SSD cache wars: Intel vs. Corsair storage accelerators

Home OfficeFeatures
by Joel Hruska, 09 Aug 2012Features

When solid-state drives first broke into the consumer market, there were those who predicted the new storage format would supplant hard drives in a matter of years thanks to radically improved performance.

In reality, the shift from hard drives (HDDs) to SSDs has thus far been confined to the upper end of the PC market. For cost-conscious buyers and OEMs, the higher performance they offer is still too expensive, and the total capacity is insufficient. If you don’t create much data or install many programs, a cheaper 128GB SSD may be perfect for your needs. People who work with large projects or file archives, however, can’t always downsize gracefully. If the entire point of upgrading was to improve data access rather than boot/launch times, then using a secondary hard drive for data storage defeats the purpose of the SSD.

SSD cache drives have emerged as a means of addressing this gap. They are small, typically containing between 20 to 60GB of NAND flash, and they’re paired with a standard hard drive. Once installed, drivers monitor which applications and files are accessed most often, then those files are cached on the SSD. It can take the software a couple of runs to start caching data, but once this process is complete, future access and boot times are significantly enhanced.

In theory, this type of solution offers the best of both worlds. Intel began offering its own caching software when it released the Z68 chipset in 2011 – and, leveraging Nvelo’s Dataplex software, Corsair, Crucial and OCZ have all launched cache drives of their own. We’ve tested both solutions head-to-head in order to answer three questions:

1. What type of performance boost can users expect from an SSD cache drive?

2. How do the Dataplex and Intel solutions perform head-to-head, and what are the salient characteristics of each?

3. Is overall performance improved by using a faster mechanical HDD?

If you’re curious about a cache drive, I highly recommend you don’t skip over the “Dataplex and Intel SRT” section below. Dataplex and Intel Smart Response Technology (SRT) are not perfect substitutes for one another. They aim to accomplish the same goal, but have very different restrictions and capabilities. Depending on the age of your system, you may only have one option.

The Software: Dataplex and Intel SRT

Intel first offered Smart Response Technology as a free download with the Z68 chipset, and included it again with the Z77. It’s not included as an option with the X79, despite that solution’s premium price tag, though some vendors offered their own solutions. SRT can be installed to any SSD, but requires at least 18GB of storage space and maxes out at 64GB. If you plug in a drive with a larger capacity than that, SRT will lop off part of the drive for its own use.

Smart Response Technology also offers users the chance to switch between Enhanced and Maximised modes. Enhanced mode accelerates data reads, but not writes. The advantage of this mode is that you don’t have to worry about an unexpected failure disrupting a cache update. The disadvantage is that the performance increase is much more modest and confined to read tasks. Seagate’s Momentus HHDs (hybrid hard drives) use a read-only cache. Microsoft’s Ready Boost technology, which is integrated into Vista and Windows 7, is also limited to accelerating reads.

Maximised mode accelerates both reads and writes. That’s the only mode Dataplex currently offers, so that’s the setting we tested for this review.

Integrating SRT into an existing OS installation isn’t as easy as it is with Dataplex. SRT only works with the southbridge set to RAID mode in BIOS. Switching from IDE to RAID is typically no problem for the operating system, but if you installed the OS using AHCI, you’ll need to tweak some registry settings or reinstall the entire OS in order to make the change. Once you’ve done one or the other, getting the cache up and running is simple and straightforward.

Intel Smart Response Technology strengths:

1. Offers a range of configuration options – users can change the cache size and choose between two acceleration modes

2. Works with any SSD from any vendor

3. SRT software is a free download

4. Tools like Intel’s SSD toolbox are still useable (if Intel SSD is used)

Intel Smart Response Technology weaknesses:

1. Only available on Z68 and Z77 chipsets

2. May require registry edits/manual driver swaps

3. Relies on Microsoft’s default AHCI drivers

Now, we turn to Nvelo’s Dataplex.

Unlike SRT, Dataplex supports the full range of Intel and AMD chipsets stretching back into the Core 2 era. Nvidia chipsets are not supported. Dataplex told us that it originally intended to support Nvidia chipsets as well, but had no end of trouble with nForce’s… unique SATA drivers. Given that NV hasn’t updated its nForce drivers in over a year, Nvelo had no choice but to drop the chipset.

Even with that omission, Dataplex supports a much wider range of chipsets and hardware than Intel’s SRT. The caveat, in this case, is that Dataplex is locked to certain drives. That’s less of a downside than you might think, however, for two reasons. Firstly, the drives themselves are typically priced competitively relative to their capacities (with one exception, which we’ll address shortly). Secondly, multiple companies have jumped on this bandwagon. We tested a Corsair product, but Crucial and OCZ also have cache drives available as well.

Dataplex is easier to install than SRT and doesn’t require you to adjust any BIOS or driver settings. Like SRT, it can be used to cache a RAID 0 or RAID 1 array. The program’s downsides are its lack of any sort of software toolbox/information kit, and the DRM that’s required to use it. Each SSD comes with a specific product key printed on the back. That key has to be entered when you install the software. You can move the drive from system to system, provided you uninstall the software first, and if you change more than two components in your rig, you’ll have to re-validate.

The DRM is annoying, but the company deals with user concerns promptly. At one point, when the drive hadn’t reset the key properly during these tests, I contacted Corsair to ask for a manual reset. My code was reset in minutes.

Nvelo Dataplex strengths:

1. Drop-in, dead simple upgrade to any system

2. Compatible with full range of Intel and AMD-based motherboards

3. No need to switch between IDE, RAID, or AHCI mode

Nvelo Dataplex weaknesses:

1. Requires specific hardware

2. DRM/online registration required

3. No adjustable settings or control options

4. Only caches boot drive, won’t cache drives bigger than 2TB

Is your data safe?

A number of users have expressed concerns about the possible loss of data in the event of a sudden hardware failure. Both SRT and Dataplex synchronise their write data aggressively. This means it’s uncommon for critical data to be “stuck” on the SSD with the HDD in an unbootable state. If the cache drive detects errors or problems, including an interrupted shutdown/startup, it immediately resynchronises the cache and temporarily disables itself.

The best way to ensure data isn’t lost in the event of a power outage is to follow Microsoft’s own advice and use an uninterruptable power supply (UPS). Neither Dataplex nor SRT play nice if you decide to yank out the SSD without first uninstalling the software.

The “one exception” we mentioned on the previous page as far as pricing is concerned is OCZ’s Synapse Cache family. OCZ is the only company to explicitly offer 32 to 64GB cache drives with 50 per cent of the flash blocked off for provisioning. These drives are typically priced as though they were full-size; a 32GB drive with 32GB of overprovisioning is priced as though it offered 64GB of storage. Without substantially more data it’s impossible to say if OCZ is helping users by providing an enormous amount of replacement cache, or is loading the Synapse drives with less reliable NAND to begin with.

So, all that said, let’s move on to the meat of the matter: Which caching system is faster?

The drives

We paired Intel’s SRT technology with one of the company’s SSD 330 Series, and Dataplex with a Corsair Cache Accelerator. Both of these 60GB drives actually contain 64GB of flash (the extra 4GB is used for over-provisioning). Also, they both use the LSI Sandforce SF-2281 controller, but the Corsair Cache Accelerator is an asynchronous design while the Intel 330 uses synchronous flash.

This can have a significant impact on real-world performance; synchronous flash transfers data on both the rising and falling edge of a clock, while asynchronous flash is timed to an external (slower) signal. Whether or not this matters for a cache drive is something we’ll examine. The 330 also supports SATA 6G, though the drive may not be fast enough to benefit from it.

Because Intel’s SRT is tied to the chipset rather than a specific SSD, it’s difficult to directly compare costs. However, the basic price of the drives is practically the same – the 60GB Corsair Cache Accelerator that we tested currently retails at £51, and the Intel 330 Series is priced at £53, not a bank-breaking difference.

Our mechanical drives are an older 1TB Caviar Black and a new 1TB VelociRaptor. The Caviar Black stands in for typical 7200 rpm performance with 32MB of cache, no Advanced Format (the drive uses older 512-byte sectors rather than the newer 4K standard) and no SATA 6G support. The VelociRaptor is currently the best performing consumer hard drive on the market.

Testing the accelerators

We tested both products with an Asus P8Z77V, 16GB of RAM, and a Core i7-2700K. Performance was measured using PCMark 7, and also in a variety of real-world load/copy scenarios. In the graphs below, we’ve split results by colour and drive family. The top bar, in blue, shows the performance of the Intel 330 SSD when it’s used as a standalone drive. The VelociRaptor results are in green, the Caviar Black results are in orange. The last bar of a given colour is the drive’s baseline performance without a cache accelerator solution.

PCMark 7 encompasses a number of tests that depend on storage performance as a secondary factor, as well as a specific suite of storage tests. Here’s the overall score first:

The VelociRaptor is 14 per cent faster than the Caviar Black in PCMark 7′s overall performance test. Adding a cache drive makes a larger difference – the VelociRaptor picks up 22 per cent when paired with the Intel 330. Adding a cache drive to the Caviar Black boosts it 41 per cent. Intel SRT and Dataplex tie here; the VelociRaptor is slightly faster with SRT, the Caviar Black is better with Dataplex. The 330 alone is just 1.1 per cent faster than the VelociRaptor and 330 combination.

Switching to the storage subsystem suite, we see rather different scaling. Adding a cache drive still improves performance more than moving from the Caviar Black to the VelociRaptor, but the VelociRaptor is a full 30 per cent faster than its older cousin. Again, we see the two cache mechanisms effectively tie, and again, the HDD and cache drive configurations almost equal the performance of the 330 alone.

Boot and app load times

Both Intel’s SRT and Nvelo’s Dataplex cache data based on previous access analysis; applications and files must be loaded at least once before they’ll be fully cached. In the tests below, files and programs were launched twice before measurements were taken, with a system reboot in-between each iteration. Load times were then measured for three subsequent runs, again with a reboot in-between each.

Boot time was measured from POST to fully functional desktop. Because these tests were performed on fresh OS installations without much in the way of ancillary software, the results likely understate the benefit of a cache SSD. When I previously benchmarked a Cache Accelerator in a real-world system, the benefit was much greater. Even in this configuration, adding a cache drive slashed boot times, with Nvelo’s Dataplex solidly ahead of Intel’s SRT.

Adobe programs are notorious for long load times on conventional hardware. Photoshop CS6 attempts to hide some of this by loading certain packages at boot – which, for the sake of testing, we disabled, because it was warping boot time measurements. The trial version of the software attempts to convince you that you have to log in to Adobe as well – since this also obfuscates launch time, we disabled the system’s ethernet adapter before benchmarking the program.

CS6 is very fond of SSDs; Nvelo and Intel essentially tie. Our next test examines the impact adding a cache drive has on actually loading files into the program.

For this test, we used a set of 151 images, all of which were taken with an iPhone 4S. A handful of the photos had been previously edited, with sizes ranging from 60 to 400K, but most were between 2.6 to 3MB. This test was also run three times, with a full reboot between each load. The results break our previous pattern in several ways. The Caviar Black scarcely benefits from Intel SRT at all; adding a cache drive only improves performance by around 15 per cent.

Alone, the VelociRaptor is only 10 per cent faster than the Caviar Black, but the VelociRaptor and Intel 330 is the fastest solution of all, outperforming even the 330 by a six second margin. This is the first time we’ve seen one cache solution clearly outperform the other, and the first time we’ve seen distinctly different results depending on the mechanical HDD that’s used. It won’t be the last.

Now we turn to games.

Orcs Must Die! is an addictive third-person tower defence game that deserves an award for packing its entire plot into a three-word title. Again, the VelociRaptor and Intel SSD are the fastest combined solution, but the 2.5 second gap is slim pickings for a title. The Intel 330 SSD wins overall.

Civilization V: Gods and Kings really isn’t fond of the older Caviar Black, though a cache drive neatly alleviates the problem. The VelociRaptor’s performance, meanwhile, scarcely moves with or without a cache accelerator.

Shogun 2's load times are legendary in a Tom Cruise-jumping-on-the-sofa kind of way. It takes the Caviar Black 2.5 minutes just to launch the game. Loading one of the historical missions (not shown here) adds another 54 seconds. The VelociRaptor nearly halves this, then gets a further boost from the cache drives. The performance gap between the VelociRaptor and the Caviar Black in cache configurations is just three seconds, and again, the VelociRaptor and SRT combination captures almost all the theoretical performance of the Intel 330.

For those of you who prefer a single graph, here’s the sum total of the various load times in all the scenarios we tested.

The aggregate load times indicate that both Intel’s SRT and Nvelo’s Dataplex do an excellent job of harnessing the full performance potential of their respective SSDs. Mechanical drive performance matters significantly more for Smart Response than it does for Dataplex; the VelociRaptor and SRT is 14 per cent faster than the Caviar Black when using Intel’s software.

These results imply that the benefit of a faster mechanical drive when using a Cache Accelerator is fairly small, but we’ve got one more test to show you. Our final benchmark measures internal copy performance. Our test folder was a 77.8GB folder of some 8,000 files covering a wide range of file formats and data types. When you see the results below, keep in mind that we reinstalled the entire OS, drivers and patches when we switched from Dataplex to Intel SRT on both drives. We purposefully chose a data set that was larger than the cache drives themselves, and rebooted between each copy.

Up to this point, the VelociRaptor has been significantly and noticeably faster than the Caviar Black. Here, it completely blows the latter’s doors off; so much so that we’ve reached out to Western Digital hoping for an explanation. The VelociRaptor gains nothing from pairing up with the Nvelo-powered drive from Corsair in this scenario, but slashes its total copy time by nearly a third when working with Intel’s SRT.

We couldn’t run this test on the Intel SSD alone, as the folder is larger than the 330 itself. We settled for a stripped down version and copied 13.6GB of data within the SSD and on the VelociRaptor with SRT enabled.

We’re still waiting to hear back from Western Digital, but data like this suggests that users who do a great deal of data movement within the same drive will see a marked performance advantage from WD’s VelociRaptor.


Our results show several distinct trends. If you have a slower hard drive, Nvelo’s Dataplex offers better performance than Intel SRT, even though the Corsair Cache Accelerator is somewhat slower than the Intel 330. Dataplex does an excellent job of masking the hard drive’s performance – the difference between the WD VelociRaptor and the Caviar Black when using Nvelo’s software is minimal. That’s great news if you’re rocking an older drive, not-so-great if you were hoping to pair the fastest mechanical solution with an SSD.

SRT’s performance characteristics are somewhat different. Here, the VelociRaptor’s superior performance does make a difference, at least in certain tests. This was the fastest overall configuration and very nearly as fast as the Intel 330 alone.

Neither SRT nor Dataplex make a particularly great case for the VelociRaptor; the caching solutions shrink the gap between the two mechanical drives. The one exception to this is the VelociRaptor’s self-copy performance. Here, WD’s latest and greatest ate the Caviar Black for breakfast, even when the latter was equipped with a cache drive. The Intel 330 improved this performance even further.

As for the bigger picture of whether or not cache SSDs can improve overall hard drive performance, they definitely can. Exactly how much of a benefit you’ll see depends on your chipset and the type of SSD you choose, but adding a cache to a system is an excellent way to improve responsiveness. It’s particularly attractive on older or secondary systems, where you might not be willing to spring for an expensive new high capacity SSD, but could justify a smaller upgrade that still provides much faster performance.

blog comments powered by Disqus