U
Utku Aydonat
Researcher at Intel
Publications - 17
Citations - 805
Utku Aydonat is an academic researcher from Intel. The author has contributed to research in topics: Compiler & Multiversion concurrency control. The author has an hindex of 12, co-authored 17 publications receiving 726 citations. Previous affiliations of Utku Aydonat include Altera & University of Toronto.
Papers
More filters
Proceedings ArticleDOI
From opencl to high-performance hardware on FPGAS
Tomasz Czajkowski,Utku Aydonat,Dmitry N. Denisenko,John Freeman,Michael Kinsner,David Neto,Jason Wong,Peter Yiannacouras,Deshanand Singh +8 more
TL;DR: It is shown that the OpenCL computing paradigm is a viable design entry method for high-performance computing applications on FPGAs and that it can achieve a clock frequency in excess of 160MHz on benchmarks.
Proceedings ArticleDOI
An OpenCL™ Deep Learning Accelerator on Arria 10
TL;DR: This work shows a novel architecture written in OpenCL(TM), which is referred to as a Deep Learning Accelerator (DLA), that maximizes data reuse and minimizes external memory bandwidth, and shows how the Winograd transform can be used to significantly boost the performance of the FPGA.
Posted Content
An OpenCL(TM) Deep Learning Accelerator on Arria 10
TL;DR: This work shows a novel architecture written in OpenCL(TM), which is referred to as a Deep Learning Accelerator (DLA), that maximizes data reuse and minimizes external memory bandwidth, and shows how the Winograd transform can be used to significantly boost the performance of the FPGA.
Journal ArticleDOI
A multilevel computing architecture for embedded multimedia applications
TL;DR: The multilevel computing architecture (MLCA) as mentioned in this paper is a template architecture featuring multiple processing units for multimedia and streaming applications, which uses superscalar techniques to exploit task-level parallelism among different processing units.
Patent
Method and Apparatus for Implementing Layers on a Convolutional Neural Network Accelerator
TL;DR: In this article, a convolutional neural network (CNN) accelerator is modified to change a data flow between components on the CNN accelerator to implement a fully connected layer in response to the change in the data flow.