FPGA-based CSP accelerator module meets standard for hyperscale servers
Designed to meet the new Open Compute M.2 Accelerator Module standard intended to operate in Glacier Point carrier cards for Yosemite servers, the 250-M2D accelerator module has been released by enterprise-class NVMe computational storage company, BittWare.
The 250-M2D accelerator module is an FPGA-based computational storage processor (CSP), designed for feature-rich, dense servers, such as Yosemite, are favoured by hyperscale and cloud companies for their performance density and energy-efficiency in machine learning platforms.
The 250-M2D features a programmable Xilinx Kintex UltraScale+ FPGA directly coupled to two banks of local DDR4 memory. It can be programmed by customers developing in-house capabilities, or delivered ready-to-run and pre-configured, with application IP from Eideticom, a specialist in computational storage.
The accelerator module can be purchased directly from BittWare’s parent company, Molex. There is also a version specifically for recommendation models, complete with software for easy integration from deep learning frameworks, available from Myrtle.ai.
BittWare is a Molex company. It provides enterprise-class compute, network, storage and sensor processing accelerator products featuring Achronix, Intel and Xilinx FPGA technology. The programmable products increase application performance and energy-efficiency while reducing total cost of ownership, says the company.
BittWare has 30 years’ experience developing FPGA accelerators and claims to be the only FPGA vendor-agnostic supplier of critical mass able to address enterprise-class qualification, validation, lifecycle and support requirements for customers deploying FPGA accelerators in high volumes.
Molex has a presence in more than 40 countries. It offers a full range of connectivity products, services and solutions for markets that include data communications, medical, industrial, automotive and consumer electronics.