I recently bought three of Micron’s 15 TB versions of this drive second hand on eBay, as they were much cheaper per TB than new Samsung QVO 8TB SATA drives (seemingly the largest in the consumer space at the moment).

My server is only PCIe 3.0 with the Dell R720 NVMe cage having a PCIe 2.0 switch in it, so I can only get around 1.5 GB/sec read/write from each drive. Weirdly running ZFS across all three doesn’t give me much of a speed boost, only going a tiny bit faster than a single drive. Using hdparm on three drives at the same time lets me go over 4 GB/sec so there’s some other bottleneck in ZFS that is even slower than the PCIe 2.0 switch. It’s not encryption or compression as it’s the same speed whether they are on or off (but significantly higher CPU usage when they are on). I do see one CPU core pegged at 100% during ZFS transfers though so there appears to be something in there that can’t run in parallel across multiple cores.

So at least in my case with my old server, I can just barely saturate a 10 Gb network link with ZFS and NVMe drives, but if I went up to a 25 Gb NIC in the same machine, I’d have to switch to a different filesystem with Linux software raid in order to saturate the 25 Gb link.

Seems like the RAID controller would be a massive bottleneck. That’s the only reason I mentioned ZFS above. To bypass any hardware bottlenecks.