Samsung will manufacture the drives, but Xilinx will sell them as part of its data center portfolio. Powered by Xilinx FPGAs, the SmartSSD CSD is the industry's first adaptable computational storage platform providing the performance, customization, and scalability required by data-intensive applications.

Xilinx will showcase the SmartSSD CSD and partner solutions at the Flash Memory Summit Virtual Conference and Expo taking place November 10-12. 

Flexible programmable storage platform

The SmartSSD CSD is a flexible, programmable storage platform that developers can use to create various unique and scalable accelerators that solve a broad range of data center problems. It empowers a new breed of software developers to easily build innovative hardware-accelerated solutions in familiar high-level languages. The SmartSSD CSD accelerates data processing performance by 10x or more for applications such as database management, video processing, artificial intelligence layers, complex search, and virtualization.

"Processing data efficiently is the key to unlocking its value," said Pej Roshan, vice president of marketing, Data Center Group at Xilinx. "The SmartSSD computational storage drive empowers a broad new range of developers to harness the benefits of computational storage and quickly deliver compelling applications to solve a host of thorny data center challenges. From transparent compression to next-gen AI inferencing acceleration, the range of functions performed on the SmartSSD CSD is limited only by a developer's imagination." 

"The industry is beginning to realize just how much the SmartSSD CSD will be able to boost performance in the datacenter and far beyond, and with the latest Xilinx tools for application development we anticipate dramatic growth in a wealth of acceleration applications", said Jim Elliott, corporate senior vice president of Memory Sales and Marketing at Samsung Semiconductor, Inc.

Calculations on SSD not CPU

Incorporating a Xilinx FPGA accelerator, the SmartSSD CSD pushes high-speed computation closer to where the data lives, bypassing server CPU limits. The reduced data movement lowers latency and power consumption for accelerated data processing speed and efficiency. The SmartSSD CSD allows enterprises to keep up with user demand for faster access to data without being forced to purchase more servers, driving up CAPEX and OPEX. Additionally, host CPU resources are freed up to handle other higher-level tasks more efficiently.

The Xilinx Adaptable Platform

The SmartSSD CSD platform is designed to enable the easy creation of custom applications by harnessing the Vitis unified software platform and accelerated libraries. Both developers and enterprise appliance vendors can quickly go-to-market with pre-built customized applications.

Runtimes, libraries, APIs, and drivers can be built into the system using common high-level languages such as C, C++ and OpenCL. 

In addition, Xilinx has launched a high performance, easy to deploy data encryption solution built on dm-crypt. By leveraging standard Linux kernel modules, SmartSSD CSD developers can utilize Vitis libraries to create turnkey acceleration applications that scale effortlessly across multiple SmartSSD CSDs.

Turnkey Accelerated Applications

Xilinx ecosystem partners have developed a range of turnkey accelerated applications that deliver accelerated data processing for services like big data analytics, video transcoding and search-in-storage.

Representative accelerated solutions, include:

• Big data/data analytics – SmartSSD CSD with Bigstream's hyperacceleration layer allows customers to increase time-to-insight by up to 10x on the Apache Spark unified analytics engine for big data.
• Search-in-storage – Lewis Rhodes Labs (LRL) applies novel neuromorphic processing with the SmartSSD CSD to find "needle-in-a-haystack" data from large data lakes by searching petabytes in minutes. LRL's deterministic processing accelerates queries up to 100x faster than other solutions.
• Video file transcoding – CTAccel's pre-built SmartSSD CSD solution performs high performance image processing that improves server throughput and latency while reducing up to 60 percent of video transcoding costs.
• Transparent compression – Eideticom's NoLoad NVMe CSD enables up to 10x more storage capacity, increased line-rate compression performance, improved CPU utilization, and reduced memory bandwidth pressure.

Search Petabytes in minutes

NPUSearch-in-Storage is one of the official partners that already have a prepared solution for the market. The company used twenty-four 4 TB NVMe SmartSSD CSD drives in a 2U rack and powered them on with Dual AMD EPYC 7282 and 64GB RAM. NPUSeach also used two M.2 480GB SSDs (RAID 1) for boot drives.


The results they managed to achieve are nothing short of extraordinary. Compared to traditional search with a CPU NPU search in storage using Xilinx's FPGA.

The company demo ½ Rack with 240 Smart SSDs getting the total storage just shy of 1PB. NPUSearch-in-Storage finished searching 1 PB within 25 minutes maximal time and in some optimal searches even faster. CPU would take dramatically more time. A smart storage solution can scale linearly while the CPU maxes out and stays flat with performance.

Fast fixed-throughput regular expression-based search.NPUSearch leverages novel insights into how the brain processes data and places this processing on the  FPGA right next to storage.

Drive Specifications and Availability

The SmartSSD CSD is a standard 2.5-inch (U.2) form factor solid-state drive in a low 25-Watt power envelope. It offers a physical storage capacity of 4TBs, with the ability to hold up to 12TBs of data or more when combined with accelerated transparent compression. The integrated Xilinx Kintex UltraScale+ FPGA accelerator features more than one million system logic cells and nearly 2,000 DSP (digital signal processing) slices for hardware acceleration.

SmartSSD CSDs are available for pre-order today and will begin shipping with general availability in January through Xilinx and its authorized distributors, by agreement with Samsung.