Collective Knowledge Aggregator proof-of-concept
Crowd results Raw CK browser Graphs Reports Datasets Models Home

This page is outdated! New version is available here.


Share
50b2cef8499ebf06
Date: 2017-04-24 13:38:09.686863
License:
Module: experiment.bench.dnn
Repo: upload
CID (DOI-like distributed ID): 1eb2f50d4620903e:50b2cef8499ebf06

DNN engine and model evaluation statistics per layer (crowd-tuning)


DNN engine name: BVLC Caffe framework (opencl,libdnn,viennacl)
DNN engine version: master-880111e
DNN engine type: opencl
DNN engine dependencies: cmake=3.5.1
compiler=6.2.0
lib-blas=0.2.19-85636ff
lib-boost=1.58.0
lib-gflags=2.2.0
lib-glog=development-e56ed55
lib-hdf5=1.8.16
lib-lmdb=0.0.0
lib-opencl=8.0.61
lib-opencv=2.4.9
lib-protobuf-host=3.1.0
lib-viennacl=master-5494bef
DNN model name: Caffe model (net and weights) (bvlc, alexnet)
DNN model version: trunk
Batch size: 10
Platform: Hewlett-Packard (HP Z640 Workstation)
OS: Ubuntu 16.04.2 LTS
CPU: Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz
GPU: NVIDIA Corporation Device 1b80 (rev a1)

Name Direction Min time (ms.): Expected time (ms.): Max time (ms.): Repetitions:
16: fc6 backward 8.4 8.4 8.5 3
16: fc6 forward 5.9 5.9 6.0 3
19: fc7 backward 3.5 3.6 3.6 3
22: fc8 backward 3.1 3.3 3.9 3
05: conv2 backward 2.8 2.9 2.9 3
19: fc7 forward 2.5 2.5 2.5 3
09: conv3 backward 2.3 2.3 2.4 3
05: conv2 forward 2.1 2.1 2.1 3
01: conv1 backward 2.0 2.1 2.2 3
11: conv4 backward 1.9 1.9 2.0 3
09: conv3 forward 1.8 1.8 1.8 3
22: fc8 forward 1.6 1.6 1.6 3
13: conv5 backward 1.4 1.5 1.6 3
11: conv4 forward 1.4 1.4 1.4 3
01: conv1 forward 1.3 1.3 1.3 3
13: conv5 forward 0.9 0.9 0.9 3
03: norm1 backward 0.7 0.8 0.8 3
07: norm2 backward 0.7 0.7 0.7 3
15: pool5 backward 0.5 0.6 0.6 3
04: pool1 backward 0.5 0.5 5.1 3
08: pool2 backward 0.4 0.4 0.5 3
03: norm1 forward 0.2 0.3 0.3 3
18: drop6 forward 0.2 0.3 0.3 3
21: drop7 forward 0.2 0.3 0.3 3
07: norm2 forward 0.2 0.2 0.2 3
23: prob forward 0.2 0.2 0.2 3
23: prob backward 0.1 0.1 0.2 3
02: relu1 forward 0.1 0.1 0.1 3
04: pool1 forward 0.1 0.1 0.1 3
06: relu2 forward 0 0 0 3
08: pool2 forward 0 0 0 3
10: relu3 forward 0 0 0 3
12: relu4 forward 0 0 0 3
15: pool5 forward 0 0 0 3
14: relu5 forward 0 0 0 3
17: relu6 forward 0 0 0 3
20: relu7 forward 0 0 0 3
00: input forward 0 0 0 3
10: relu3 backward 0 0 0 3
21: drop7 backward 0 0 0 3
14: relu5 backward 0 0 0 3
20: relu7 backward 0 0 0 3
02: relu1 backward 0 0 0 3
18: drop6 backward 0 0 0 3
06: relu2 backward 0 0 0 3
12: relu4 backward 0 0 0 3
00: input backward 0 0 0 3
17: relu6 backward 0 0 0 3
View entry in raw format

Developed by Grigori Fursin           
Implemented as a CK workflow 		                         
   
                      Hosted at