cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Highlighted
386 Views
Registered: ‎01-13-2020

Vitis AI compiler DPU V3

Jump to solution

Hi,

I'm trying to deploy a Neural Network on a ZCU104 using Vitis AI. 

I have compiled a custom quantized model using the V2.1 in the tools docker. But I got an error while trying to execute my model on the the DPU V3.1: because my DPU version is 3.1 and my executable is 2.1. 

However I have tried to execute DNNDK Adas sample (V2.1 too) on a DPU (V3.1 too) on the ZCU104 and it works perfectly...

I would to know what Xilinx has done in order to make the V2.1 samples work on a V3.1 DPU. I don't understand why did my model get a version error and why Adas sample not.

To solve my problem I thought to use the V3 compiler. However,I have seen on the web that the V3 compiler is only for Alveo Accelerator:

https://github.com/Xilinx/Vitis-AI/blob/master/doc/release-notes/1.x.md 

https://developer.xilinx.com/en/articles/dpuv3e-for-alveo-accelerator-card-with-hbm.html 

Do you have any suggestion that could help me with this issue?

 

Thank you,

Erwann

0 Kudos
1 Solution

Accepted Solutions
Highlighted
Xilinx Employee
Xilinx Employee
338 Views
Registered: ‎03-21-2008

Are you using the Vitis-AI prebuilt Linux Image for the ZCU104?

It can be found at: https://www.xilinx.com/bin/public/openDownload?filename=xilinx-zcu104-dpu-v2019.2.img.gz

You will also probbaly want to install the Vitis-AI Library packages on the board as well (see instrcutions at https://github.com/Xilinx/Vitis-AI/tree/master/Vitis-AI-Library)

Also make sure you  use the .json file located at opt/vitis_ai/compiler/arch/dpuv2/ZCU104/ZCU104.json. for the compiler arch option This will match the dpu config in the ZCU104 image.

One common area of confusion is between the the DPU version and the compiler verson that it supports. You might think we would have alligend these to be the same, but unfortunatly we did not do that.

So the DPU version 3.1 that you see on the Viti-AI TRD is a Vivado IP version, but the Vitis AI compiler version that supports it is different. To make it more confusing we call this a v2 DPU.  

At some point I hope we will  change this to the make versoins make more sense.

 

Here's the bottom line:

If you use the ZCU104 Vitis AI Linunx image, and use the file /opt/vitis_ai/compiler/arch/dpuv2/ZCU104/ZCU104.json for the arch option for the Vitis-AI compiler you should be good to go.

 

 

View solution in original post

1 Reply
Highlighted
Xilinx Employee
Xilinx Employee
339 Views
Registered: ‎03-21-2008

Are you using the Vitis-AI prebuilt Linux Image for the ZCU104?

It can be found at: https://www.xilinx.com/bin/public/openDownload?filename=xilinx-zcu104-dpu-v2019.2.img.gz

You will also probbaly want to install the Vitis-AI Library packages on the board as well (see instrcutions at https://github.com/Xilinx/Vitis-AI/tree/master/Vitis-AI-Library)

Also make sure you  use the .json file located at opt/vitis_ai/compiler/arch/dpuv2/ZCU104/ZCU104.json. for the compiler arch option This will match the dpu config in the ZCU104 image.

One common area of confusion is between the the DPU version and the compiler verson that it supports. You might think we would have alligend these to be the same, but unfortunatly we did not do that.

So the DPU version 3.1 that you see on the Viti-AI TRD is a Vivado IP version, but the Vitis AI compiler version that supports it is different. To make it more confusing we call this a v2 DPU.  

At some point I hope we will  change this to the make versoins make more sense.

 

Here's the bottom line:

If you use the ZCU104 Vitis AI Linunx image, and use the file /opt/vitis_ai/compiler/arch/dpuv2/ZCU104/ZCU104.json for the arch option for the Vitis-AI compiler you should be good to go.

 

 

View solution in original post