Sign in with
Sign up | Sign in
One SSD Vs. Two In RAID: Which Is Better?
By , Achim Roos,
1. Are Two SSDs Any Better Than One?

In a few short years, SSDs went from exotic high-end hardware to a staple in most performance-oriented PCs and notebooks. As we already know, that's due to a few things. First, they're tremendously faster than hard drives, and orders of magnitude more responsive. Second, the price per gigabyte of NAND flash continues dropping thanks to advanced manufacturing and economies of scale.

Today, many 128 GB drives sell for less than $100. Stepping up to 256 GB often gets you even better pricing on a per-gigabyte basis (almost always under $1, and sometimes as low as $.60 per gig). 

Now SSDs are at a point where enthusiasts have to make an important decision: do I buy one large SSD or do I grab a pair of smaller drives and stripe them in a RAID 0 configuration? This is largely an issue of performance. We know that single drives are hitting the limits of SATA 6Gb/s. So, are any of the workloads you're running that are taxing enough to necessitate the throughput (particularly in sequential transfers) two SSDs working cooperatively can deliver?

We're setting out to answer that question by testing the performance of modern SSDs at different capacities. Samsung sent over a sextet of drives: two 128 GB 840 Pros to test against a 256 GB model and two more 256 GB drives to test against a 512 GB version.

If you want more information on the 840 Pro, check out our launch coverage: Samsung 840 Pro SSD: More Speed, Less Power, And Toggle-Mode 2.0, along with our last experiment with these things: Is A SATA 3Gb/s Platform Still Worth Upgrading With An SSD?

2. Benchmark System And Software

We used the same benchmark suite found in our SSD Charts to measure and compare the performance of different single- and multi-drive configurations of Samsung's 840 Pro. First, we tested the drives on their own: one 128, 256, and 512 GB SSD alone. Then, we then put two 128 GB and two 256 GB SSDs in RAID 0, sending each array through our benchmark suite.

Real-World Benchmarks

  1. Booting up Windows 8. The clock starts when the POST screen vanishes and stops when the Windows desktop appears.
  2. Shutting down Windows 8. After Windows 8 runs for three minutes, we shut it down and start the clock. The clock stops once the system powers off.
  3. Booting up Windows 8 and Adobe Photoshop. After Windows 8 boots up, a script starts the image editor Adobe Photoshop CS6 and loads a photo with a resolution of 15,000x7,266 pixels and a size of 15.7 MB. Once this is complete, Adobe Photoshop is closed. The clock starts after the POST screen and stops when Adobe Photoshop closes. We perform this benchmark five times.
  4. Five applications. After booting up Windows 8, a script starts five different applications. The clock starts when the first application launches and stops when the last one closes. We perform this benchmark five times as well.

Script for the Five-Application Benchmark

  • Load a Microsoft PowerPoint presentation and then close Microsoft PowerPoint.
  • Start the Autodesk 3ds Max 2013 command line renderer and render a picture with a resolution of 100x50 pixels. The picture is so small because we’re benchmarking the SSD, not the CPU.
  • Start the built-in ABBYY FineReader 11 benchmark and convert a test page.
  • Start the built-in MathWorks MATLAB benchmark and execute it once.
  • Start Adobe Photoshop CS6 and load the same picture used in the third real-world benchmark, but in the original TIF format with a resolution of 29,566x14,321 pixels and a size of 501 MB.


Benchmark System

Benchmark System Hardware
HardwareDetails
LGA 1155 PlatformAsus P8Z77-V Pro, Chipset: Intel Z77 Express, BIOS: 1805
LGA 1155 Processor
Intel Core i7-3770K (22 nm, Ivy Bridge, D2), 4C/4T, 3.5 GHz, 4 x 256 KB L2, Cache, 6 MB L3 Cache w/ HD Graphics 4000, 95 W TDP, 3.9 GHz max. Turbo Boost
Dual DDR3 RAM2 x 8 GB DDR3-1600 CL10-10-10-27 (Corsair Vengeance CMZ16GX3M2A1600C10)
SSD System Drive (I/O and General Performance Benchmarks)Samsung 840 Pro, 256 GB, Firmware DXM04B0Q, SATA 6 Gb/s
Benchmark Drive SSDs
Samsung 840 Pro, 128 GB, Firmware DXM04B0Q, SATA 6 Gb/s

Samsung 840 Pro, 256 GB, Firmware DXM04B0Q, SATA 6 Gb/s

Samsung 840 Pro, 512 GB, Firmware DXM04B0Q, SATA 6 Gb/s
Power Supply
Seasonic X-760, SS-760KM Active PFC F3
Benchmarks
General Benchmarks
h2benchw 3.16
PCMark 7 1.0.4
I/O Performance Benchmarks
IOMeter 2006.07.27
Fileserver Benchmark
Webserver Benchmark
Database Benchmark
Workstation Benchmark
Streaming Reads
Streaming Writes
4K Random Reads
4K Random Writes
Real World Benchmarks
3ds Max 2013
Finereader 11
Matlab 2012b
Photoshop CS6
Powerpoint 2010
System Software and Drivers
Software/DriverDetails
Operating System
Windows 8 x64 Pro
Intel RST11.7.0.1013
3. Results: Sequential Read And Write Performance

As expected, the RAID 0-based setups, consisting of two 128 GB and two 256 GB SSDs, solidly beat the single-drive configurations in our sequential read and write test. In fact, the two striped 256 GB drives provide about twice the performance of the single drives. The two 128 GB SSDs aren't quite as strong in sequential writes, which we might have predicted based on the throughput numbers gleaned from one 128 GB 840 Pro.

According to Samsung’s specs, the 128 GB 840 Pro should be expected to post less read and write performance than the larger models. So, we're getting consistent results.

4. Results: 4 KB Random Read And Write Performance (AS-SSD)

At the low queue depths you normally encounter in a desktop environment, all of these configurations perform fairly similarly. In fact, the striped setups are even a bit slower than single drives. This is because we're taxing the NAND flash's throughput. Parallelism is needed to distribute the workload across multiple dies on multiple channels.

Once we jump up into very high queue depths, both RAID-based arrangements distinguish themselves. It's only a shame that this is very atypical of any desktop workload, so you won't see it unless you take the 840 Pros into a more enterprise application.

5. Results: 4 KB Random Read And Write Performance (Iometer)

Our Iometer benchmark demonstrates the link between queue depth and random read and write performance very well.

The bar charts below are indicative of average performance from a queue depth of one to 32. As you can see from the top measurement, 4 KB random reads, the two striped configurations are already dominating. Writes aren't as clear of a victory, though. The 512 GB 840 Pro manages to outmaneuver the two 128 GB SSDs, which themselves are only slightly faster than the single 256 GB drive.

A look at performance scaling over multiple queue depths is even more telling, though. Reads again appear bound by the NAND at low queue depths, only scaling in favor of RAID 0 once we apply plenty of parallelism. Writes are far more taxing, and two 256 GB drives barely beat one 512 GB SSD, while two 128 GB 840 Pros are just slightly faster than a 256 GB drive.

6. Results: Access Time

Perhaps a result of overhead, all of the single-drive configurations offer lower access times than striped arrays. The differences are extremely subtle, though.

7. Results: I/O Benchmark Profiles (Iometer)

Again, the bar charts represent average performance from a queue depth of one through 32 in three benchmark profiles: database, Web server, and workstation.

The two RAID 0 arrays manage to clearly draw ahead of the single drives across all queue depths in Iometer’s Web server benchmark profile. Can anyone guess the access pattern in play? That's right, 100% reads. Could have seen that one coming, right?

However, the database and workstation profiles convey less scaling at queue depths up to eight. Up until that point, the Samsung 840 Pro 512 GB performs about the same as the two 256 GB SSDs in RAID 0. The same is true for the single 256 GB SSD and two striped 128 GB SSDs.

8. Results: PCMark 7 And PCMark Vantage

The synthetic PCMark 7 and PCMark Vantage benchmarks demonstrate almost identical performance for all of the SSD setups. There are more notable differences in the individual workloads, but they seem to cancel each other out.

PCMark 7

PCMark Vantage

9. Results: AS-SSD Copy Benchmark And Overall Performance

Copying files is one way to take advantage of fast storage, SSDs in RAID included. In our three copy benchmarks, two fast SSDs working cooperatively overcome the limits of a single SATA 6Gb/s interface, pushing more throughput than any single drive can manage.

Overall Performance

The RAID 0 arrays win both the AS-SSD overall score and desktop performance crowns based on superior performance in a number of synthetic metrics. But not so fast: the real-world benchmarks paint a different picture. Blazing-fast peak sequential transfers don't necessarily translate into better real-world numbers.

10. Real-World Benchmarks: Booting Up And Shutting Down Windows 8

Our Windows 8-based benchmark system boots up fastest with the single 256 GB 840 Pro, followed by the 512 GB model. The two RAID 0 arrays place third and fourth, while the individual 128 GB drives brings up the rear. However, the difference between first and last place is only 1.1 seconds.

When it comes to shutting Windows 8 back down, the twin 128 GB SSD RAID array does best. Then again, it's only ahead by a slim margin. The difference between first and last place is a miniscule 0.4 seconds.

11. Real-World Benchmarks: Booting Up Windows 8 And Adobe Photoshop

Our third real-world benchmark provides us with pretty much the same results as the first two. There’s practically no difference between the SSD-based setups, whether you're looking at a single drive or two in RAID 0. In this one, we're booting Widows 8, starting Adobe Photoshop CS6, and loading a picture.

12. Real-World Benchmarks: Five Applications

It’s déjà vu all over again for our fourth and last real-world benchmark. We launch several applications after Windows 8 boots up. The different SSD-based setups all perform about the same, with a very narrow win going to the 512 GB 840 Pro.

13. RAID 0: Great For Benchmarks, Not So Much In The Real World

They're a funny thing, SSD benchmarks. You can run synthetics all day long and create these unrealistically demanding workloads that make solid-state storage look one way. Then, you can tinker around with real-world metrics that paint another picture entirely.

For enthusiasts, the truth often lies somewhere in between. A majority of the tasks we perform do tend to involve basic operations like opening Web browsers, editing images, composing email, and watching video. But sometimes we do need big performance from our systems: compiling a big project, moving tens of gigabytes of media files, or capturing uncompressed AVIs for FCAT analysis. In those instances, you want responsiveness on demand.

As we expected, twin SSDs in RAID 0 post phenomenal numbers when we hammer them with sequential reads and writes. Twin 256 GB 840 Pros nearly hit 1 GB/s in both disciplines. Largely as a result of the SATA 6Gb/s interface, single drives max out at a little more than half of those numbers.

The RAID-based configurations undoubtedly scored the first touchdown given exceptional sequential results, but the game didn't end there. Individual SSDs regained ground in the tests that followed, even posting better scores in some of them. Random I/O performance is a good example. Striped drives are certainly better equipped to push more IOPS, but only when you're stacking commands more than four high. Jumping up to a queue depth of 32, 16, or even eight is really uncommon in a desktop or workstation environment. As a result, the performance differences are far less pronounced in the real world.

One SSD on its own scores again in the contrived tests we put together. The performance differences when we boot up and shut down Windows 8, then fire up different applications, are marginal at best and not noticeable in practice. Single drives actually manage to outperform the striped arrays some of the time, even.

If you're planning an upgrade and want to know whether to buy a couple of 128 GB drives and put them in RAID 0 or just grab a single 256 GB SSD, for example, the answer still seems clear enough to us: just grab the large drive and use one. Using Samsung's 840 Pros as an example, a pair of 128 GB drives will run you $300 on Newegg right now. The 256 GB model sells for $240 (maybe that's why it's out of stock currently). There's also the issue of reliability. If one drive in a RAID 0 configuration fails, the entire array is lost. At least for a primary system drive, one SSD on its own is safer.

There are of course exceptions. SATA 6Gb/s currently limits us to 500+ MB/s reads and sub-500 MB/s writes. Sometimes, that's just not enough. Just take those raw AVI captures mentioned earlier as an example. We use four Crucial m4s in RAID 0 to make sure we aren't dropping any frames. In a case like that, RAID 0 is a must-have, and the fact that only captured video resides on the array means that a failure would be a fairly superficial loss (except the cost of the drive). If you have an application like that, well, then you already know what you need, and you know that a large, single drive isn't going to get the job done.