[SOLVED] 2 M.2 NVMe SSDs in RAID_0 Boot Drive on Z390 Platform

Endre

Honorable
Hello!

Recently, I upgraded my storage from one Samsung 970 PRO NVMe 1TB SSD, to two of the same.

For anyone out there wondering if you can create a RAID configuration and boot from it, the answer is “Yes”, you can.

My system:
•Motherboard: Gigabyte Z390 Aorus Master.
•CPU: Intel Core i7-9700K.
•Memory: 32GB DDR4-2666 @1.2V.

Is RAID a “game changer” on the Z390 platform?
•I’d say it isn’t, but it works pretty well; no errors, and you obtain a double sized LPDDR cache.
•In benchmarks, I obtained a slight increase in sequential transfer speeds (Crystal DiskMark).
•You’ll “feel” a bit of increase in speed when copying huge files.

Other than that, everything’s pretty similar to as running a single drive of the same model.

PS: I hope this informations will help someone.
 
Solution
The nature of raid 0; data gets split. the split is usually between the high & low bytes - if you write 44442222 to the array, one drive will get 4444 and the other drive will get 2222, so both caches take the same hit as a single drive.

As for the sequential speeds, sorry, I thought he 970 pro could also write at 3300 as a single drive, so yes you'd see the increase there but not much in the reads. since you're maxiing out the DMI bus between the CPU & Chipset.

Endre

Honorable
So basically, this?
"You’ll “feel” a bit of increase in speed when copying huge files. "

I'm asking, just because I'm genuinely interested.

Yeah. Pretty much.
But even a slight improvement is a plus.
Let’s remember that users are enabling GPU resizable-BAR, or use XMP on their memory DIMMs which doesn’t lift the quality of their experience with much more than a RAID_0 configuration.
 

Endre

Honorable
So basically, this?
"You’ll “feel” a bit of increase in speed when copying huge files. "

I'm asking, just because I'm genuinely interested.

RAID_0 makes a bigger difference on platforms with more PCIe lanes assigned by the CPU.

On Intel Z390 platform, there are only 16 PCIe lanes assigned by the CPU, which are “taken” by the GPU.
Everything else runs through the chipset, through DMI 3.0 connection (which is equal to PCIe 3.0 x4).
So, all drives are competing for bandwidth, bottlenecking each other.
 

USAFRet

Titan
Moderator
RAID_0 makes a bigger difference on platforms with more PCIe lanes assigned by the CPU.

On Intel Z390 platform, there are only 16 PCIe lanes assigned by the CPU, which are “taken” by the GPU.
Everything else runs through the chipset, through DMI 3.0 connection (which is equal to PCIe 3.0 x4).
So, all drives are competing for bandwidth, bottlenecking each other.
In a theoretical sense, yes.

I'm still looking for actual user facing performance enhancements, before I'd jump into that swamp.
 
  • Like
Reactions: Endre

popatim

Titan
Moderator
Raid 0 on that intel chipset is not faster then a single drive, even with huge file transfers. Not even in benchmarks.
It's actually a little slower but it is not typically perceptible to the user.

You do get the "Pro's" you listed; single drive letter, larger C: drive, etc. (Btw cache is not doubled)

I'm glad you learned from this and happy to hear you have backups. Enjoy !
 

Endre

Honorable
Raid 0 on that intel chipset is not faster then a single drive, even with huge file transfers. Not even in benchmarks.
It's actually a little slower but it is not typically perceptible to the user.

You do get the "Pro's" you listed; single drive letter, larger C: drive, etc. (Btw cache is not doubled)

I'm glad you learned from this and happy to hear you have backups. Enjoy !

Actually the scores in CrystalDiskMark were a bit higher, especially the sequential writes, which increased from 2700+ to 3400+ (I could post a screenshot of that).

Question:
You said that the cache isn’t going to get doubled. But why is that?
I mean, if the storage gets doubled, isn’t the LPDDR cache going to get doubled too? (1GB LPDDR x2).
 

popatim

Titan
Moderator
The nature of raid 0; data gets split. the split is usually between the high & low bytes - if you write 44442222 to the array, one drive will get 4444 and the other drive will get 2222, so both caches take the same hit as a single drive.

As for the sequential speeds, sorry, I thought he 970 pro could also write at 3300 as a single drive, so yes you'd see the increase there but not much in the reads. since you're maxiing out the DMI bus between the CPU & Chipset.
 
  • Like
Reactions: Endre
Solution

Endre

Honorable
The nature of raid 0; data gets split. the split is usually between the high & low bytes - if you write 44442222 to the array, one drive will get 4444 and the other drive will get 2222, so both caches take the same hit as a single drive.

As for the sequential speeds, sorry, I thought he 970 pro could also write at 3300 as a single drive, so yes you'd see the increase there but not much in the reads. since you're maxiing out the DMI bus between the CPU & Chipset.

Yeah.
And I have to admit that certain benchmark cathegory numbers were a bit lower on RAID_0 vs the single drive in CrystalDiskMark.

So, how about heat?
Was I right about that, assuming that they’ll split workloads?
 

popatim

Titan
Moderator
I don't know for sure ; it depends one what happens at the sector level.
While classic R0 splits the data there is also another approach that assigns sectors to disks and the fills the whole sector. For example in a 2 disk scenario, one drive gets all the even numbered sectors, The other gets odd. So if you have 64k sectors and write a 28k file, only one drive gets written to. Write a 65k file and both drives get written too but one drive only gets 1k of the file and 63k is wasted space.

At least this is how I understand it, I've haven't found a technical paper that actually goes in depth on NVME raids inner workings.
 

Endre

Honorable
I don't know for sure ; it depends one what happens at the sector level.
While classic R0 splits the data there is also another approach that assigns sectors to disks and the fills the whole sector. For example in a 2 disk scenario, one drive gets all the even numbered sectors, The other gets odd. So if you have 64k sectors and write a 28k file, only one drive gets written to. Write a 65k file and both drives get written too but one drive only gets 1k of the file and 63k is wasted space.

At least this is how I understand it, I've haven't found a technical paper that actually goes in depth on NVME raids inner workings.

OK.

I have one other question, since this is my very first RAID configuration.

QUESTION:
Will updating the BIOS version erase my RAID volume?
(Since it’s been created in the BIOS menu).