GTC 2019, Artificial Intelligence, and Computational Storage

NVIDIA’s 2019 GPU Technology Conference (GTC) kicked off on Monday of this week at the San Jose Convention Center. While the name of the conference might suggest to the uninitiated that it would be around gaming and other graphics technology, GTC is arguably the leading conference on artificial intelligence (AI), due to the nearly-universal use of general-purpose graphics processing units (GPGPUs) for AI workloads and high-performance computing (HPC). The 2019 GTC Conference covers topics including AI/Deep Learning, Accelerated Data Science, Graphics/Simulation, Intelligent Machines/IoT, HPC, Data Center/Cloud, GPGPU software development, and using AI for business. Vertical industries will include automotive/transportation, media/entertainment, healthcare/life sciences, national/university labs, industrial/manufacturing, telecommunications, finance, and retail.

Like NVIDIA, NGD Systems is also very interested in AI and machine learning (ML). Our Newport family of computational storage products provides dedicated hardware as well as multi-core ARM resources to accelerate AI workloads that utilize extremely large (100s of gigabyte to petabyte scale) datasets. Since GPGPUs are also used for AI/ML, one might wonder how these two technologies compliment each other. By providing scalable AI and compute resources inside the SSD, NGD Systems can accelerate GPGPU-based AI/ML workloads by “tailoring” the data that is sent to the GPGPU. This can both accelerate the GPGPU-based portion of the solution (less data, with significantly reduced time to transmit the data form storage to the GPGPU) and improve results by providing data pre-filtering. One potential use case like this would be for machine vision in manufacturing – the GPGPU application could provide in image to Newport, which then could provide data relevant to that image, such as the attributes of the components on the image.

And, this is only the first level of potential integration between GPGPUs and NGD Systems Newport computational storage SSD. A second-level approach could be to have the GPGPU portion of an application “tune” the data filtering that occurs in the Newport SSD. This would provide a “closed loop”, and provide additional learning capabilities to optimize the performance of the overall solution. From our perspective, GPGPUs and computational storage are not competing technologies – they are very much complimentary. NGD Systems will be at GTC this week in Booth 530 – stop by and see us! If you would like to set up a meeting, you can contact me as well. See you at GTC!

2019-03-19T09:26:03+00:00