Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

6. numPYNQ - StartUp101


Published on

numPYNQ aims to create an hardware library that features accelerated versions of the core functions of numPy, a software library for scientific computing commonly used in Python applications.

In this presentation we present the work done in the Startup101 course that we followed at NECSTLab - Politecnico di Milano.

Published in: Devices & Hardware
  • Be the first to comment

  • Be the first to like this

6. numPYNQ - StartUp101

  1. 1. StartUp 101 numPYNQ 21/06/17, DEIB Seminar Room Luca Stornaiuolo Riccardo Pressiani Filippo Carloni
  2. 2. CONTEXT DEFINITION 2 Signal Processing Machine LearningData MiningImage Analysis Huge amount of data that need to be processed to get aggregated information
  3. 3. TECHNICAL CHALLENGES 3 Performance Energy Consumption Design Time The need of higher and higher computational power brings some challenges to overcome
  4. 4. SOLUTION 4 numPYNQ numPYNQ is an FPGA-based hardware library that offers all the data scientists and software developers an accelerated version of the core functions of NumPy to be used transparently from their applications.
  5. 5. INNOVATION & TECHNOLOGY 5 Performance Energy Consumption
  6. 6. INNOVATION & TECHNOLOGY 6 Performance Energy Consumption Design Time
  7. 7. INNOVATION & TECHNOLOGY 7 Overlay Bitstream Python API
  8. 8. MARKET 8 The Big Data market is estimated to top $61 billion in 2020, a 26% compound annual growth rate for 2011-2020. Source: Wikibon research as covered by
  9. 9. CLIENTS 9 Two feasible options Data Analysis-based Companies
  10. 10. CLIENTS 10 • The PYNQ platform goal is to let FPGA non-expert users to easily exploit the power of such devices • numPYNQ would expand the available features for the PYNQ platform • Xilinx would be able to increase the revenues for the PYNQ-based products
  11. 11. CLIENTS 11 Data Analysis-based Companies • Data scientists need high-performing, flexible and easy to use systems • FPGAs are a great opportunity, although the high expertise and the design time required are often an important barrier • Data scientists could save money and time exploiting the numPYNQ features
  12. 12. GO TO MARKET 12 Data Analysis-based Companies Partnership deal to complete the development with the company support Revenue sharing Direct channel marketing through sales force, email and online marketing numPYNQ as a Service
  13. 13. REVENUE MODEL 13 Freemium model Basic Plan Premium Plan Advanced Plan Basic functions Advanced Functions — Dedicated Training — Dedicated Support — Custom Functions — — Cost Free $ $$$
  14. 14. REVENUE MODEL 14 An online store to sell packages of advanced functions to extend the Freemium Plans
  15. 15. COMPETITION ANALISYS 15 numPYNQ TensorFlow  PyCUDA
  16. 16. ROADMAP 16 Jun ‘17 M ar‘18 Jun ‘18 D ec‘18 Proof Of Concept at XOHW17* Second Phase of S2P Contest Complete development of the library Porting on F1 Instances Business Model Testing * Xilinx Open Hardware Contest 2017:
  17. 17. TEAM 17 Luca Stornaiuolo MSc student in Computer Science and Engineering at Politecnico di Milano Role in numPYNQ project: Hardware Design and Optimization Manager Riccardo Pressiani MSc student in in Computer Science and Engineering at Politecnico di Milano and University of Illinois at Chicago Role in numPYNQ project: Hardware Implementation and Social Media Manager Filippo Carloni BSc student in Computer Science and Engineering at Politecnico di Milano Role in numPYNQ project: PYNQ Overlay and Interfaces Manager
  18. 18. CONTACTS 18 numPYNQ Thanks for the attention! Luca Stornaiuolo Riccardo Pressiani Filippo Carloni