An efficient Binarized-Neural-Network (BNN) design accelerated by NVIDIA Turing Bit-Tensor-Cores. Please see our paper on arXiv) for details.
For our referencing BSTC SBNN design, please see our SuperComputing-19 paper for detail and our SBNN repository.
Latest version: 0.1
Despite foreseeing tremendous speedups over conventional deep neural networks, the performance advantage of binarized neural networks (BNNs) has merely been showcased on general-purpose processors such as CPUs and GPUs. In fact, due to being unable to leverage bit-level-parallelism with a word-based architecture, GPUs have been criticized for extremely low utilization (1%) when executing BNNs. Consequently, the latest tensorcores in NVIDIA Turing GPUs start to experimentally support bit computation. In this work, we look into this brand new bit computation capability and characterize its unique features. We show that the stride of memory access can significantly affect performance delivery and a data-format co-design is highly desired to support the tensorcores for achieving superior performance than existing software solutions without tensorcores. We realize the tensorcore-accelerated BNN design, particularly the major functions for fully-connect and convolution layers — bit matrix multiplication and bit convolution. Evaluations on two NVIDIA Turing GPUs show that, with ResNet-18, our BTC-BNN design can process ImageNet at a rate of 5.6K images per second, 77% faster than state-of-the-art.
Update Makefile accordingly and make. You will need a NVIDIA Turing GPU (Compute Capability-7.5) to be able to run.
make
Ang Li, Pacific Northwest National Laboratory (PNNL)
For research articles, please cite our paper:
- Ang Li, Simon Su, "Accelerating Binarized Neural Networks via Bit-Tensor-Cores in Turing GPUs" [arXiv:2006.16578].
Bibtex:
@article{li2020accelerating,
title={Accelerating Binarized Neural Networks via Bit-Tensor-Cores in Turing GPUs},
author={Li, Ang and Su, Simon},
journal={arXiv preprint arXiv:2006.16578},
year={2020}
}
This project is licensed under the BSD License, see LICENSE file for details.
PNNL-IPID: 31925-E, ECCN: EAR99, IR: PNNL-SA-152850
This research was supported by PNNL's DeepScience-HPC and DMC-CFA LDRD projects. This research was supported by the U.S. DOE Office of Science, Office of Advanced Scientific Computing Research, under award 66150: "CENATE - Center for Advanced Architecture Evaluation". The Pacific Northwest National Laboratory (PNNL) is operated by Battelle for the U.S. Department of Energy (DOE) under contract DE-AC05-76RL01830.
Please contact us If you'd like to contribute to TC-BNN. Thank you!