posted Nov 15, 2020, 1:56 AM by MUHAMMAD MUN`IM AHMAD ZABIDI
[
updated Nov 15, 2020, 1:57 AM
]
Table 3 from M. G. S. Murshed, C. Murphy, D. Hou, N. Khan, G. Ananthanarayanan, and F. Hussain, “Machine Learning at the Network Edge: A Survey,” arXiv Prepr. arXiv1908.00080, pp. 1–28, 2019.
|
Device
|
GPU
|
CPU
|
RAM
|
Flash
memory
|
Power
consumption
|
Example
applications
|
|
Raspberry Pi
|
400MHz
VideoCore IV
|
Quad
Cortex A53 @ 1.2GHz
|
1 GB SDRAM
|
32 GB
|
2.5 Amp
|
video analysis [78, 114]
|
|
Coral Dev Board (Edge TPU)
|
GC7000 Lite
Graphics + Edge TPU coprocessor
|
Quad Cortex-A53, Cortex-M4F
|
1 GB LPDDR4
|
8 GB LPDDR4
|
5V DC
|
image processing [18]
|
|
SparkFun Edge
|
-
|
32-bit ARM
Cortex-M4F 48MHz
(with 96MHz
burst mode) processor
|
384KB
|
1MB
|
6uA/MHz
|
speech recognition [31]
|
|
Jetson TX1
|
Nvidia
Maxwell 256 CUDA
cores
|
Quad ARM A57/2 MB L2
|
4 GB
64 bit LPDDR4
25.6 GB/s
|
16 GB eMMC, SDIO, SATA
|
10-W
|
video, image analysis [68] [61]
robotics [33]
|
|
Jetson TX2
|
Nvidia Pascal
256 CUDA
cores
|
HMP Dual
Denver 2/2 MB L2 +
Quad ARM
A57/2 MB L2
|
8 GB
128 bit LPDDR4
59.7 GB/s
|
32 GB eMMC, SDIO, SATA
|
7.5-W
|
video, image analysis [68], [92]
robotics [24]
|
|
Intel Movidius Neural Compute Stick
|
High Performance VPU
|
Myriad 2 Vision Processing Unit
|
1 GB
|
4 GB
|
2 trillion
16-bit operations per second within
500 mW
|
classification [73] computer
vision [14, 43]
|
|
ARM ML
|
-
|
ARM ML
processor
|
1 GB
|
-
|
4 TOPs/W
(Tera Operations)
|
image, voice recognition [107]
|
|
RISC-V GAP8
|
-
|
nona-core
32-bit RISC-V
microprocessor
@250 MHz
|
16 MiB SDRAM
|
-
|
1 GOPs/mW
|
image, audio processing [35]
|
|
OpenMV
Cam
|
-
|
ARM 32-bit
Cortex-M7
|
512KB
|
2 MB
|
200mA
@ 3.3V
|
image
processing [6]
|
|
BeagleBone AI
|
-
|
Cortex-A15
Sitara AM5729 SoC with 4 EVEs
|
1 GB
|
16 GB
|
-
|
computer vision [25]
|
|
EMC3531
|
-
|
ARM Cortex-M3
NXP Coolflux DSP
|
-
|
-
|
-
|
audio, video analysis38
|
|
|