The NGD Systems and AIC Solution for Artificial Intelligence Applications

We are now into Day 4 of the 2019 NVIDIA GPU Technology Conference (GTC), and there has been a lot of exciting developments! NGD Systems is sharing Booth 530 with AIC, where we are demonstrating a jointly-developed computational storage solution. Our solution provides customers who want to utilize computational storage with a turnkey solution to accelerate their applications. The solution combines an AIC FV2019-LX 2U/24-bay NVMe server with up to 24 NGD Systems ICS-8100 NVMe U.2 Computational Storage solid-state drives (SSDs). This solution provides up to 384 TB (raw) of storage capacity (16TB per drive) in 2U of rack space.
While this amount of storage capacity is impressive for 2U, it is not the most impressive aspect of this solution. By utilizing NGD Systems’ Newport computational storage platform, each of the (up to) 24 ICS-8100 drives can utilize its multi-core ARM processors running the Ubuntu OS, and dedicated artificial intelligence (AI) acceleration hardware to act upon the data that they contain. This is the basic premise of in-situ processing – move the computational resources to the data, rather than using up power and latency moving the data from the storage device to the server CPU and its memory. In-Situ processing allows the computational storage compute resources to filter and/or pre-process the data in the SSD, and return to the host server CPU only the data that is specifically required for a particular query. This can reduce the latency of data movement by up to 95% in most applications, which is significant for petabyte-scale data sets.
The block diagram below shows how the Newport Computational Storage Controller interacts with the host CPU to accelerate applications. We have also built a development environment and management tools that simplifies the deployment and provisioning of applications across multiple Newport computational storage SSDs. The level of acceleration realized varies by application, but our testing across a variety of applications show that NGD Systems computational storage devices allow workload performance can be scaled linearly with the size of the data set. If you would like to find out more, please contact me, or better yet stop by Booth 530 at GTC. Looking forward to seeing you!