# PYNQ board and GPU integration

**URL:** <https://discuss.pynq.io/t/pynq-board-and-gpu-integration/5790>\
**Category:** Support\
**Created:** [July 7, 2023, 11:00am UTC](https://discuss.pynq.io/t/pynq-board-and-gpu-integration/5790 "2023-07-07T11:00:25Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![mselva2](https://avatars.discourse-cdn.com/v4/letter/m/bb73d2/32.png) [@mselva2](https://discuss.pynq.io/u/mselva2)\
**Post date:** [July 7, 2023, 11:00am UTC](https://discuss.pynq.io/t/pynq-board-and-gpu-integration/5790/1 "2023-07-07T11:00:25Z")

</div>

Has anyone tried connecting a FPGA board with a GPU like the NVIDIA Jetson Board?

---

<div class="post-metadata">

**Author:** ![briansune](https://sea1.discourse-cdn.com/flex019/user_avatar/discuss.pynq.io/briansune/32/3087_2.png) [@briansune](https://discuss.pynq.io/u/briansune)\
**Post date:** [July 7, 2023, 12:05pm UTC](https://discuss.pynq.io/t/pynq-board-and-gpu-integration/5790/2 "2023-07-07T12:05:33Z")

</div>

@mselva2

Out of this forum support ground.  
But I will personally provide some info:  
FPGA + GPU is commonly found in research paper.  
Most design need FPGA support PCIe x4 or x8 Gen2 or Gen3 above.  
Aka Riffa or Xilinx DMA PCIe  
Any network like 1Gbps can also make FPGA connect to host(PC) with GPU but the latency and bandwidth is suboptimal.

Such that the computation can be divided into each best location.  
Specific algorithm can be accelerate on FPGA while Tensor Core can do neural computation very well these days.

ENJOY~

---

<div class="post-metadata">

**Author:** ![cathalmccabe](https://sea1.discourse-cdn.com/flex019/user_avatar/discuss.pynq.io/cathalmccabe/32/3348_2.png) [@cathalmccabe](https://discuss.pynq.io/u/cathalmccabe)\
**Post date:** [July 7, 2023, 12:16pm UTC](https://discuss.pynq.io/t/pynq-board-and-gpu-integration/5790/3 "2023-07-07T12:16:31Z")

</div>

This group built a soft GPU in the programmable logic, and used PYNQ that may be of interest:

> **[Edge GPU based on an FPGA Overlay Architecture using PYNQ](https://ieeexplore.ieee.org/document/9893229)**
>
> Using Graphical Processing Units (GPUs) to accelerate applications such as Deep Neural Networks (DNNs) or image processing tasks has been gaining lots of attention for some time now due to their high performance. This has brought a lot of attention...

(You may need IEEE sign-in to access the full paper)

Cathal