The Memory bottleneck problem

Several computer applications today face memory constrains, sometimes due to limited memory capacity (not enough memory), sometimes due to limited memory bandwidth (memory too slow).

How we solve the problem

Most data requested by the FPGA kernel from memory share similarities, trivial and non-trivial. We have invented novel techniques to detect and keep track of such similarities to utilize memory capacity and memory bandwidth significantly better. Our technology is lightning fast and increase effective memory bandwidth and capacity on-the-fly.

How we can add value

Several applications today are performance limited by memory bandwidth and capacity: In financial trading it is paramount to calculate derivative prices before your competitors, requiring the best memory bandwidth to access historical prices in order to determine future prices. In many machine learning applications it is often critical that the Neural Network can process large volumes of input data fast enough to make reliable and accurate real time decisions. In several server applications huge volumes of metadata concerning stored information is kept in memory. Such server applications often require Terabytes of DRAM memory.