PC Drivers & Software

[otison]

ISSUE DESCRIPTION:

               Raidxpert2 is not picking up one of the drives from what I can tell.  I have managed to create an 8-disk bootable NVME raid 0 array.  Any more than that and I’m greeted with an error message, but I can make a second array.  The sequential read speed is far too low than what I have seen in reviews and online.  I am unable to reproduce the Iometer v1.10 readings I have seen.

https://community.amd.com/community/gaming/blog/2017/10/02/now-available-free-nvme-raid-upgrade-for-...

https://linustechtips.com/main/topic/839042-der8auer-8x-samsung-960-pro-m2-nvme-in-raid-0-with-threa...

I have tried taking out the on-board m.2 drives and creating a raid with just the Asus Hyper m.2 cards enabled.  This oddly only lets me make an array with 6 drives and automatically assigns one as a global hot-spare.

Der8auer and Linus were both using the Zenith board for these tests and PRO’s not EVO’s, but their read/write speed is virtually identical.  I did notice a boost when I overclocked the Ram and CPU, but it won’t go any faster than 10 Gbps reads and writes. 

Where is rcadm.efi ???  The documentation talks about installing from UEFI shell and to contact your motherboard or system supplier for the file?

BIOS:  Have tried v.10 , .11, and .12.   I have downloaded the beta-bioses but have not tried them yet.

I can’t tell if I am missing something obvious as to why I can’t get read speeds that should be higher. 

These raid cards should be balanced between the dies (0 & 1 in the manual)

A lane or block diagram for the MSI MEG x399 creation board might help shed some light, but I can’t find anything.  I am stumped as to why I get the same performance as three nvme's in raid 0 for reads compared to eight.

Please advise, Thank you.

HARDWARE:

MOTHERBOARD:  MSI MEG CREATION X399

CPU:  Thread-ripper 2990WX

RAM:  128GB Corsair Vengeance 2667 mHZ – 

Module Part Number:

CMK128GX4M8A2666C16

GPU:  Two Titan V’s in PCIE_3 & PCIE_5 running at 8x

PSU : Enermax 1500W Gold Fully Modular PSU

STORAGE:  11 500 GB SAMSUNG 970 EVO Nvme m.2 – Eight of these drives are interfaced with two Asus M.2 Hyper Cards in PCIE Slots 1 & 4 running at 4x 4x 4x 4x and three are on-board.

Software:  AMD Raidxpert2, Windows 1809,

Cooling:  Custom Loop EKWB, ram, cpu, gpus

[daethan]

I too would like to better understand why user deployed raid 0 is far from what reviewers and amd is claiming. In fact I have yet to find a single post where an end user has achieved the read speeds of linus, deb8ur or amd. I made a post in the level1tech forms about the difficulties trying to get a raid 0 to its full potential.

https://forum.level1techs.com/t/m-2-raid-x399m/137253

Right now I get the fastest speeds when running in numa(Memory Interleaving = channel) mode with all drives on the same node rather than split as is suggested. Regardless of the config it seems to hit a read wall. In my system with 6 970 pro drives in raid 0 I can get 12 or 16GB/s reads on the same node depending on what thread is running the testing software and 14GB/s when split between nodes. I have tried a lot of different bios settings and configurations but nothing gets close to the 20ish GB/s you would expect this type of raid to top out at. The top read speed scales pretty hard with memory speed so not sure if that's a software limitation or infinity fabric or what.

If any user has any experience with an m.2 raid on tr4 I hope you can post your experience here and whatever suggestions you may have. If anyone from AMD would like to post here that would be awesome.