Unable to run OpenVINO on three or more FPGAs

[Tony7]

We tried to load the same neural network on three different FPGAs as shown in the program below.

#include "inference_engine.hpp"
int main(int argc, char *argv[])
{
InferenceEngine::Core core;
auto network1 = core.ReadNetwork(".xml");
auto exeNetwork1 = core.LoadNetwork(network1, "FPGA.0");
auto network2 = core.ReadNetwork(".xml");
auto exeNetwork2 = core.LoadNetwork(network2, "FPGA.1");
auto network3 = core.ReadNetwork(".xml");
auto exeNetwork3 = core.LoadNetwork(network3, "FPGA.2");
return 0;
}


However, the OpenVINO program crashed with the error shown below. However, commenting out any of the three networks would fix the problem.

Error initializing DMA: 5
Error initializing mmd dma
Error initializing bsp
Error Code: 147
Error Description: Error: Failure due to generic standard exception.User requested device 2 is invalid
File: /home/jenkins/agent/workspace/private-ci/ie/build-linux-ubuntu16/b/repos/fpga-plugin/thirdparty/dla/runtime/core/src/new_device.cpp
Function: initialize
Line #: 101
terminate called after throwing an instance of 'InferenceEngine::details::InferenceEngineException'
what(): Failed to create FPGA device handle with ID 2
Aborted (core dumped)

Environment:
Ubuntu 16.04
OpenVINO version: 2020.0301
Aocl version: 19.2
Intel® Programmable Acceleration Card Stack version: 1.2

[Zulkifli_Intel]

Hello Tony7,

Thank you for contacting us.

The error of “Failure due to generic standard exception. User requested device 2 is invalid” is due to the Inference Engine Core class only supports 2 FPGA devices, FPGA.0 and FPGA.1. ID 2 is not recognized as mentioned in InferenceEngine::Core Class Reference GetAvailableDevices().

Regards,

Zulkifli

[Tony7]

Hi Zulkifli,

Thanks for the reply.

I created the program below to print the available devices, and FPGA.2 is recognized as one of the available devices. 

InferenceEngine::Core core;

auto devices = core.GetAvailableDevices();

for (auto device: devices){

    cout << device << endl;

}

The output of the program is shown below.

(aocl) autodrive@AUTODRIVE-BOLT:~/dla/new_openvino_test/benchmark_ws$ ./devel/lib/zeus_inference_engine/YoloTest 

CPU

FPGA.0

FPGA.1

FPGA.2

Regards,

Tony

[Zulkifli_Intel]

Hello Tony7,

Multiple FPGA devices can be supported for OpenVINO simultaneously if they are running different models in each respective cards as per below configuration example:

Card 1: Object Detection with SSD-VGG Sample

Card 2: Image Segmentation Sample with FCN-8.

Card 3: Object Detection for Faster R-CNN Sample

Card 4: Image Classification Sample with Alexnet.

The step can be obtained at FPGA Plugin - OpenVINO Toolkit

However, the current configuration only supports 2 FPGA devices for the same model.

Regards,

Zulkifli

[Zulkifli_Intel]

Hello,

This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.

Regards,

Zulkifli