Collective Knowledge Aggregator
proof-of-concept
Crowd results
Raw CK browser
Graphs
Reports
Datasets
Models
Home
Select CK-powered unified experimental workflow:
crowd-benchmark DNN libraries and models using mobile devices
crowd-benchmark DNN libraries and models
open ReQuEST @ ASPLOS'18 tournament (Pareto-efficient image classification)
crowd-benchmark DNN libraries and models (Caffe - dev)
crowd-benchmark DNN libraries and models (Caffe2)
crowd-benchmark DNN libraries and models (TensorFlow)
crowd-benchmark DNN libraries and models (dividiti desktop app)
auto/crowd-tune GCC compiler flags (minimize execution time)
auto/crowd-tune LLVM compiler flags (minimize execution time)
auto/crowd-tune GCC compiler flags (minimize execution time, do not degrade code size)
auto/crowd-tune GCC compiler flags (minimize total binary size, do not degrade execution time)
auto/crowd-tune LLVM compiler flags (minimize execution time, do not degrade code size)
auto/crowd-tune OpenCL-based CLBlast (GFLOPs)
crowd-test OpenGL compilers (beta)
crowd-test OpenCL compilers (beta) - crowdsource bug detection via CK
crowd-benchmark shared workloads via ARM WA framework
(under development) auto/crowd-tune CUDA work size (execution time)
(under development) auto/crowd-tune OpenCL local work size (execution time)
(under development) auto/crowd-tune OpenCL local work size (execution time/FPS vs energy)
(under development) crowdsource program scalability
(under development) crowdsource program numerical stability
(under development) crowdsource OpenCL bug detection
(under development) crowdsource modeling of program behavior
auto/crowd-tune GCC compiler flags (custom dimensions)
[ 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
]
[
CK intro
,
universal workload crowd-tuning
;
vision
and
YouTube lecture
]
Routine (CK wrapper):
clblast-tune-xgemm-fp32
GPGPU:
GeForce GTX 1060
GeForce GTX 1080
Intel(R) HD Graphics 4400
Mali-T628
Mali-T880
CPU:
ARM Cortex-A53 ; Samsung Exynos M1
Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz
Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz
ODROID-XU3
Platform:
Hardkernel (Hardkernel odroid-xu3 board based on EXYNOS5422 )
Hewlett-Packard (HP Z640 Workstation)
LENOVO ThinkPad X240 (20ALCTO1WW)
Razer 6.06 (Blade)
SAMSUNG SM-G930F
OS:
Android 6.0.1
Ubuntu 16.04.2 LTS
Windows 10
#
GPGPU
CPU
Platform
OS
Routine (CK wrapper)
GFLOPs
Time (s)
Dataset (M N K)
Best parameters
Choices (env)
CLBlast engine
Power consumption (W)
min / max
Memory usage (MB)
Bug detected?
User
Replay
1
GeForce GTX 1080
Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz
Hewlett-Packard (HP Z640 Workstation)
Ubuntu 16.04.2 LTS
xgemm-fp32
5804.0 ± ?
0.370 ± ?
1024 x 1024 x 1024
Version
master-953a5a9
seti@ck
[
See raw files
]
2
GeForce GTX 1060
Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
Razer 6.06 (Blade)
Ubuntu 16.04.2 LTS
xgemm-fp32
3549.6 ± ?
0.605 ± ?
1024 x 1024 x 1024
Version
master-4e51b1e
d664668acaf475a9
[
See raw files
]
3
GeForce GTX 1080
Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz
Hewlett-Packard (HP Z640 Workstation)
Ubuntu 16.04.2 LTS
xgemm-fp32
3273.6 ± ?
0.082 ± ?
512 x 512 x 512
Version
master-953a5a9
seti@ck
[
See raw files
]
4
Mali-T880
ARM Cortex-A53 ; Samsung Exynos M1
SAMSUNG SM-G930F
Android 6.0.1
xgemm-fp32
19.0 ± ?
112.886 ± ?
1024 x 1024 x 1024
Version
master-953a5a9
fgg-lenovo2
[
See raw files
]
5
Intel(R) HD Graphics 4400
Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz
LENOVO ThinkPad X240 (20ALCTO1WW)
Windows 10
xgemm-fp32
15.9 ± ?
0.263 ± ?
128 x 128 x 128
Version
master-953a5a9
fgg-lenovo2
[
See raw files
]
6
Intel(R) HD Graphics 4400
Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz
LENOVO ThinkPad X240 (20ALCTO1WW)
Windows 10
xgemm-fp32
9.9 ± ?
0.053 ± ?
64 x 64 x 64
Version
master-953a5a9
fgg-lenovo2
[
See raw files
]
7
Mali-T628
ODROID-XU3
Hardkernel (Hardkernel odroid-xu3 board based on EXYNOS5422 )
Ubuntu 16.04.2 LTS
xgemm-fp32
6.5 ± ?
41.521 ± ?
512 x 512 x 512
Version
master-953a5a9
fvella
[
See raw files
]
8
Mali-T880
ARM Cortex-A53 ; Samsung Exynos M1
SAMSUNG SM-G930F
Android 6.0.1
xgemm-fp32
1.5 ± ?
2.757 ± ?
128 x 128 x 128
Version
master-953a5a9
fgg-lenovo2
[
See raw files
]
Developed by
Grigori Fursin
Implemented as a
CK workflow
Hosted at
