Are all the banks duplicates of each other? That feels like a significant amount of redundancy. Can anyone with experience or knowledge shed light on whether this trade off has been worth it?
yey1
My personal guess is that it is not a replication, but a portion of memory.
kevinle1
If that's the case, then a scheduler would have to identify memory accesses that are in different banks and then pipeline them together right in order to get this kind of high pin utilization right?
hanzhoul
That is to say, throughput will be higher if one program tries to access data from different DRAM banks, right?
Are all the banks duplicates of each other? That feels like a significant amount of redundancy. Can anyone with experience or knowledge shed light on whether this trade off has been worth it?
My personal guess is that it is not a replication, but a portion of memory.
If that's the case, then a scheduler would have to identify memory accesses that are in different banks and then pipeline them together right in order to get this kind of high pin utilization right?
That is to say, throughput will be higher if one program tries to access data from different DRAM banks, right?