The new XD2000F accelerator module targets financial data analytics, data communications system design, bioinformatics, and video transcoding applications. It plugs into servers via the HyperTransport technology bus at 800MT/s to significantly speed compute-intensive computer processing.
The HyperTransport bus offers a high-bandwidth, low-latency interface for other processors and memory in dual, quad, and even octal platforms. Placing the XD2000F in one or more of the CPU sockets gives the In-Socket Accelerator a high-speed link to memory as well as a fast connection to the host processor without requiring board modifications, while providing a simple hardware integration path.
The module was created for system designs requiring high memory bandwidth by implementing two separate DDR2 interfaces to the motherboard as well as four additional on-board QDRII memories, each with separate address capabilities. The result is substantially greater memory bandwidth locally, overcoming a typical system bottleneck.
"The combination of the XtremeData XD2000F In-Socket Accelerator with the Quad Core AMD Opteron processor allows our ACTIV MPU to keep its competitive advantage when it comes to low latency and overall throughput," said Mike Dunne, chief technical officer at ACTIV Financial. "We continue to be very satisfied with the benefits that XtremeData technology brings to our ActivFeed MPU platform. Hardware acceleration is key to our architecture and our customers are enjoying the new benefits of the ACTIV MPU performance increase."
Additionally, the XD2000F maintains full compliance with the space and cooling requirements for AMD Opteron processor sockets. By adhering to physical design requirements, XtremeData's modules can make the unique claim that they fit into any AMD processor socket, regardless of the manufacture.
"The combination of AMD's native quad-core processor and XtremeData's XD2000F module will allow our mutual customers in the financial, high-performance computing, and bioinformatics industries to achieve amazing compute power and performance levels," said Doug O'Flaherty, division manager of Acceleration Strategies at AMD. "Partners like XtremeData are taking advantage of the open Torrenza initiative and are clearly pushing the co-processing envelope on AMD platforms."
When XtremeData co-processor modules are installed into commercial off-the-shelf (COTS) servers, blade systems, and ATCA form-factors, the algorithmic parts of applications are substantially accelerated via parallelization and pipelining. The XD2000F provides high-performance application acceleration ranging from 10X to 100X compared to processors alone, while simultaneously reducing overall system power consumption and overall system latency. It is the ideal solution for accelerating applications such as Algorithmic Trading, Market Data, Deep Packet Inspection, Bioinformatics, and Video Transcoding.
"The XD2000F delivers up to five times the memory bandwidth over previous generation products," said Jim Benbow, chief hardware architect at XtremeData. "We designed the module with two FPGAs, one as a bridge, the other running the application. This industry first design allows users to reconfigure the application FPGA in-system, without rebooting the server and lets system designers focus on getting their code working in the application FPGA without having to be concerned about low-latency system interfaces."
The XD2000F is pinned out and ready accept Altera's Hardcopy II Structured Asic family, so when customers want to move to high volume deployment they have an easy, cost-effective migration path.