Since their beginnings, FPGA's have been notorious for being hard to program. That could be changing with the new Vitis Unified Software Platform from Xilinx. Five years in the making, the Vitis unified software platform is designed to allow a whole new user base of software engineers and AI scientists to take advantage of the power of hardware adaptability. "The Vitis unified software platform automatically tailors the Xilinx hardware architecture to the software or algorithmic code without the need for hardware expertise. Rather than imposing a proprietary development environment, the Vitis platform plugs into common software developer tools and utilizes a rich set of optimized open source libraries, enabling developers to focus on their algorithms. Vitis is separate to the Vivado Design Suite, which will still be supported for those who want to program using hardware code, but Vitis can also boost the productivity of hardware developers by packaging hardware modules as software-callable functions. With exponentially increasing compute needs, engineers and scientists are often limited by the fixed nature of silicon,” said Victor Peng, president and chief executive officer, Xilinx. “Xilinx has created a singular environment that enables programmers and engineers from all disciplines to co-develop and optimize both their hardware and software, using the tools and frameworks they already know and understand. This means that they can adapt their hardware architecture to their application without the need for new silicon.” Learn more: https://www.xilinx.com/products/design-tools/vitis.html Sign up for our insideHPC Newsletter: http://insidehpc.com/newsletter

1. Introducing the Vitis
Unified Software Platform
Ramine Roane
VP Software & AI Product Management
Under embargo until Oct. 1, 2019 @ 9 am Pacific, noon Eastern

2. Announcing…
• Available in a month
• Standards, Open
• Free!
Unified Software Platform

3. VITAL
VITALITY

4. Heterogeneous
Compute
Cloud to Edge AI Proliferation
Industry
Trends

5. Industry Trend: Cloud/Edge Unification

6. Genomics Video Analytics Healthcare Finance
Data Center 5G Autonomous Driving Security
Power efficient inference
along with traditional
software
AI Proliferation
Industry Trend: AI Proliferation

7. 2012 2019
#DEVELOPERS
Vivado
OS and
Firmware SDK
SDSoC
(Embedded)
SDAccel, Data Center
(FaaS, Alveo)
AI Inference
Acceleration
Vitis Unified
Software Platform
Platform Transformation

8. Industry Trend: Heterogeneous Compute
Cache
Cache Cache
1980-2000
2x/ 1.5y
process → Dennard scaling
2000-2010
2x/ 3.5y
multithreading → Amdahl’s law
2010-2020
2x/ 10y
density → Moore’s law
SINGLE CORE MULTICORE HETEROGENEOUS ADAPTIVE
HETEROGENEOUS
Cache
Scaling from: Silicon process Architecture-aware software Software-aware architecture
AcceleratorCPU Multicore CPU Multicore CPU FPGA, ACAP

9. 2012 2018
80
70
50
AlexNet
60
BN-AlexNet
BN-NIN
ENet
GoogLeNet
ResNet-18
VGG-16
VGG-19
ResNet-34
ResNet-50
ResNet-101
ResNet-153 ResNeXt-101
Inception v3
Inception-v4
DenseNet-264 ShuffleNet 2x
SENet-154
MobileNet v2
Top-1Accuracy(1%)
Silicon Design Cycle
Pace of AI/ML Innovation
Speed of Innovation Outpaces Silicon Cycles
Innovation Cycle

10. Custom Data Flow
Custom Precision
Custom Memory
APPLICATION
DOMAIN
ARCHITECTURE
Architecture Adaptability

11. Programmable OR Adaptable
Application Architecture
1
ASIC
ADAPTABLE (once)
COMPUTE EFFICIENCY
PROGRAMMABLE
CPU, GPU, ASSP
1
COMPUTE EFFICIENCY
1
3
3
2
2

12. Why Not Programmable AND Adaptable?
FPGA, ACAP
PROGRAMMABLE
COMPUTE EFFICIENCY
1
DSA2
ADAPTABLE
DSA1
2
21
Application Architecture
2

13. Xilinx runtime library (XRT)
Vitis target platform
Domain-specific
development
environment
Vitis core
development kit
Vitis accelerated
libraries
OpenCV
Library
BLAS
Library
Vitis AI Vitis Video
Partners
Genomics,
Data Analytics,
And moreFinance
Library
Analyzers DebuggersCompilers
Vitis: Unified Software Platform
Coming soon…

14. 1997 2004 2009 2014 2019 2024
IBM Watson becomes
Jeopardy champion!
Image
classification
Classification
better than humans
AlphaGo beats
Lee Sedol
AlphaZero
chess champion!
ADAS
Deep Blue (traditional software)
beats Garry Kasparov
Complexity: 10^120
Robo-taxis
(geofenced) Fully
autonomous
vehicles
Deep Learning vs. Traditional Software

15. Announcing Vitis AI
Xilinx runtime library (XRT)
Vitis target platform
Domain-specific
development
environment
Vitis core
development kit
Vitis accelerated
libraries
OpenCV
Library
BLAS
Library
Vitis AI Vitis Video
Partners
Genomics,
Data Analytics,
And moreFinance
Library
Analyzers DebuggersCompilers
Coming soon…

16. Frameworks
Vitis AI
development kit
Vitis AI
models
DSA
Vitis AI: Deep Learning Acceleration

17. DNN Processing Unit (DPU)
Direct Framework Compilation
Minutes of Compile Times
AI Model
Quantize
Optimize
Vitis AI
Rapid Iterations from TensorFlow

18. Shell
Hardware
Developers
Application
Software Developers
AI Scientists
(iterations in minutes)
Embedded
Developers
Putting it All Together

19. Trace
if (is_uyvy) {
uyvy2bgr (in_mat, in_rgb);
}
else {
yuyv2bgr (in_mat, in_rgb);
}
resize <INTERPOLATION_AREA,
MAX_IN_HEIGHT,
MAX_IN_WIDTH,
MAX_OUT_HEIGHT,
MAX_OUT_WIDTH,
NPC,
MAX_DOWN_SCALE> (in_r, out_r);
cv.cpp
Application Example: Smart City

20. In Programmable Logic
AI acceleration
in AI Engine
Preprocess
Running in CPU Preprocess
Vitis Dataflow
Pipelining P
P
AI
AI
AI
Postprocess
Acceleration in
Programmable Logic
P AI Postprocess
AI
AI 6 FPS
30 FPS
40 FPS
80 FPS
Postprocess
Postprocess
In AI Engine
Adaptive Architecture for Smart City Application

21. Live Demo

22. Standard, Open
Free for Xilinx Boards
Easily Accessible Examples, Tutorials, Documentation
Extensive Open Source Libraries
Empowering Software Developers

23. Launching developer.xilinx.com
Easy Access to Examples, Tutorials, Documentation
Connecting developers to experts
30+ expert articles & projects and growing
Covering 9 segments
Xilinx Developer > Authors
Authors

24. 400+ functions across 8 libraries
Open source, performance-optimized out-of-the-box acceleration
Extensive Open Source Libraries
Library
Docs
Source
Tests
Examples
Benchmarks
25 functions 12 99 114
365525 37 Models

25. Compilers
AI optimization
LLVM
User Since 2001
Contributor Since 2007
Now Core to Xilinx Strategy
Committed to Open Source
2007 Contributions2019
Runtime
Libraries
AI Models
20192019

26. Hardware Developers
Software Developers
AI & Data Scientists
Development Platforms
for ALL Developers

27. Unified Software Platform
Cloud to edge, software and AI
Comprehensive libraries and models
Work at Speed of Innovation
Hardware adaptable to software
Software programmable DSA
Standards, Open Source, Free
Embracing & participating in open source
Use of standard environments & APIs
Key Takeaways

28. Building the Adaptable,
Intelligent World
Mission