x399 Taichi, NVMe drive stuck at x1 speed

Arundor,  what have you used to measure the speed?  Looks like to me that it should be Gen3x4 like my X399 FPG.  All three of my 960 EVOs show Gen3x4 and test around 3.0 MBps.  Your M2_3 also supports SATA3 6.0 Gb/s.  Or you just reading the Gen3x1 of running a performance test?  Have you tried another socket?  Please post your specifications in your signature as I have, including BIOS and OS versions and power supply.  Enjoy, John.

MisterJ and parsec, thanks for the replies.

Edited by Arundor - 14 Nov 2017 at 9:11am

Arundor wrote: MisterJ and parsec, thanks for the replies. I tried remounting the drive as suggested and it appears the standoff and come a little loose.  I tightened it and remounted the drive, and the performance has improved.  Here is the speed I am getting now: https://i.imgur.com/f5Qv5oj.jpg  I'm still a little concerned about the 4KiB Q32T1 speed being lower than what many other owners of this drive have reported so I will continue trying to improve my configuration to address that.  But for now, I think I can say that the original problem is solved. Oddly, Samsung Magician still reads the link speed as Gen 3x1, however the benchmark I linked above shows that the drive is achieving speeds greater than would be possible at x1 speed so I'm content to conclude that Magician is in error. For completeness I'll fill in the other details I was asked about in case anyone else with a similar problem finds this thread. - Operating system is Windows 10 Professional version 1709 (Fall Creators Update) - I tested drive speed using the built in performance tests in Samsung Magician, and also the 3rd party CrystalDiskMark software. - Only one PCIe slot is in use, a GTX 760 video card is in the top x16 slot. - I tried altering the options in the AMD PBS section of the UEFI/BIOS. The options were Auto/Gen1/Gen2.  I didn't see a Gen3 option but I tried setting it to Gen2 since that was the highest option available, and it didn't help with the problem. I have since set it back to Auto.

Great, glad that was a simple fix. The M.2 connections are so close together that alignment and fitment are critical. The screw position insures the drive is pushed in all the way.

The Magician software might not be quite compatible with X399 at this point, which might account for the error in the link bandwidth. There is no chance that you are running at PCIe 3.0 x1 with your Sequential Q32T1 read and write speed results.

The AMD PBS options have Auto being Gen3 (PCIe 3.0) and I would leave it on Auto. For an experiment you could try setting it to Gen2 or Gen1 and see what the Magician software shows for the PCIe bandwidth, if you trust it.   I'm curious if those options affect the M.2 slots.

You think you have poor results for the 4K Q8T8 results? Here's my 500GB 960 EVO on a Ryzen X370 system:



The 4K Q8T8 test is new, I'm curious to try it on my Intel PCs with the same and other NVMe SSDs.

IMO, the 960 EVO can be erratic in its benchmark results. But on Ryzen boards we have seen consistently lower 4K high queue depth speeds for some reason. As MisterJ said, use as an OS drive affects benchmark results considerably. To demonstrate both the 4K and OS difference, a 960 EVO on an Intel Z270 PC, empty data drive:



Comparing speeds on the two X399 PCs, we have a 250GB SSD vs a 1TB SSD. The 1TB 960 EVO has a much larger SLC cache, so I would expect it to do better.