a bit delayed reply to old comment here.. :
x9dri-f and samsung nvme 970 evo plus. Guys, very low speed crystaldiskmark read/write RND4K Q1T1. Does everyone have it? Is this a physical limitation of the bus?
Unfortunately It´s NOT so strange low speed for me.. I´ve got 2 old Supermicro board - X9SRH-7F (C602 chipset) and very old C2SBA+ II. (G33 chipset). I tested them on low-cost Axagon PCEM2-ND (
https://forums.servethehome.com/ind...furcation.31172/page-3#lg=post-353379&slide=0) , which is x8 electrically but due to limitation of its ASMedia self-bifurcation chip only PCIe x4 in real...and only PCIe Gen2... And when I tested it with inserted Samsung 970 Evo Plus NWMe M.2, the speed was only 2,2-3x better comparing to my another Samsung 860 Pro SSD SATA disk (connected to onboard SATA3 or 2 respectively
). Tested through AS SSD and ATTO benchmark (could share results).
I know it´s a bit funny setup and results, especially on ancient C2SBA+ II. with Core2Quad Q9550s (socket 775 CPU)
I think such bad speed results are due to not-so-good ASMedia bifurcation chip, due to its PCIe x4 limit and of course bottlenecked by CPU and MB chipset. But still it should give you some imagination towards running NWMe M.2 SSD disks on PCIe cards on old systems...
For me the strange is another situation. I always check CPU utilisation during read and write speed on NWMe 970 Evo Plus. And even on ancient C2SBA+ II and even under Win7x64SP1
, the % CPU utilisation of C2Q was still relatively low. So someone should presume, that if CPU is still NOT fully loaded, than why Axagon PCEM2-ND with 970 Evo Plus is not much faster ? chipset speed limitation etc. ?