Version master-880111e

 bvlc, googlenet Batch size=1

 602 ms.

(602 ± 1 ms.)
3 repetitions 		161 ± 3 ms. 435 ± 3 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 54 MB 0.689299 / 0.891441 deploy.prototxt 4.0 / 35.0 0 / 40 55 MB ck@home

[ See raw files ]

(6506005d2f48cebd)
49 SAMSUNG SM-G930F Android 6.0.1 ARM Cortex-A53 ; Samsung Exynos M1 cpu BVLC Caffe framework (cpu)

Version trunk-c61d487

 deepscale, squeezenet, 1.1 Batch size=1

 621 ms.

(999 ± 480 ms.)
9 repetitions 		554 ± 163 ms. 544 ± 316 ms. Stats per layer



 700 EUR - Samsung Galaxy S7 Black 32GO (20160101) 5 MB 0.58388 / 0.810123 deploy.prototxt 0.7 / 7.2 - 32 MB Samsung S7

[ See raw files ]

(3828420194c39b5e)
50 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 657 ms.
3 repetitions 		379 ± 5 ms. 279 ± 1 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB tegra_crowdtuning

[ See raw files ]

(1604fc892ab33491)
51 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=10

 661 ms.
3 repetitions 		199 ± 3 ms. 464 ± 14 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 551 MB -

[ See raw files ]

(44a8fb8dfee79380)
52 jetson_tx1 (jetson_tx1) Ubuntu 16.04.2 LTS jetson_tx1 NVIDIA Tegra X1 cuda BVLC Caffe framework (cudnn)

Version trunk-eeebdab

 bvlc, googlenet Batch size=10

 677 ms.
3 repetitions 		201 ± 6 ms. 480 ± 17 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 551 MB tegra_crowdtuning

[ See raw files ]

(ad435acde9b45279)
53 Shuttle Inc V20 (SP35) Ubuntu 14.04.5 LTS Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, googlenet Batch size=1

 701 ms.

(754 ± 218 ms.)
3 repetitions 		495 ± 111 ms. 419 ± 107 ms. Stats per layer

 54 MB 0.689299 / 0.891441 deploy.prototxt - - 55 MB Yes 51d443519307ce92

[ See raw files ]

(4de58d9d78e9df11)
54 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 cuda BVLC Caffe framework (cuda)

Version trunk-eeebdab

 bvlc, googlenet Batch size=100

 744 ms.

(744 ± 0 ms.)
3 repetitions 		374 ± 0 ms. 369 ± 0 ms. Stats per layer



 54 MB 0.689299 / 0.891441 deploy.prototxt - - 5515 MB nvidia@home

[ See raw files ]

(a88b24a81bb5a995)
55 LENOVO ThinkPad X240 (20ALCTO1WW) Windows 10 Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz cpu BVLC Caffe framework (cpu)

Version trunk-90e8abc

 resnet, resnet50 Batch size=1

 781 ms.

(790 ± 8 ms.)
3 repetitions 		391 ± 16 ms. 391 ± 16 ms. Stats per layer



 1600 USD - Lenovo ThinkPad X240 (France) (20151201) 98 MB 0.0 / 0.0 deploy.prototxt 4.0 / 35.0 0 / 40 231 MB win-caffe

[ See raw files ]

(bb7de095b1ea4754)
56 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, alexnet Batch size=10

 835 ms.

(896 ± 50 ms.)
3 repetitions 		495 ± 5 ms. 390 ± 45 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 244 MB 0.568279 / 0.799501 deploy.prototxt - - 83 MB ck-dnn-as-a-self-optimizing-service-by-dividiti

[ See raw files ]

(5c9747438059e294)
57 Raspberry Pi (Raspberry Pi 3 Model B) Raspbian GNU/Linux 8 (jessie) BCM2709 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.0 Batch size=1

 860 ms.
3 repetitions 		506 ± 2 ms. 350 ± 3 ms. Stats per layer



 50 EUR - (UK) (20151101)  MB 0.576801 / 0.803903 deploy.prototxt 1.2 / 4.5 0 / 0 53 MB raspberry pi volunteer

[ See raw files ]

(c65aecfecd482588)
58 Raspberry Pi (Raspberry Pi 3 Model B) Raspbian GNU/Linux 8 (jessie) BCM2709 cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 deepscale, squeezenet, 1.1 Batch size=2

 883 ms.
3 repetitions 		541 ± 4 ms. 356 ± 19 ms. Stats per layer



 50 EUR - (UK) (20151101) 5 MB 0.58388 / 0.810123 deploy.prototxt 1.2 / 4.5 0 / 0 64 MB raspberry pi volunteer

[ See raw files ]

(15faa429b49a0aec)
59 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 bvlc, alexnet Batch size=10

 940 ms.

(951 ± 10 ms.)
3 repetitions 		513 ± 12 ms. 436 ± 3 ms. Stats per layer



 120 USD - cTuning Azure cost per month (depends on usage) (20160101) 244 MB 0.568279 / 0.799501 deploy.prototxt - - 83 MB -

[ See raw files ]

(1099e435bb397e85)
60 Hewlett-Packard (HP Z640 Workstation) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz GeForce GTX 1080 opencl BVLC Caffe framework (opencl,libdnn,viennacl)

Version master-880111e

 resnet, resnet152 Batch size=10

 1017 ms.

(1074 ± 49 ms.)
3 repetitions 		264 ± 9 ms. 800 ± 55 ms. Stats per layer



 236 MB deploy.prototxt - - 5063 MB nvidia@home

[ See raw files ]

(0a74c99ba5763e9f)
61 Microsoft Corporation 7.0 (Virtual Machine) Ubuntu 16.04.2 LTS Intel(R) Xeon(R) CPU E5-2673 v3 @ 2.40GHz cpu BVLC Caffe framework (cpu)

Version trunk-eeebdab

 resnet, resnet101 Batch size=1

 1033 ms.

(1049 ± 14 ms.)
3 repetitions 		301 ± 18 ms. 749 ± 15 ms. Stats per layer