Copyright © 2014 BDTI 1

Jeff Bier, President, BDTI

May 29, 2014

Trends and Recent Developments in

Processors for Vision

Copyright © 2014 BDTI 2

• BDTI provides:

• Product development engineering services

• Emphasis on optimization for performance, cost and power

• Benchmarking and evaluation

• For technology selection, feasibility studies, competitive

analysis and proof points

• Focused on:

• Algorithm-intensive applications: vision, video, audio, wireless

• Embedded processors, tools and techniques

“BDTI’s deliverables were orders of magnitude better than other

vendors’. BDTI has set a new standard for quality.”

– Group Program Manager, Fortune 500 Systems & Software Co.

See our fun vision demos in the Technology Showcase!

About BDTI

Copyright © 2014 BDTI 3

The Fundamental Challenge

High Performance

Low Cost

Low Power

Small

Ease of

Development

(and Optimization)

Flexibility

Copyright © 2014 BDTI 4

Processing Efficiency vs. Development Effort

Performance/$

Performance/W

Development

Effort

Copyright © 2014 BDTI 5

Reality

Performance/$

Performance/W

Development

Effort

Copyright © 2014 BDTI 6

A Rising Tide

Performance/$

Performance/W

Development

Effort

Copyright © 2014 BDTI 7

Trend: Heterogeneous Architectures

Performance/$

Performance/W

Development

Effort

Copyright © 2014 BDTI 8

• Very heterogeneous processors

• Benefit from huge investments by suppliers, because mobile market is

huge

• Hardware performance, efficiency, integration

• Application development infrastructure

• Mobile apps have become a primary locus of software development

• APs can be difficult to buy and use for embedded applications

• APs are used in some embedded applications (sometimes in mobile

device form, sometimes via a system-on-module)

Trend: Mobile Application Processors

CPU GPU

DSP

ISP

VPU

Copyright © 2014 BDTI 9

• Graphics processing units (GPUs) are massively parallel machines

• Over the past decade, GPUs and their tools have evolved to support

non-graphics workloads (“general-purpose GPU” or “GPGPU”)

• Widely used in demanding workstation and data center applications

in which data parallelism is abundant

• E.g., Photoshop, FFTs

• NVIDIA pioneered this concept with CUDA

• Others have joined via OpenCL and RenderScript

• Important recent developments:

• Now in mobile application processors and embedded processors

• Expanding support via libraries, code examples,

optimized middleware

• OpenCL support

• HSA (Heterogeneous System Architecture)

Trend: General-Purpose GPU

Copyright © 2014 BDTI 10

• As more vision applications achieve high volumes, vision-specific

processors are emerging

• All are co-processors, working in tandem with a CPU

• Many are sold as licensable IP for custom chips:

• CogniVue, CEVA, Cadence (Tensilica), videantis, Apical

• A few are sold as chips:

• Analog Devices (PVP), Mobileye, Movidius, TI (EVE)

• Some are do-it-yourself kits:

• For chip designers: Synopsys Processor Designer

• For system designers: Xilinx Zynq

• Challenges:

• Unique programming models and environments

• Limited libraries

• Important (potential) trend: OpenVX

Trend: Vision-Specific (Co-)Processors

Copyright © 2014 BDTI 11

• Different processors are better suited for different applications

• Applications are diverse, and so are processors

• Selecting a processor is a multi-dimensional optimization problem with

incomplete data

• Allow time to evaluate your options

• Macro trend: Increasing industry investment in processors for vision

• Including mobile application processors

• You can benefit by riding on the coattails of high-volume applications

But, progress is uneven and unpredictable Risk for users

• A processor is only as good as its tools, libraries, etc.

Lessons Learned

Copyright © 2014 BDTI 12

• Processors that are:

• Optimized for vision

• Flexible in I/O options

• Readily available (eval boards, small quantities)

• Easy to program (and optimize for)

• Supported by a roadmap we can be confident in

Wish List

Copyright © 2014 BDTI 13

• “Vision: Promise & Challenge of Heterogeneous Processing,”

http://bit.ly/1rjdBT5

• “Processing Options For Implementing Vision Capabilities in Embedded

Systems,” http://bit.ly/1wq16oW

• “Targeting Computer Vision Algorithms to Embedded Hardware,”

http://bit.ly/1nY2wkD

• “Using Heterogeneous Computing for Mobile and Embedded Vision,”

http://bit.ly/1nY2wkD

Resources