Hardware RAID or NVMe storage?

H/W RAID controllers do not require system resources to perform RAID parity calculations for redundant arrays; they employ onboard processors and cache memory to offload all RAID parity-related operations from the host system. However, conventional hardware RAID I/O processors are designed for use with SAS or SATA HDD and SSD devices, which generally have a maximum transfer rate of 1100MB/s (12Gb/s SAS SSD’s). Most hardware RAID arrays are comprised entirely of hard disk drives, which perform at a much slower rate (200MB/s max.).

Current generation NVMe SSD’s (PCIe 4.0, aka Gen4) support transfer rates 6-7 times that of the fastest 12Gb/s SAS SSD. To put this in perspective, a single M.2 SSD is capable of outperforming an entire array of SAS SSD’s. Today’s commercially available hardware RAID architecture is unsuitable for NVMe configurations; NVMe hardware RAID requires more considerable processing and caching resources. Outside of a small number of very specialized applications, NVMe hardware RAID is impractical and cost-prohibitive.

It’s also impossible to ignore the fact that today’s computing platforms provide ample system resources. Your average entry-level server motherboard now features one or more multi-core CPUs, terabytes of memory, and up to 128 lanes of direct to CPU PCIe connectivity. An order of magnitude more than what was available 5-10 years ago when Hardware RAID controllers were the storage solution of choice. Modern server platforms can easily accommodate NVMe RAID configurations without impacting other system-related activity. Current generation AMD-based systems offer even more potential due to their adoption of the PCIe 4.0 interface.