One SSD Vs. Two In RAID: Which Is Better?

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
This articles is build for people that use just their computer for Internet and few applications.
I think the problem is here. If you build a RAID 0 in first place, it's not because you want to run Word and IE, you want fast PERFORMANCE with gaming and file transfert speed!

On RAID 0, you don't care that your Windows 7 take 0.1 seconds more to start (load the RAID driver take more time than if you don't need it) because the point is having PERFORMANCE with application/game that required an extra boost. None of the test really point out this.

It's like give a proof that 32bits memory addressing run faster than 64bits memory addressing. This sure that 32bits memory addressing is faster if you need less than 3.2gb of memory :) (when most pc have more than 4gb...). Sound pretty much what travish82 point out "People believe what they want" and it's worst when they don't understand the context or don't define any. Context should be "Why I need to use a RAID 0 and where I have advantage to use it?". Same as all technologies. You just point out few drawbacks of RAID 0 without showing any advantage that RAID 0 is built for.
 
However, RAID is a great way to go if you you have some SSDs that you initially purchased for small operating systems (linux, xp) or for boot drives and you would like to repurpose them.

The MTBF is high and the average gamer would be hit with an inconvenience if a SSD failed - many, not all, people store mass data on disks still, I store my mass data on a NAS.

Additionally, this enables expandability with a performance increase, however marginal. My 60 GB SSD is fine for now, why should I buy a huge SSD when I could simply double my storage for a fraction of the price when I need to?

Last point. I don't notice a difference in a computer shutdown in 3, 4, 5, or 6 seconds; or a game loading in 10 seconds compared to 15. However, RAID would show me the difference in Photoshop, video editing, CAD, and other large-data processing/serving compared to a single drive.
 
[citation][nom]americanbrian[/nom]I would like to see a video transcoding benchmark. I know that the author mentions RAW video editing, but it would be nice to see some real world encoders being used with some results. A lot of people do edit/transcode videos nowadays and I suspect the gains would be sufficiently large that it may well be worth the small price differential even if you only transcode stuff occasionally...[/citation]I have an Asus P9X79 with an i7 3930K and 32GB RAM. I initially built it with (2) 120GB OCZ Vertex 3's in a RAID0 on the Intel SATA3 RAID controller. I was hitting a max speeds over 1GB/s. This was built for video production. During the times I had to do video file conversions, the CPU had the biggest impact (along with software that would support multiple cores), but this shouldn't really come as any surprise.

I have since switched to a single Samsung 840 PRO 256GB. And there is no difference. And I didn't really expect there to be much difference. Standalone, the drive still pushes 500MB/s on sequential read/writes. I don't expect any DVE or NLE workstations to be able to process more than the read/writes of a capable single SSD on a 6Gb controller.

And I'll tell you what...the primary reason I switched to the single drive? I was constantly getting parity errors. I got burnt twice with the RAID0 failing. So I then switched to a RAID1 setup. I figured that while my storage capacity just got cut in half, having my data saved in the event of a catastrophic drive failure was more important. Unfortunately, I was still getting parity errors. (I spent months troubleshooting this on various tech forums...OCZ, ASUS, Intel, where ever.) Eventually I dropped the idea of a RAID altogether and went with the Samsung 840 PRO. System has been pretty much trouble free since then. And the upside is now I have two Vertex3's I was able to repurpose to some aging laptops.

So yeah, this article by Toms simply confirms what has been said for years and what I knew deep down was a concern. Single drive failure ruins everything. And quite frankly, if I have data that's not important, I'm going to store it on slower HDDs and not on costlier (per MB) SSDs. I'll save my SSD for the most important, mission-critical things like OS, applications, and immediate data needs.

On a side note (re: threadjack), y'all need to check out the SK Hynix line of SSDs on NewEgg (or where ever). The SH910's are, based on my benchmark results, the Strontium Hawk series. (I haven't cracked open the cases to confirm.) The 120GB drives are giving me better max speeds than my Sammy 840 PRO 256GB drive! Sometimes you can even find the SH910's for $90. Definitely worth a look!
 
Its quite logical that access time will get some overhead from the RAID logic, and all the benchmarks tied to access times will see no improvement from going to RAID.
Probably, with an AMD mobo, things would get worse... and with a dedicated enterprise class RAID card things should improve.
BUT THE WORST THING IS NOT HAVING TRIM SUPPORT ON YOUR SSDs. I know Intel decided to add suport from trim to RAIDs, but when I looked at (some months ago) they only supported spare disks (not in a RAID volume, but accessed through BIOS RAID mode) on last chipset.
Then you have to rely on Garbage collection and / or overprovisioning...
I knew this, and went for a 2 x 128GB Vertex 4 RAID 0 (I already had one of them attached on a SATA/2 interface, which I substituted for a cheaper Kingston, which is able to saturate SATA/2: The result was a 256GB RAID 0 array 1€ cheaper than a 256GB Vertex 4...
I know there are other issues related to RAIDing SSDs (you need to take the disks out of RAID mode to flash them with new firmware...) and probably I will end breaking down the array: but 120GB is enough for a boot disk, so I probably will upgrade another computer to SSD sometime...


 
Evolution2011 and americanbrian: Transcoding, exporting or editing have a really small impact on a SSD. They'll never max out your SSD disk output speed (around 500mb/s). So RAID 0 is useless (1000mb/s) for those softwares.
Thoses tests is a burden for your CPU. It's also why you don't see that kind of test in this article.

You can run transcoding and export on normal HD and you still won't see a difference.
SSD will speed up your "preview screen" (during editing) and it's why SSD is prefer in those softwares.
 
I'd like to comment on the people who argue that 2 smaller SSD's is cheaper than one big one. One of the characteristics of SSDs is that the larger the drive, the faster it performs. So you might be spending a bit more for the same capacity, but you'll be increasing your system performance as well, without any of the downsides of Raid 0.
 
Im thinking the performance numbers may be bound to the chipset platform in the sense that some perform better then others.. in this situation id say just go with a pci express ssd and call it a day but that's just me.
 
I run 4x Vertex 4 256GB SSD in RAID0 with my LSI SAS 9211-8i controller in my gaming rig, I'm dishing out a whopping 2gb per second read/write, now that's called awesome, you kids still got a long way to go.
 
[citation][nom]americanbrian[/nom]I would like to see a video transcoding benchmark. ...[/citation]

I was going to post a lengthy comment, but Evolution2001 pinched the words right out of my mouth. 😀


[citation][nom]Evolution2001[/nom]I have an Asus P9X79 with an i7 3930K and 32GB RAM. I initially built it with (2) 120GB OCZ Vertex 3's in a RAID0 on the Intel SATA3 RAID controller. ...[/citation]

Doubly freaky dude, I have a very similar system. 😀 P9X79 WS, 3930K, 64GB, initially a 120GB
MAX IOPS but now a Samsung 840 250GB (might replace it with a 256GB Vector I recently obtained).
Quadro 4000, 3 x GTX 460, etc.


The article mirrors my own SSD RAID tests:

http://www.sgidepot.co.uk/misc/ssd_tests.txt

The benchmark numbers look great in places, but the real-world tests show that what matters
is having any kind of SSD setup. SSD-RAID0 is indeed slower in some cases. Note these tests
deliberately used a P55 board to provide data to those who keep asking about upgrades for
older systems.


wittermark, I managed to get a lot more than that with 8 x Vertex3 120GB MAX IOPS using an
HP P410. 😀

Ian.

 
Lest ye not forget that unless you are setting up RAID with SSD's on the latest intel chipset that support RAID level trim for your SSD's you will lose that feature. In which case your SSD's will eventually lose their performance and drop back down to merely fast hard drive status, based on their use and amount of free space.
 
I have 2 64GB vertex 3 in raid0 and I have seen real world performance increases in daily tasks of about 10-15%, plus now instead of 2 spereate 64GB drives i have 1 128gb drive
 
I used two 64gb Samsung 830s in RAID 0 with Windows 8 for over a year...until performance degraded no matter what I did to try to help them along. I migrated over to a single 128gb 840 Pro, and now those 64gb drives just hold VMs in my test lab.

If you buy two small drives and just want them combined to one big storage space, Windows 8 has this feature...I can't remember what it's called...oh right it's called Storage Spaces. It will do this for you without the pitfalls of RAID 0.
 
This article is right on time as I've been wondering about putting RAID 0 SSDs in my Asus G75VW. Looks like just getting two 512GB drives and using a 3GB of cache (Fancy Cache) will be a better option.

Great article, right on time, and highly appreciated.
 
warezme WRITES:
> ... In which case your SSD's will eventually lose their performance and drop back
> down to merely fast hard drive status, based on their use and amount of free space.

Not necessarily, it depends on the model of SSD used. Some drives have very good
garbage collection characteristics. This is especially true of the Vertex2E, a model
which was optimised for surviving without TRIM because when it was new many users
were still on XP. Hence, it's a good model for UNIX systems aswell; I have three in my
SGI Fuel, performance is if anything actually better than when first installed 2 years ago.

Some models are quite different though, as you say their garbage collection is poor.
All depends on the model/firmware.


chumly writes:
> So what we need is:60 GB Boot Drive. 256 GB Game/App Drive. ...

Sure, that kind of split is quite common, though these days I don't see the point of using
anything less than a 120GB/128GB for the boot drive, given the current pricing, and there
usually is a useful extra performance edge from the 120s vs. 60s. (or 128s vs. 64s). If you
don't mind the hunt, for the cost of a new 60GB one can easily bag a used 120GB off eBay
these days (I've obtained loads of them, most recently another 120GB MAX IOPS for only
64 UKP total).

My gaming PC has a 120GB boot drive (Vertex3), 120GB game data drive (Vertex3 MAX IOPS)
and a 1TB Samsung F3 for general data & backup.


> 512 GB x 4 RAID 5 Storage Drive.Right?

I would recommend RAID10 rather than 5, but if it's for storage rather than speed (though
one will get some extra speed anyway), it makes more sense to use normal HDDs for this,
eg. 4 x 2TB RAID10 would be quite good.


Ninjawithagun writes:
> I strongly disagree with the findings within the article. I've been running my main gaming rig for

Not sure that comparing to a laptop is quite fair... differences in CPU, RAM, I/O paths, etc.


> ... For me, an SSD RAID 0 configuration is the way to go as I utilize the full benefits it provides. ...

Just curious, what do you do for backup? I use Macrium Reflect to create backup images files on
a 1TB SATA. If the boot drive SSD should fail (120GB Vertex3), I can then restore the backup
image onto a completely unused & safely stored away SSD of the same model.

Btw, I suspect your gaming setup behaves nicely for coping with multiple players, etc.,
because the RAID0 is likely being hit with a more threaded load, in which case the reads
will flow much better. I saw this effect with my own testing of up to 5 x 120GB Vertex2E
on a good P55 board:

http://www.sgidepot.co.uk/misc/ssd_tests.txt

Scroll down to the section entitled, "Asrock P55 Deluxe (RAID0/10 via Intel RST 11.6.0.1030,
64KB Stripe, 4K Cluster; multiple OCZ Vertex2E 120GB)". AS-SSD low queue-depth 4K read
doesn't increase that much (little parallelism) but 4K/64thrd scales very nicely, giving 511MB/sec
for 5xV2E/120. Or if one should desire some protection, then even 4xV2E/120 in RAID10 gives
a very respectable 375MB/sec for 4K/64thrd read; it's probably this behaviour that allows your
system to behave they way it does. As you say, it depends on one's task at hand. Someone who
was doing something that involved a lot of small random reads with no threading would observe
little gain from SSD RAID0, eg. processing pattern data for textile manufacturing (thousands of
small files involved in this).

4xV2E RAID10 loses some write performance vs. RAID0, but it's still good. Naturally, the results
with SATA3 models tested on later chipsets would be even more impressive (I'll test this later -
I have 8 x Vertex3 120GB MAX IOPS, various different X58/P67/Z68/X79 boards).

Ian.

 
[citation][nom]Kamab[/nom]Putting them in RAID0 doubles your chance of data failure, aka either drive fails and you probably lose everything.[/citation]
This isn't true at all, failure chance doesn't work like that: IE; compounding percentages. Yes the chance of failure is higher, but the chance of failure for a RAID 0 of SSDs is lower than a chance of failure in a standard HDD. That said, there's only one reason anyone whether they RAID 0 or not, wouldn't have an enterprise -or at least a WD Red- HDD for backup purposes; and that is if they can't afford it yet.

Now let's talk odds of failure here. "How do percentages compound together?" you may be asking. It's actually very simple, but you may want to crack out the calculator if you can't run the numbers in your head. Let's say SSD (A) has a failure rate of 10% within 5 years (which is extremely high for SSDs, so it's an example ONLY). So, it's obvious that ordering 10 of them, odds are high one or more may fail within 5 years. How do these odds stack up though? Instead of considering failure-rate, consider non-failure-rate which is 90%. Now let's consider two drives at that rate: 90% x 90%. "Wait, what?" You're likely asking. "Where did that formula come from?" Well, it's the formula for compounding percentages. Just like in RPGs that use multiple % increases/decreases for particular stats like chance to critical strike. Only in this case, it's getting scaled to a percent instead of a base number. Moving on; 90% x 90% = 81%. "Wait, what? How did you calculate that?" Well, instead of actually using percentages, just use decimals: 0.9 x 0.9 = 0.81 and convert back to percentages. Simple.

Now you're probably thinking, "It's close enough to double the failure rate, why argue semantics?" And my response to that is that it scales with more drives in the array. If it truly doubles and at a failure rate of 10% within 5 years (again, exaggerated example) then ten drives would promise a failure in five years, but random odds don't work like that. Yes if you flip a quarter 10 times, odds are you'll get 5 heads and 5 tails, but not all the time. Saying RAID 0 doubles chance of failure is a technoob term that makes you sound like a know-it-all wannabe. RAID 0 SSDs will still have a lower failure rate than single drive HDDs, so what's the point mentioning it even? "You may lose your data..." Which should be backed up regularly anyway.

Anyway, the clear and concise winner in these tests is to RAID 0 two drives. If you want 512GB of high-speed storage, get 2x256GB and RAID 0 away. You'll see no real performance hit compared to a single 512GB, and have the perks of the RAID 0 when your system can actually take advantage of it. Now, that's not taking into consideration other factors mentioned by other commentors, such as lack of TRIM support, which gamers may care about. However, even they are unsure of that claim as they leave out details and don't firmly believe certain points. I am aware of TRIM alternatives though that negate that claim too. People, please do your research on things before posting what you believe. My post was to clarify how failure chance scales.
 
RAID 0 has never been about performance in the real world. As Jack pointed out on page 1, RAID0 has ALWAYS posted disappointing scores in real world tests, and only really shows promise in high end corporate applications where data integrity makes RAID0 useless in the first place.

BUT, there is a case for RAID0, and it is the reason I use it in my system; My initial 240GB drive was too small for my system drive, and so it was much cheaper to get a 2nd 240GB SSD than to get a larger 480GB drive (especially over a year ago when 480+GB drives were insanely priced compared to smaller sizes). Having 2 240GB SSDs in RAID0 gave me an upgrade path, made the capacity affordable, and does break the 500MB barrier on those rare occasions when real world performance actually gets full performance out of the drives.

However, if I was building my system today, there is no question I would get a single 512GB drive. It is more reliable, more appropriately priced, and a simpler setup. It just makes sense now.


All that said, I will never have a rig without RAID1 ever again. Completely saved my butt last year when I had a drive failure! Recovery with the onboard Intel controller was seamless! I literally only had to plug the new drive in, everything else was automatic and in the background while I used the system. Ya, it sucks to buy 2 drives for the space of 1, but it is worth it as a relatively safe place to put important stuff.
 
3 years ago I built my system basically to capture uncompressed AVIs files, with only an HDD -- it was quite disappointing.

Then I got a SSD -- it is faster, but not as I´d expected.

From what I understood in my particularly case RAID 0 is the way to go, eh?
 
When I put together my current machine, I chose an SSD for the boot drive (plus "important" software that would rarely, if ever be updated), and two 2TB hard disks in a RAID 5 configuration for "everything else" (the idea being to minimize SSD writes -- I tend to keep my computers a long time).

The SSD performance was nothing short of startling. Applications opened and loaded in a fraction of a second. But... the HD RAID wasn't far behind. It was subjectively almost as fast. Because all my "user data" are on the RAID HD, the SSD could not have performed so well unless the HD was reasonably quick.

My suggestion is a single SSD for the OS and "large" applications that need to load quickly, with an HD RAID system (for security) for smaller apps and user data (especially for e-mail and browser temp files).
 
Thanks for this review Toms! I would have liked to see a comparison with say OCZ SSDs. Any idea why the Samsung SSD's have such low read speeds? 240 mb/s is pretty slow in comparison.
 
[citation][nom]jemm[/nom]... From what I understood in my particularly case RAID 0 is the way to go, eh?[/citation]

What kind of video? SD 4:2:2? HD 4:4:4:4? Serial 10bit? 2K? 4K? 8? More details please!

Btw, multiple 15K SAS on a SAS RAID card with battery backup gives excellent general
performance, while cache RAM means small reads/writes are fast aswell. Or there's
Enterprise SATA.

Ian.

 
Hello everyone, I am going run16 Samsung 840 Pro SSD's in my system and I am looking for a really good raid card that can handle 16 drives. Are there any suggestions? Thanks.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom