Computational Storage- A Compelling Solution for Bringing Intelligence to the Edge

Just like improving performance in complex systems, improving compute density and reducing power usage at the edge requires a comprehensive assessment of the sources of the problems, and not just through the incremental application of technologies. A recent study has shown that up to 50% of power consumption in datacenters results from data movement between systems and components. Given that, it would seem that minimizing data movement might be a good place to start the efforts around power reduction. Since less data movement also often means a smaller footprint, anything to reduce power consumption is also likely to help increase solution density as well.

One technology that holds significant promise to reduce data movement is computational storage. Computational storage embeds processing capabilities within storage systems or storage devices. While you can’t embed an Intel Xeon processor or a GPGPU inside of a solid-state drive, it is quite easy to embed multiple ARM cores inside of one – in fact, the bulk of modern flash controllers utilized in SSDs have multi-core ARM processors inside of them. The NGD Systems Newport computational storage platform provides these capabilities, plus hardware for AI, search, and encryption acceleration (watch the video here), all while maintaining a power profile of less than 12 watts (U.2 form factor, 32TB capacity). In addition to these capabilities, the Newport controller has up to 12.8GB/s of connectivity to its flash ASICs (over 3X the capacity of 4-lane Gen3 PCI Express bus). This is an extremely powerful set of tools for edge computing applications, and it is packed into an under 6.5 cubic inch package (2.75” x 3.95” x 0.59”).

How does this compare to the capabilities of typical single-board computer (SBC)? An Intel S72000APR SBC contains an Intel Xeon Phi 7200 processor, can hold 384GB of RAM, consumes 320W of power, and take up 338 cubic inches of space (6.8” x 14.2” x 3.5”). Throw in a 32GB SSD, and the power consumption goes up to 332W. That is almost 28 times the power consumed by the U.2 version of the Newport Computational Storage SSD. It is also 52 times the space taken up by the U.2 Newport. For an edge application such as a self-driving vehicle or an engine management system, the power and space savings provided by computational storage is substantial. Find out how Computational Storage can help your edge computing application save power and space by contacting us at sales@ngdsystems.com.

2019-02-19T09:38:28+00:00