SCIMA: a novel processor architecture for high performance computing
Masaaki Kondo, Hideki Okawara, Hiroshi Nakamura, Taisuke Boku, Shuichi Sakai · 2000
Clock acceleration and ILP extraction have drastically improved processor performance. However, the performance of memory has not improved as much as that of processors, thus the problem of performance gap between memory and processor, which is called the memory wall problem, is becoming very serious. This will become more serious in the near future. We propose a new VLSI architecture which can solve this problem. Our architecture integrates software-controllable memory into the processor chip. We have evaluated our architecture on the Linpack benchmark and QCD simulation. These results indicate that our processor achieves high performance by using on-chip memory effectively.