PYNQ Juliabrot Fractal Factory

A Xilinx FPGA accelerated near real-time interactive Python Jupyter Lab Notebook

Compatible with the Avnet PYNQ ZUBoard 1CG, Avnet PYNQ Ultra96 v1/v2 and Xilinx PYNQ Z1/Z2 development boards

To demonstrate potential, several full-rez up to 16K sized images are available: click here

The U96 board execution is faster than the ZUBoard 1CG & Z1/Z2 but otherwise the features are the same for both platforms.

Setup and installation:

• Install PYNQ v3.0.1 or newer on your ZUBoard 1CG, Ultra96 or PYNQ Z1/Z2 board

  • SD images and instructions can be found here: http://www.pynq.io

• (For PYNQ Z1/Z2 board only), open a PYNQ board console:

  • Use ssh xilinx@<your board IP address> (default password 'xilinx') OR
  • Open Jupyter Lab's root console

    • http://<your board IP address>/lab

    • If requested, enter xilinx for the default password

    • Click the Terminal Icon to open a console

    • Enter the commands below in the console to complete the installation, installing ipycanvas will take a while.

      sudo pip3 install ipycanvas

    • Next, install the notebook itself

      cd $PYNQ_JUPYTER_NOTEBOOKS
      git clone https://github.com/FredKellerman/pynq-juliabrot
      cd pynq-juliabrot
      git checkout origin/master

• If not already open, in your browser go to http://<your board IP address>/lab

• Use the Jupyter Lab Folder/File Explorer and under the folder pynq-juliabrot open juliabrot-zoom.ipynb

• Execute the notebook

• After executing jui.draw_roaming_ui() you should see:

• Use mouse click to start selection, click again to compute area within selection
• Enjoy!

Mandelbrot / Julia FPGA Compute Engine Attributes

Up to 16K x 16K grid sizes
Max Iterations up to 4,294,967,296

juliabrot Overlay	Precision	N Kernels	Logic MHz	# DSP48s
96b	64-bits	6	300MHz	300
96b_mid	95-bits	4	300MHz	320
96b_deep	160-bits	1	214MHz	248
z1 (z2 also)	64-bits	3	125MHz	150
ZUBoard 1CG	90-bits	3	250MHz	216

Note: at this time the Python front end used for initial conditions only supports up to 80-bit precision, future plans are to remove this limitation. Due to dev board memory limitations, images larger than 4K for coloring and formating should be completed off target on a PC.

The author would like to thank Github users @francof2a, @martinRenou for their kind replies and awesome contributions to the Open Source community.

https://github.com/martinRenou/ipycanvas

https://github.com/francof2a/fxpmath