cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Highlighted
Voyager
Voyager
1,461 Views
Registered: ‎08-16-2018

Accelerated Pynq

Jump to solution

My understanding is that the Pynq image is 'just' a Linux OS with any Python on it for the Zynq PS.

If it was just this, I could just have any processor, so the question is, is it possible/ easy to add accelerated functions on the PL for Pynq to use access them and get that plus in speed? Any pointers/ examples how to do it?

Note: by 'accessing the PL' I don't mean writing to GPIOs and things like that but sending/ receiving data batches through AXI, etc.

0 Kudos
1 Solution

Accepted Solutions
Highlighted
Moderator
Moderator
1,444 Views
Registered: ‎11-09-2015

HI @johnvivm,

I recommend you to visit the web page for pynq. There is a dedicated forums for it.

Pynq will call an overlay which is basically a bitstream. You can lod you own overlay if you want to:

https://pynq.readthedocs.io/en/latest/pynq_overlays/loading_an_overlay.html

Regards,


Florent
Product Application Engineer - Xilinx Technical Support EMEA
**~ Don't forget to reply, give kudos, and accept as solution.~**

View solution in original post

0 Kudos
8 Replies
Highlighted
Moderator
Moderator
1,445 Views
Registered: ‎11-09-2015

HI @johnvivm,

I recommend you to visit the web page for pynq. There is a dedicated forums for it.

Pynq will call an overlay which is basically a bitstream. You can lod you own overlay if you want to:

https://pynq.readthedocs.io/en/latest/pynq_overlays/loading_an_overlay.html

Regards,


Florent
Product Application Engineer - Xilinx Technical Support EMEA
**~ Don't forget to reply, give kudos, and accept as solution.~**

View solution in original post

0 Kudos
Highlighted
Voyager
Voyager
1,436 Views
Registered: ‎08-16-2018

I see... lots of fluff, nothing serious. I was looking for a platform for accelerated computing and/ or ML/ NN inference. 

0 Kudos
Highlighted
Advisor
Advisor
1,428 Views
Registered: ‎04-26-2015

@johnvivm I get the feeling that there are really two options with the Pynq:

 

(1) Modify the provided overlay as much as necessary, ensuring that the essentials (eg. HDMI controller) remain the same. Then recompile the overlay, update the Python code to handle any blocks that you've added, and load that onto the board.

(2) Treat the Pynq as a standard Zynq board. Scrap Python and all of the provided software, just run bare-metal or Petalinux with your own IP cores in the PL. No need to worry about keeping any of the existing PL blocks. This way is, I think, probably better-documented - but even so it's not exactly trivial.

Highlighted
Voyager
Voyager
1,422 Views
Registered: ‎08-16-2018

(1) I don't need any HDMI for accelerated ML. I would scrap it, otherwise it's taking FPGA resources.

(2) What's the advantage in taking a Pynq overlay and scrap Python if I can start with my Zynq and set up my Linux?

Appreciate your input. If these are the options, I'm thinking Pynq is not the way for hw-accelerated ML.Or at least, there is no special advantage on it.

0 Kudos
Highlighted
Advisor
Advisor
1,318 Views
Registered: ‎04-26-2015

@johnvivm

 

(1) Excellent, that solves one problem.

 

(2) The only reason to use the Pynq in this case is that with the education discount, I'm pretty sure it's the cheapest Zynq 7020 board around. If you're not eligible for the education discount, don't bother.

0 Kudos
Highlighted
Moderator
Moderator
1,313 Views
Registered: ‎11-09-2015

Hi @u4223374,

You need to differenciate two things when talking about PYNQ:

  1. The framework
  2. the boards

You can do pynq on an Ultra96 board for example (or any other board).

The main advantage of pynq is that you can leverage project from the community on Zynq/ZynqMP without knowing anything about FPGA. You would only need to know python.

Then about the PYNQ board, note that with the v2 of the board, it is the same price for everyone (~120$), academic or not.

 


Florent
Product Application Engineer - Xilinx Technical Support EMEA
**~ Don't forget to reply, give kudos, and accept as solution.~**
0 Kudos
Highlighted
Voyager
Voyager
1,296 Views
Registered: ‎08-16-2018

"you can leverage project from the community on Zynq/ZynqMP without knowing anything about FPGA"

Quite a bold statement. Lets suggest buying a plane to drive her without knowing anything about flying.

0 Kudos
Highlighted
Moderator
Moderator
1,289 Views
Registered: ‎11-09-2015

Hi @johnvivm,

I would say it is more driving a car without know anything about mechanic.


Florent
Product Application Engineer - Xilinx Technical Support EMEA
**~ Don't forget to reply, give kudos, and accept as solution.~**