Does RAID 0 hurt CPU performance?

Since laptops don't have a dedicated hardware RAID controller, this means that all the operations are done with the help of the CPU right? so like in a game or a benchmark, leaving the load times and SSD RAID 0 performance aside, would it cause lower CPU performance since the CPU has to do some work for the RAID 0 array?

I will be getting a Eurocom Sky X9 (Clevp P870DM-G) which claims to have hardware RAID in their ads but in the OC forums told me if it doesn't have an actual hardware RAID controller, than that's software RAID that is done through the BIOS so it is not a pure hardware RAID

 

Yeah, there is a balance to achieve and if your workloads need constant/sustained CPU+RAM performance more than constant/sustained storage subsystem performance, using RAID0 may hurt your overall productivity.

Without a dedicated hardware RAID, as the storage subsystem gets faster, the cpu utilization increases substantially too. More work = more sweat for the platform overall (and that always means for the CPU+RAM).

A proper RAID controller will have it's own RAM cache, a battery backup system to prevent data corruption and a dedicated cpu highly optimized for the different RAID levels it supports - in addition to at least a half dozen available ports to connect drives to. This needs a TDP as much or more than most notebooks sold today. For these reasons, a 'real' RAID controller is not offered on any mobile platform I know of (or at least; not one that I would put too much faith - or cash - in).

If your workloads and overall workflow demands maximum, sustained performance of the CPU+RAM combo, having RAID enabled may actually make you less productive depending on the actual work that is being produced by the platform.

If you are worried about a workflow with demands of anything less than that (of the CPU+RAM) and are also not running multiple NVMe x4 SSD's in that RAID0 array***, I would stop worrying now.

So, what is the intended workflow? How sustained is the load on the CPU+RAM? What concrete benefits will the RAID0 array bring? And is this multiple SATA3 drives or NVMe x2 or x4?



*** Assuming that those multiple NVMe x4 SSD's were properly cooled and not throttling in the first place, of course.

 

Thanks a lot for your detailed responses as usual.

Workflow is browsing the net 80% of the time, watching movies 10% of the time, and playing games 10% of the time

The only reason why I might have wanted to have RAID 0 is to make full use of the space as in, create a RAID array of the two 512GB 950 PROs to almost 900GB, C: gets 200GB for the OS + programs and D: gets 700GB for the games and rest goes for OP. That way, I get more space in D: for the games

If I were to use the SSDs as a single drive then each SSD gets a single 400 GB partition, you see, this way my D: space went down from 700GB to 400GB (due to OP as well)

The negatives though as you confirmed taxing the CPU and the fact that the 4K Random Read/Write speeds becomes lower by around 7-10% using the IRST Driver vs. the Samsung NVMe driver

Yes they are PCIe x4 NVMe SSDs.

CLEVO/SAGER claim on their website they have hardware RAID but as the guys on OC forums said, that is just a marketing gimmick because unless as you said you had a dedicated RAID controller, it can in no way be hardware RAID

See:
http://www.sagernotebook.com/Notebook-NP9870-S.html

 

First off: OP'ing... 67% of the actual capacity offered would be closer to less than ~640GB of actual storage in RAID0 or not. With 200GB for C:\Drive, that leaves 'only' ~440GB for games. Is 33% OP'ing overkill?

No. Not if you want to get as close as possible to the theoretical performance you paid for your storage subsystem. And even then it will be far below reasonable (mine) expectations.

Secondly: getting a 45% increase in Sequential performance with a ~10% decrease in Random R/W performance is not worth risking your setup to the fickleness of RAID0. How reliable/dependable and available do you want this system to be? If these expectations are high, RAID0 is not worth considering - even for a larger D:\DATA drive.

Thirdly; if optimum setup of your platform includes the biggest capacity possible for the D:\DATA drive; consider a 1TB or (better) a 2TB SATA3 SSD instead and a single NVMe x4 SSD for your O/S + Programs + 33%OP'ing for all.

Using separate drives, the platform is optimized as the O/S can do what it wants on it's drive and the DATA can be manipulated as needed on it's drive concurrently.

With your workflow, this would be beyond overkill; but if you still want the second NVMe x4 SSD... I would set it up as either the Program drive or the Scratch Disk/Temp drive. But again; OVERKILL for your workflows.


Not only is a +45% and a -10% difference unacceptable - it also highlights that two or more drive bays does not a hardware RAID make... Especially when that +45% is an 'Up to' marketing 'promise'.


Myself, for your workflows? I would put in two 2.5" SATA3 SSD's instead. An SanDisk Extreme Pro 480/960GB for the O/S+Programs drive. And the same drive or a 2TB Samsung 850 EVO for the DATA drive. Yeah; this is how much I fear that the NVMe x4 drives will throttle not only themselves, but the rest of your components too.


To answer your question; RAID0, in your workflows, will not detract significantly*** from your overall performance if having the largest D:\DATA drive is a high priority.

But that doesn't mean it is the only or even the best option to pursue.

Good luck.




*** Again; assuming that no throttling is taking place in the NVMe x4 SSD's or any other component they'll affect with their extreme heat output.

 

Thanks a lot bro! you are as I always say, the SSD Doctor!

Cheers

 

RAID-0/1/10 has almost zero CPU usage. RAID-5/6 CPU usage is negligible. For instance, an i7-2600 can process 12,000 mega*bytes*/second of RAID-6 (ie: double-parity) writes.

Code:

[    0.048136] smpboot: CPU0: Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz (fam: 06, model: 2a, stepping: 07)

[    0.286071] raid6: mmxx1     5621 MB/s
[    0.342726] raid6: mmxx2     6059 MB/s
[    0.399370] raid6: sse1x1    4781 MB/s
[    0.456024] raid6: sse1x2    5799 MB/s
[    0.512677] raid6: sse2x1    9619 MB/s
[    0.569330] raid6: sse2x2   11748 MB/s
[    0.569416] raid6: using algorithm sse2x2 (11748 MB/s)

IMHO, its totally silly for anyone to be buying custom ASICs to do RAID-6 these days.

 

I second this approach. Using different drives for different purposes is wise and flexible practice.

 

RAID O does not hurt CPU performance. There is almost nothing to calculate. Even with using RAID levels that use parity bits, on modern processors the CPU usage is minimal.

 

Not true, the "almost nothing to calculate" ends up being 4.5% reduction in single core performance.
(near the end where the guys compares all the various combos).

RAID 0 definitely has a performance impact, you have to weigh if you want a small dip in processing power for a massive increase of disk i/o.

 

I benched [email protected] GHz in my BGA KILLER laptop. Tested raid 0 vs single ssd in Wprime 1024M stress test and Cinebench R15. No difference between raid 0 vs. single SSD in these benchmarks. Perhaps you wiil notice slightly with a weak low powered Intel BGA processor, but not with something that is much more powerful than this low entry trash

 

Were you also stressing the storage subsystems at the same time though?

 

exactly, a CPU test only means that storage sub system is not being taxed, thus, even if there was a little performance hit as the CPU is doing the RAID operations, you won't see it in such a benchmark unless the storage was being tested at the same time.

 

It's been some time since I checked RAID performance. RAID 0 splits (or stripes) the data onto mulitple drives. When the computer reads it back, you get two storage devices loading into RAM instead of one. It should only boost performance unless the devices are stepping on each other due to poor bus design. RAID 1 mirrors two storage devices with identical data in case one fails. If I recall, writing is about the same speed. Reading from RAID 1 could be faster since a clever operating system would read half of the file from both devices, assuming the RAM is connected well with the persistent storage. There is no checksum or parity to compute for RAID 0 or 1; error correcting codes are for the higher number RAIDs.

For laptops, I'd question the assumption of independance of drive failures too. Be sure to back up your data onto something other than your laptop!

 

Unless you have some really strange usage patterns, leave RAID on and forget it.

Like people pointed out above, you will only see the CPU performance impact when your storage subsystem is stressed. But when you reach the point where a single SSD can be stressed, chances are your I/O-heavy app will be I/O bottlenecked with RAID off anyway. Suffering from a overall system performance lose means I/O load and CPU load need to be in a ratio just right so you're not benefiting from the I/O throughtput boost but do lose significant CPU time from soft RAID. This kind of situation is easy to create in a benchmark (by running multiple unrelated apps), but in real life usage where people tend to do one thing demanding at a time, it's almost never relevant.