To implement an AI algorithm on DE1-SOC FPGA using OpenVINO toolkit

[vinmor]

Good morning everyone,

my goal is to implement an AI algorithm on DE1-SOC FPGA using OpenVINO toolkit.

I installed the OpenVINO toolkit (version 2022.3.0) on my PC (Windows 11 OS) and I followed some interactive tutorials to learn how to do inference on my PC's CPU and GPU.

Then, I connected FPGA DE1-SOC to PC via USB cable but it is not detected as available device with "ie.available_devices" command (only CPU and GPU of PC are shown).
I'm sure the FPGA is properly connected to the PC because I manage to transfer programs to it using Quartus Prime 21.1 (Lite Edition)

So, I would like to ask if my goal is attainable.
How could i fix this?

Thanks in advance,
Best regards.

[Hairul_Intel]

Hi vinmor,

Thank you for reaching out to us.

The last version of OpenVINO™ toolkit that supported FPGA based cards was 2020.3.2 LTS, and that support ended in October 2021.

The latest versions of Intel® Distribution of OpenVINO™ toolkit (2020.4 onwards) no longer support Intel® Vision Accelerator Design with an Intel® Arria® 10 FPGA and the Intel® Programmable Acceleration Card with Intel® Arria® 10 GX FPGA.

For your information, Intel has transitioned to the next-generation programmable deep-learning solution based on FPGAs. Inquiries regarding Intel® FPGA AI Suite should be directed to your Intel Programmable Solutions Group account manager or subscribe for the latest updates.

Hence, we recommend you contact PSG via the form here.

Regards,

Hairul

[vinmor]

Hi Hairul,
thank you for answering.

I'm new in the world of artificial intelligence and Openvino, so I apologize if my questions are trivial.

I read that you named "Intel® Arria® 10 FPGA" and "Intel® Arria® 10 GX FPGA" but I have DE1-SoC with the following specifications:

- Altera Cyclon V FPGA (5CSEMA5F31C6N) with 85K LEs

- 800MHz, Dual-core ARM Cortex -A9 MPCore Processor

- Gigabit Ethenet and two 2x20 GPIO Headers

- 64MB SDRAM /16-bit data bus) and 1GB DDR3 (32-bit data bus)

- USB Host x2, Micro SD Card Socket, Uart-to-USB

- VGA Out, Video-In, Audio

So, if I understand correctly, I can't read an Intermediate Representation model and execute it on my device with OpenVINO, is that correct?
Or can I do it with Intel FPGA AI Suite?
If not, are there other ways?

Thanks in advance,
Best regards.

[JohnT_Intel]

Hi,

DLA is not supported in Cyclone V device. You will need to migrate to Arria 10 or newer device. It will need to be enable with AI Suite implementation.

[vinmor]

Hi JohnT,
thank you for answering.

So, I necessarily need to have Intel FPGA AI Suite to implement AI algorithm on FPGA with OpenVINO, is that correct?
However, I've read that Intel FPGA AI Suite has very restrictive requirements like:

- Ubuntu operating system
- at least one PCIe slot
- 48GB RAM

and is compatible with Intel Quartus Prime Pro Edition and supports only the following FPGA families:

- Intel Agilex 7,
- Intel Arria 19,
- Intel Cyclon 10GX,
- Intel Stratix 10.

Did I get it right?
Do you confirm this?

Thanks in advance,
Best regards.

[JohnT_Intel]

Hi,

The requiement is correct to build the bitstream for the FPGA. The listed FPGA is the verrified device with AI Suite.