A memory addressing and access design for real time volume rendering

This paper presents a memory addressing design that uses buffering to achieve a cache hit ratio of 95% for a PCI based Volume Rendering hardware accelerator. The target system for this memory interface is VIZARD II, a second generation PCI board using several XILINX chips. To improve the performance of this and possibly other hardware accelerators, a cubic addressing scheme is presented that improves the percentage of cache hits to misses. To further improve the performance the cubic addressing is coupled with several FIFO buffers to minimise the pipeline stalling effect of cache misses in the eight parallel memory modules. This combination of addressing scheme and memory access buffering raises the cache hit to miss ratio from 63% to 95%. Most Volume Rendering systems are fully pipelined and can utilise the design presented here to increase the number of frames per second and the quality of rendered images