search

doonny / PipeCNN

An OpenCL-based FPGA Accelerator for Convolutional Neural Networks
Apache License 2.0
1.26k stars 369 forks source link

Performance On KCU1500 with SDx Environment #98

Closed yuanm2 closed 3 years ago

yuanm2 commented 6 years ago

We successfully execute the PipeCNN's Alexnet model and here is our performance summary for testing first 10 images from imagenet. We notice that the coreConv kernel takes 339.24ms on average and macPool 521.841ms. Does anyone have the similar results?

performance_sdx.pdf

doonny commented 5 years ago

Anyone has optimization suggestions for Xilinx FPGAs ?

cococener commented 4 years ago

My

We successfully execute the PipeCNN's Alexnet model and here is our performance summary for testing first 10 images from imagenet. We notice that the coreConv kernel takes 339.24ms on average and macPool 521.841ms. Does anyone have the similar results?

performance_sdx.pdf Sorry,I have a question,I want to know if this project can be compiled by SDAccel ,Because when I compiled it ,it can't produce .xclbin,I used HW and System mode,just don't have the .xclbin,and also build finished.And if you can ,can you tell me what change do you do for the pipe project,Because my board is KCU1500,too.