论文部分内容阅读
在高性能计算等应用领域,FPGA器件的复杂程度越来越高,已经进入16核、500多万逻辑单元的级别,DSP也可以达到10 TFLOPS的性能,但是,在实现高性能计算的路途上仍存在一个瓶颈,这就是数据的传输速率,虽然处理单元的性能和速度足够强大,但如果数据从存储器到处理单元的传输速度跟不上,运算性能也会大打折扣。目前,在高性能计算的应用中,FPGA和存储器之间的数据传输速率就是一个短板。最近,Altera公司宣布了一种集成HBM2DRAM和FPGA的异构SiP器件,在单个封装中高效地集成了FPGA和宽带存储器管芯,突破
In high-performance computing and other applications, FPGA devices become more sophisticated, has entered the 16-core, more than 500 million logic levels, DSP can achieve 10 TFLOPS performance, but in the road to achieve high performance computing There is a bottleneck. This is the data transfer rate. Although the performance and speed of the processing unit are strong enough, the computational performance will be greatly reduced if the data transfer speed from memory to the processing unit can not keep up. At present, in high-performance computing applications, the data transfer rate between FPGA and memory is a short board. Recently, Altera Corporation announced a heterogeneous SiP device that integrates HBM2 DRAMs and FPGAs to efficiently integrate FPGAs and broadband memory die in a single package, breaking through