Are the basics of HBM the same as that of DRAM? i.e., when an LLC miss request goes to DRAM, the whole 64B block corresponding to that address is fetched from DRAM. But that block is transferred in 4 transactions(16B at a time, because channel width is 16B). In HBM memory, a requested block of 64B will also take 4 transactions or it will be transferred in 1 transaction only?