NVME in 2x Slot, full speed IOPs?

[Haughernaut]

Does anyone know if you stick a NVME drive in a 2x slot would the drive get full IOPs or would they be halved; even though the full IOPs is far below the rated speed of even 2x PCIE lanes? The motherboard I have has a NVME slot that is limited to 2x PCIE lanes. If I can't get full IOPs from 2 lanes then I would be willing to just buy an add in card.

[dizmo]

It would help if you listed what drive and motherboard you have.

If your drive is rated at more than 780MB/s (limit of x2), then yes, you'll lose that extra "performance."

Welcome to the forum!

[Sakkura]

7 minutes ago, dizmo said:

It would help if you listed what drive and motherboard you have.

If your drive is rated at more than 780MB/s (limit of x2), then yes, you'll lose that extra "performance."

Welcome to the forum!

The limit of PCIe x2 is either 1GB/s or just under 2GB/s depending on whether it's PCIe 2.0 or 3.0.

[Sakkura]

What SSD is it? And are you talking about read IOPS, write IOPS, mixed IOPS, what queue depth, and what transfer size?

 

An insufficient interface bandwidth will mostly hurt sequential transfers and high queue depth random IOPS. Low queue depth performance is more dependent on latency, which is not affected by a lower interface bandwidth.

[dizmo]

10 minutes ago, Sakkura said:

The limit of PCIe x2 is either 1GB/s or just under 2GB/s depending on whether it's PCIe 2.0 or 3.0.

Ah yes, I misread the article. It listed PCIe for some reason, then 2.0 and 3.0 later on.

Still, if it's 2.0 it will likely still cause a potential loss of performance...in the odd circumstance the drive is being used to its fullest.

[Haughernaut]

I'm more taking along the lines of Random 4k read/write IOPs. For example on a 960 Evo you get roughly 110-130 MBs 4k random read, this is definitely smaller than the available bandwidth of even PCIe 2.0 2x and PCIe 3.0 2x. Who cares about sequential reads and writes when user experience in Windows lies more around random 4k.

[Sakkura]

16 hours ago, Haughernaut said:

I'm more taking along the lines of Random 4k read/write IOPs. For example on a 960 Evo you get roughly 110-130 MBs 4k random read, this is definitely smaller than the available bandwidth of even PCIe 2.0 2x and PCIe 3.0 2x. Who cares about sequential reads and writes when user experience in Windows lies more around random 4k.

At low queue depth, 4K random IOPS won't be hurt much at all. Higher queue depth performance would still be affected to a certain degree.