There are several factors affecting speed. Make sure those lanes are on PCIE gen3 and at X4 without sharing lanes with anything else. No matter what you do the current generation of NVMe drives installed on the motherboard (not a PCIE card) are going to be bottlenecked by the DMI link between the PCH controller and the CPU which has 4 lanes. Marketing misrepresentation of the PCH lanes leads folks to believe they get 24 more lanes which is true BUT and a HUGE BUT, those are all downstream of the PCH controller and in the end they all go back to the CPU on 4 lanes.*
You may see a marginal speed increase on sequential reads but raid adds latency which will give you a hit where it counts, in the random 4K shallow depth ques. If you intend on using the drive as storage only of large media files then the slight gain will be better on paper and snythetic benchmarks but in reality you won’t be able to tell the difference. If using for an OS/apps drive you will notice the hit in the 4K speeds with things taking longer to load.
@deepcuts
That link says nothing. Just a VERY old comparison and does not mention what drives and makes zero mention of low que depths. That blog is from 2016! NVMe in its current state did not yet exist. Spinners, SATA SSDs and even the first generation of NVMe saw more benifit because their native speeds were/are far slower than the DMI link. Current drives get you 3500MB/S sequential for a single drive which is pretty much the saturation point of the DMI link. You can squeeze out perhaps another 300MB/S but what’s the point? Then add to that the destruction of your 4K performance and it’s a lose, lose situation.
*
My recommendation FWIW would be to run them as single drives. *
You can get the crazy speeds for sequential (never 4K ramndom) by using *a PCIE card that is not behind the PCH controller but the caveat is they can’t be used to boot from and you have to do software raid.*
“Two things are infinite: the universe and human stupidity, I'm not sure about the former” ~ Albert Einstein