Loading…
October 31 - November 1, 2019 | Lyon, France
View More Details & Register Here
Thursday, October 31 • 15:50 - 16:30
AI Inference Acceleration with components all in Open Hardware: OpenCAPI and NVDLA - Peng Fei Gou, IBM

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
NVDLA is a deep learning accelerator with both hardware and software available in open source community. IBM announced open hardware strategy recently, which focuses on building open ecosystem of OpenCAPI based heterogeneous computing system. To align with this open strategy, we proposed the solution to connect NVDLA with OpenCAPI so that the AI inference acceleration becomes possible with all components in open hardware community. The NVDLA RTL is modified and optimized for being integrated to OC-Accel framework, while the whole software stack of NVDLA is ported to OpenCAPI with kernel mode driver eliminated.  The proposed NVDLA+OpenCAPI solution is able to run various deep  neural networks  such as AlexNet and YOLO for image classification and object detection with competitive performance speedup. 

Speakers
avatar for Peng Fei Gou

Peng Fei Gou

Hardware Architect, IBM
Technical lead of IBM China OpenPOWER chip development team. Years of experience on state-of-the-art processor architecture research/development, industrial level CPU and SoC verification and compiler development. Works for projects such as IBM POWER9, OpenPOWER, OpenCAPI/CAPI, Nvidia... Read More →



Thursday October 31, 2019 15:50 - 16:30 CET
Lumiere (Track 3)
  OpenCAPI  Machine Learning
  • Session Slides Included Yes