Gidel’s new strategy for Accelerator Boards
21 June, 2016
Developed an accelerator board based on Altera’s Arria10 FPGA. Gidel’s technology drove the FPGA database revolution
The Israeli company developed an accelerator board based on Altera’s Arria10 FPGA, with competitive pricing aimed for small companies. Gidel’s technology was crucial for the FPGA database revolution
Israeli Gidel developed accelerator board dubbed HawkEye. The newly developed component is based on Altera’s programmable Arria10 FPGA component. The new board family supplies up to 8 PCIe channels for up to 56 GB/s, an up to 18 GB DDR4 on-board memory and and MTBF of more than a million working hours. Alongside the board, Gidel supplies to different developer kits – OpenCL based Altera SDK, and Gidel’s own Gidel Proc, enabeling the simultaneous acceleration of several application at the same time, as well as the optimization of HDL level development.
One of the most interesting points in Gidel’s announcement of the its new product line is pricing: the boards are sold for less than $1,000 in entry level, and for less $2,000 for the high end versions of this product line. The competitive pricing policy is part of Gidel’s new strategy – a move from big customers requiring massive support by the company, to a large scale distribution of components to price aware customers.
The brain of Replay Technology
Gidel was founded in 1993 by CEO Reuven Weintraub. The company employs 30 workers. It began its way as a FPGA solutions developer and ASIC service provider, and soon developed into a supplier of FPGA based boards, active in two main fields: computer vision and high performance computing (HPC).
Weintraub told Techtime said that most of the company’s sales are to global markets, although the company does have some Israeli customers requiring high performance computing such as Orbotech, Camtech and several defense companies. One of Gidel’s most renowned customers is Replay, purchased this March by Intel. Replay developed the FreeD system for capturing sporting event using multiple cameras, producing a 3D scene that is capable of producing replay from every possible angle.
In Replay’s method, multiple 5K cameras (the highest pixel per mm cameras available in the market) are placed in the stadium. A soccer match for example requires 21 such cameras. The camera feed is gathered from the cameras in real time, producing a 3D representation of the Stadium and the Information. The technology relies on extremely high computing capabilities – for cameras as well as for the server center processing the various feeds. Intel has previously reported that the process is carried out by an Intel server cluster, but now we learn that these server are accelerated by Gidel’s boards.
Why are you implementing a new strategy?
Weintraub: “The FPGA market is growing fast, and we have to act faster than before. IBM started using FGPA to accelerate the performance of its data centers, Microsoft running its search engine Bing with the aid of FPGA components, and Intel took over Altera for its data center strategy”.
In fact, the first attempt to use FPGA boards to accelerate data bases was conducted as part of the CHREC project in the University of Florida in 2010. 448 FPGA were than integrated into servers, achieving a 1,000 performance improvement. The groundbreaking project was awarded many international awards, as was the virtual starting point of the FPGA data center revolution. For Gidel, this was a major success, as the project was based on its FPGA boards. Another major customer is Altera itself – the company uses Gidel’s components in some of its quality control systems.
Why FPGA accelerates Data Centers?
“The industry discovered that FPGA accelerates analytics performance, because channel width (number of bits) can be accurately adjusted to the exact number of bits required for the specific processing. Instead of allocating 8, 16, 32 or 64 bits to a task requiring less, just because of hardware restrains, FPGAs allocates the exact number of bits. The basic principle is simple: using less bits to process a given task, increases processing speed. FPGA can achieve this easily and simultaneously for many tasks in parallel.”