Here unlike compressing data in cache in the previous slides, memory data are compressed because in GPUs, memory bandwidth is usually the bottleneck and we want to transfer as much data as possible from the memory. While when compressing data in cache, we want to store as much data as possible in cache so that the cache hit rate can increase.
Here unlike compressing data in cache in the previous slides, memory data are compressed because in GPUs, memory bandwidth is usually the bottleneck and we want to transfer as much data as possible from the memory. While when compressing data in cache, we want to store as much data as possible in cache so that the cache hit rate can increase.