Will a second m.2 nvme drive bottleneck performance on my setup?

[theNEOone]

I currently have a single 950 pro and will be replacing it with a 960 pro. I'm thinking about keeping my "old" 950 and running it off an m.2 nvme x4 adapter since my mobo only has one m.2 slot. Will there be any bottleneck in my system w/ the two drives? I keep hearing about "lanes" and there being a limited number of them...

Here's my config, excl. the nvme drive(s)

• Asus Maximus VIII hero
• 6700k @ 4.7gz
• 32gb (4x8gb DDR4 3000)
• gtx 1070
• 512gb 850 evo
• 1tb Sandisk x400

=|

 

[Valantar]

I'd put it in the black (x4 electrical) PCIe x16 slot on the board, but it's possible that slot is deactivated when you use the m.2 slot. Read your manual. If the slot is disabled, you could put it in one of the other PCIe x16 slots, but that will reduce your GPU to running at x8 speed. You probably wouldn't be able to measure the difference, let alone notice it. Should be fine.

 

[13Gigatons]

Funny that we are already running out of bandwidth. The new Zen CPU only provides 24 lanes total and 16x are used for the GPU. Maybe Zen 2 will double that.

 

[Valantar]

Funny that we are already running out of bandwidth. The new Zen CPU only provides 24 lanes total and 16x are used for the GPU. Maybe Zen 2 will double that.

The introduction of PCIe storage has ensured that PCIe bandwidth is at a premium. Has been for years. OTOH, switches and lane splitting works fine. The latest Intel PCH is essentially a PCIe 3.0 switch with some other parts tacked on. GPUs perform ~identically at 16x and 8x speeds, so splitting off lanes for storage is no problem. In edge cases, such as transfering sequential data from one PCIe SSD connected off the PCH to another connected off the PCH (if that's at all possible?), you might see a bottleneck as the link between the PCH and the CPU is effectively PCIe 3.0 x4 itself. (OTOH, it might be that drive-to-drive transfers through the PCH go point-to-point, in which case you'd not see any performance drop. I don't know.) For day-to-day usage, even with heavy desktop workloads, you'll never notice.

 

[theNEOone]

I'd put it in the black (x4 electrical) PCIe x16 slot on the board, but it's possible that slot is deactivated when you use the m.2 slot. Read your manual. If the slot is disabled, you could put it in one of the other PCIe x16 slots, but that will reduce your GPU to running at x8 speed. You probably wouldn't be able to measure the difference, let alone notice it. Should be fine.

This worked, thanks. Everything working as expected.

=|

 

[BonzaiDuck]

This worked, thanks. Everything working as expected.

=|

I've become very slow and deliberate as work on what seems like a perfect PC ties up loose ends. I had been thinking about this for a while. I have a 960 EVO 256GB I can test it with. One must either limit SLI and CF possibilities, occupy an x4 slot that could conceivably have alternate usages or requires disabling something like a pair of SATA ports,, fret about your remaining SATA-III ports that might share bandwidth with the M.2-in-motherboard, your need for an ODD, hot-swaps, etc.

I figured with the speed bump from NVMe, I wouldn't care so much about SATA SSD performance on a PCIe x1 Marvell controller, and that all seems to work well.

I'd put it in the black (x4 electrical) PCIe x16 slot on the board, but it's possible that slot is deactivated when you use the m.2 slot. Read your manual. If the slot is disabled, you could put it in one of the other PCIe x16 slots, but that will reduce your GPU to running at x8 speed. You probably wouldn't be able to measure the difference, let alone notice it. Should be fine.

I read a review and test that concludes the difference for a single VGA card between x16 and x8 is about 1% performance.

I'll just pick the stock clock in Afterburner and start testing at the tweaked profile.

Anyone have any insight about that? Overclocking with card running in x8 as opposed to x16? I haven't a clue. It just seems intuitively worth some attention.