Linear throughput is the wrong way to think about this. Random access is much more important for DRAM workloads. Persistent memory has a block size of 4 cache lines (256 bytes) vs 4KB for an SSD. This drastically cuts down on read/write amplification.
