RND4k Q32T1 Performance Trends Opposite to Other Metrics Across Drives

[DanielDC88]

Hi, I have three drives in Windows:

• A 512GB Samsung 970 EVO NVMe
• 2x512GB Samsung 950 SATA SSDs running in RAID 0
• a 4TB Samsung SATA SSD (i forget which)
  The motherboard is an MSI B450 Mortar Max with an R9 3900X installed.

I ran Crystal Disk Mark 8.0.4 on all of them at once, and performance looks as to be expected (drives are slower as you go down the list) except in one area, which is RND4k Q32T1.

Here the single SATA SSD is fastest, followed by the RAID array then the NVMe, which is odd! Switching between the units on Crystal Disk Mark, the IOPS for that task are ~10x RND4K Q1T1 and that is 10X the sequential tests.

This seems really odd to me so I was wondering if anyone could explain why this might be the case.

[Trooper_ish]

Number of actual nand chips to run in parallel?
Or packages or whatever?

4T drive might have 4chips on, dual drives one each, and nvme single?

Just throwing a possibility out there…
The generation of chip/controller might also play a factor, with 4T newest, then nvme then dual sata, as a contributing factor?

[jode]

Connections on the mb? NVMe direct to CPU vs. chipset.
Didn’t look up the mb, but I assume the SATA ports connect to the chipset.