What if all the data requests are not in one row and are just short part of locations in one row? Will this method help?
BestBunny
I believe the idea for this approach was that we send the entire row regardless of whether the entire row is requested so that if other parts of the row will be used later, the user already has it. If we only end up using a byte per row, this method won't be very efficient for sure but if data from the same row is requested, we save the cost of preparing the row buffer and activating the row.
paramecinm
This method won't help if all data accesses are in different rows. Besides, since the row size is 2K bits, will it be helpful to set the first dimension a multiplication of 2K bits when we are using two dimensional arrays?
yes
I think that perhaps the reasoning behind the design is that in most programs, data accesses tend to be contiguous and use multiple parts of the row.
What if all the data requests are not in one row and are just short part of locations in one row? Will this method help?
I believe the idea for this approach was that we send the entire row regardless of whether the entire row is requested so that if other parts of the row will be used later, the user already has it. If we only end up using a byte per row, this method won't be very efficient for sure but if data from the same row is requested, we save the cost of preparing the row buffer and activating the row.
This method won't help if all data accesses are in different rows. Besides, since the row size is 2K bits, will it be helpful to set the first dimension a multiplication of 2K bits when we are using two dimensional arrays?
I think that perhaps the reasoning behind the design is that in most programs, data accesses tend to be contiguous and use multiple parts of the row.
A more detail description for burst mode on Wiki