Snowflake: A Model Agnostic Accelerator for Deep Convolutional Neural Networks

Gokhale, Vinayak; Zaidy, Aliasger; Chang, Andre Xian Ming; Culurciello, Eugenio

doi:10.48550/arxiv.1708.02579

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Caffeine [33] is another CNN accelerator for Caffe-specified models targeting Xilinx devices that support a co-processing environment with a PCIe interface between the FPGA and a host. Snowflake [34] is a scalable and efficient CNN accelerator with models specified in Torch [35] and a single, sequential computation architecture designed to perform at near-peak hardware utilization targeting Xilinx system-on-chips (SoCs). In Ref.…”

Section: Related Workmentioning

confidence: 99%

Fast convolutional neural networks on FPGAs with hls4ml

Aarrestad,

Loncar,

Ghielmetti

et al. 2021

Preprint

View full text Add to dashboard Cite

We introduce an automated tool for deploying ultra low-latency, low-power deep neural networks with large convolutional layers on FPGAs. By extending the hls4ml library, we demonstrate how to achieve inference latency of 5 µs using convolutional architectures, while preserving state-of-the-art model performance. Considering benchmark models trained on the Street View House Numbers Dataset, we demonstrate various methods for model compression in order to fit the computational constraints of a typical FPGA device. In particular, we discuss pruning and quantization-aware training, and demonstrate how resource utilization can be reduced by over 90% while maintaining the original model accuracy.

show abstract

Section: Related Workmentioning

confidence: 99%

Fast convolutional neural networks on FPGAs with hls4ml

Aarrestad,

Loncar,

Ghielmetti

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…To improve inference throughput, (fast) GPU solutions have been proposed to process a large amount of data [16], [17]. Field Programmable Gate Arrays (FPGAs), on the other hand, have been extensively used as an alternative to this problem as they offer good performance and reconfigurability [18]- [22]. Nevertheless, these architectures are not efficient power-performance solutions for critical edge applications, like surveillance cameras and cellphone face recognition, etc., which have stringent execution and power consumption constraints.…”

Section: The Nmp Architecturementioning

confidence: 99%