x99 Xeons soldier on.
2x 1.5TB HGST HUSMM series SAS-3 12gbs SSDs on an HP 240 HBA (on a x99 based Dell workstation mobo

). CPU is a 14 core Xeon E5-2697 2.6/3.0 ("Haswell") with 40gb of DDR4 2133 (registered ECC). This is dedicated storage, the system is running on its own SATA SSD. Half of each disk is mirrored, half striped, using Windows LVM/StorageSpaces. The test shown is of the striped logical volume.
According to the thread instructions:
Defaults:
The NVME suite, for comparison:
These disks show up reliably on the 'bay as SAN pulls. I've never had any issues reformatting them with
sg_utils when neccesary.
Performance with more queue depth and six threads (for more parallelism, to exercise the cores):
Significant boost to RND4K, but probably not meaningful for anything but development, virtual machine hosting, keeping up w/streaming I/O, etc.
E5 Xeons and x99 boards aren't quite as old as a 386, but still not bad for a platform that's nearing a decade.
Though I'd still like to figure out why I'm not exceeding 12gbps in sequential reads despite the striped volume having drives on different SAS-3 ports and the card being PCI-E 3 x8. Maybe an issue with the benchmark? Seems too coincidental to be limited right at the SAS3 bandwidth, allowing for a little overhead.