Samsung’s 840 EVO solid-state drive reviewed

Samsung is one of the biggest names in technology. The Korean giant makes everything from vivid 4K televisions to powerful smartphones to cutting-edge washing machines—with accompanying Android apps, of course. You can’t throw a rock in an electronics store without hitting at least half a dozen Samsung products.

Your odds of striking a PC equipped with a Samsung SSD are pretty good, too. The company claims 60% of SSD-equipped PCs shipped this year will have Samsung drives. Samsung’s SSDs are also popular components on their own. In the US, the firm says it makes up 36% of the market for branded solid-state drives.

A lot of Samsung’s recent sales have come from the 840 Series. Over 2.5 million of these entry-level drives have sold worldwide, and for good reason. The 840 Series offers solid performance at extremely affordable prices. It does have a weakness, though. The drive’s TLC NAND costs less per gigabyte than the MLC flash typically used in consumer SSDs, but its write performance is somewhat compromised by the higher bit density.

Samsung has attempted to address that shortcoming with its next-generation successor to the 840 Series. The 840 EVO features a faster controller, newer TLC NAND, and a dedicated write cache populated with speedier SLC flash. There’s a second layer of caching that leans on system memory, too.

Despite these enhancements, the 840 EVO retains the budget-friendly pricing of its predecessor. We’ve been testing one pretty much around the clock for the past few days. Let’s see how it stacks up.

The finest dual-mode flash
If you look at Samsung’s semiconductor businesses, it’s no wonder the company is a formidable force in the SSD market. Samsung is the biggest flash producer in the world, after all. It also has expertise in customizing multi-core processors and fabricating DRAM, allowing the firm to produce all the component parts of a modern SSD entirely within its own facilities.

Most of the Samsung silicon inside the 840 EVO is second-generation TLC NAND. These chips are fabbed on a 19-nm process, just a small step down from the 21-nm node used for the 840 Series’ flash. The new NAND conforms to the same 400Mbps Toggle DDR 2.0 specification as the old chips.

Perhaps to avoid the negative connotations associated with TLC memory, Samsung refers to the 840 EVO’s NAND as three-bit MLC. That’s technically correct: TLC flash has three bits per cell, one more than MLC memory and two more than the single-bit SLC stuff. The higher storage density makes TLC NAND cheaper to produce, since more gigabytes can be squeezed out of each wafer. However, it also reduces the NAND’s write performance and endurance.

I explained the challenges associated with TLC NAND in our 840 Series review, so I’ll just give you the Coles Notes here. Understanding the problem requires some knowledge of how flash memory works. NAND cells are written by causing electrons to tunnel into an insulated floating gate. The negative charge associated with these electrons defines the data in the cell, and the range of possible voltages is independent of the number of bits.

Because there is some cell-to-cell variance in the silicon, writing to the NAND involves a verification step that reads the contents of the cell. Reading entails applying a series of control voltages to hone in on the charge in the floating gate. In TLC NAND, that charge can represent one of eight possible values between 000 and 111. MLC NAND only has to contend with four values between 00 and 01, while SLC is limited to two: 0 and 1. The more bits per cell, the more iterative steps are required to verify the data, slowing the write process.

As flash memory accumulates write/erase cycles, electrons get trapped in the gate insulator, reducing the range of voltages that can be used to represent data. This shrinkage is especially troublesome for TLC NAND, which has to contend with more bits within that narrowing window. Increasing the number of bits per cell lowers its write/erase tolerance.

NAND cells are made even more vulnerable by smaller process geometries, which reduce the thickness of the insulator layer. Samsung claims the 19-nm flash used in the 840 EVO is just as resilient as the 21-nm chips employed by the 840 Series, though. According to the official reviewer’s guide, the 840 EVO’s NAND should endure “at least 2,500” write/erase cycles. That’s an impressive figure for 19-nm TLC chips, but we shouldn’t be surprised. Samsung makes a lot of flash, and only the best dies are selected for use in its solid-state drives. The firm says its SSD-grade TLC NAND has 20X fewer bad blocks than the flash bound for lesser devices, such as USB thumb drives.

Although most of the 840 EVO’s flash is configured with three bits per cell, a small slice of each die is addressed as single-bit SLC NAND. This portion of the flash is dedicated to TurboWrite, a buffering technology that caches host writes before transferring them to the drive’s main storage. The SLC cache promises higher write speeds than the TLC main storage, particularly for sequential transfers.

TurboWrite funnels all downstream traffic through the SLC cache. Writes only bypass TurboWrite if the buffer is full. When the drive is idle, cached writes are transferred out of the buffer and into the TLC NAND.

The size of the TurboWrite cache varies based on the 840 EVO’s total capacity. The 120 and 250GB models have 3GB reserved for TurboWrite, while each higher-capacity variant adds another 3GB to the cache. According to Samsung, 3GB is enough to accelerate “everyday performance scenarios.” Even the smallest TurboWrite cache is considerably larger than the 1GB buffer employed by SanDisk’s Extreme II SSD. The Extreme II is the only other SSD we’ve seen with an SLC cache.

Due to the lower bit density of SLC NAND, the flash footprint of the TurboWrite cache is three times higher than the buffer’s capacity. For example, the 250GB model’s 3GB SLC cache consumes 9GB of TLC NAND. With the exception of the 120GB model, the TurboWrite cache monopolizes 3.5% of the drive’s total flash capacity. The smallest member of the family devotes 7% of its flash to TurboWrite.

TurboWrite is responsible for the 840 EVO’s slightly lower capacities (250GB instead of 256GB, for example). The 840 Series comes in similar sizes, but it doesn’t have a fancy SLC cache. The older drive does, however, overprovision additional capacity to replace bad blocks caused by cell failures. Samsung says its initial TLC endurance estimates were conservative, and that the NAND is robust enough to repurpose the extra overprovisioned capacity for TurboWrite.

With just about all incoming data running through the SLC cache, that portion of the flash is going to chew through a lot more write/erase cycles than the TLC main storage. Single-bit NAND has much higher write endurance than its three-bit counterpart, though. Consolidating smaller cache writes into larger blocks before moving them to main storage could also make more efficient use of the TLC NAND, prolonging its life. We’re waiting to hear back from Samsung about whether TurboWrite repackages writes before they’re passed to the TLC NAND—and whether the SLC cache is moved around on the die to spread the wear evenly.

Less is more: The rest of the 840 EVO
Thanks to its 128Gb TLC NAND, our 840 EVO 500GB sample has the highest physical storage density we’ve seen in an SSD. Cracking open the 7-mm, 2.5″ case reveals a tiny circuit board populated with just four flash packages.

Each of those packages has eight individual NAND dies lurking inside. This incredible density allows the terabyte model to be built with just eight packages. It looks like that model uses the same diminutive circuit board, too. Flipping the PCB reveals solder points for more flash.

This side of the board houses the EVO’s updated MEX controller. Like its MDX predecessor, the chip has three ARM-based cores. The clock speed of those cores has been boosted from 300 to 400MHz, making the MEX controller more powerful than what’s found in the high-end 840 Pro. Samsung has also increased the amount of “hardware automation” by processing more commands directly on the controller rather than in the firmware. We’ve asked Samsung if this automation involves custom logic or simply makes better use of the existing ARM cores, but we don’t have an answer yet.

Samsung has tweaked the 840 EVO with an eye toward providing better performance at the low queue depths typical of consumer workloads. New flash management algorithms have been added to the firmware, and they employ more advanced signal processing techniques to eke every last bit of life out of the NAND. Signal processing and error correction are essential to maintaining endurance as flash process geometries shrink, according to Samsung. The firm appears uninterested in using SandForce-style write compression to reduce NAND wear.

To prevent overheating in cramped notebook chassis, the 840 EVO includes a Dynamic Thermal Guard feature that throttles performance when temperatures exceed 50°C. Samsung hasn’t explained exactly how this built-in protection works, but the scheme likely involves scaling back the clock speed of the controller’s ARM cores.

256-bit AES encryption support is also built into the MEX controller. Right now, the 840 EVO supports full-disk encryption gated by an old-school ATA password. Support for encryption via the Windows 8 eDrive and TCG Opal specifications is planned for a firmware update due in September.

The only other controller elements worthy of note are those that allow the EVO to double the maximum storage capacity of the other members of the 840 family. In addition to adding support for denser 128Gb flash dies, Samsung tuned the controller to work with the extra DRAM cache required to manage a terabyte of flash. Our 500GB drive has 512MB of LPDDR2 memory onboard, and the 1TB model sports an even gig of RAM.

The table above outlines the specifications for all the members of the 840 EVO family. I’ve estimated the die configurations and am awaiting clarification from Samsung, but the math works out.

With eight parallel NAND channels and the ability to address four chips per channel, the EVO’s controller needs at least 32 dies for optimal performance. At 128Gb (16GB) per die, only the 500GB and larger models deliver on the drive’s full potential. The 120 and 250GB models have lower write performance ratings.

Note that the performance table includes two sequential write speed figures for each model. The first refers to the speed of the TurboWrite SLC cache, while the second applies to data written directly to the TLC main storage. The 120 and 250GB variants have the most to gain from TurboWrite, but even the larger capacities should enjoy a healthy performance boost. Unfortunately, Samsung doesn’t specify whether the EVO’s random write performance ratings apply to the drive’s SLC or TLC zones.

Based on specifications and pricing, the 120GB model looks like a pretty raw deal. The cost per gigabyte drops substantially if you pony up for the 250GB model, and the terabyte derivative looks like a steal, relatively speaking. It wasn’t so long ago that $600 SSDs offered only 80GB of storage.

Given the 840 EVO’s affordable price tag, it’s no surprise that the drive is covered by a three-year warranty. Five-year coverage is typically restricted to high-end models like the 840 Pro.

The warranty length doesn’t tell you how many writes the drive can endure before the flash burns out. Samsung has been reticent to publish total-bytes-written or gigabytes-per-day specifications for its SSDs, and that trend continues with the 840 EVO. However, if the firm’s 2,500-cycle cycle estimate for the NAND is accurate, the drive should be able to endure hundreds of terabytes of writes—plenty of endurance for consumer workloads.

Slick software with DRAM caching
Samsung views software as a key component of its SSDs. The 840 EVO ships with an updated version of Samsung’s data migration application, which is designed to ease the transition from large mechanical drives. SSDs are typically smaller than the hard drives they replace, and the migration software includes a custom cloning option that allows excess data to be transferred to secondary storage. The software scans the system for large files that are ripe for relocation. Users can then select which files to move to a separate backup drive before the cloning process begins.

In addition to new migration software, the 840 EVO comes with the latest version of Samsung’s Magician Windows utility. This application tracks host writes in addition to overall drive health.

The Magician app includes a system optimization routine that tweaks Windows settings to make older versions of the OS friendlier to SSDs. It also features secure erase and firmware update functionality. We’ve seen these features in SSD utilities before, but the Magician has another trick up his sleeve: RAPID mode.

RAPID mode was created by Nvelo, which was acquired by Samsung late last year. The startup was best known for its Dataplex caching software, which uses SSDs to accelerate the performance of mechanical hard drives. RAPID is similar, but it taps into system memory to accelerate SSD performance. I’ll indulge the excessive capitalization, if only because there’s an appropriate acronym attached: Real-time Accelerated Processing of I/O Data.

Read acceleration is RAPID’s primary mission. The software observes user access patterns and speculatively caches both frequently and recently accessed data. Windows’ SuperFetch mechanism is similar, but Samsung says it’s limited to applications and can’t cache user data. RAPID looks at all host reads, and it excludes things like large media files to avoid polluting the cache with data that won’t benefit from higher storage performance.

RAPID also provides “write optimization,” allowing it to maintain performance with higher queue depths. There are other optimization techniques Samsung is keeping close to its chest; consolidating smaller writes into larger blocks would be a good candidate. Since this write-caching capability stores writes in volatile DRAM, there is some danger of data loss due to an unexpected power failure. To avoid caching writes in DRAM for too long, RAPID transfers them to the SSD every time the Windows write cache is flushed. RAPID mode requires the Magician software and is disabled by default, so there’s little chance of turning it on accidentally.

Up to 1GB of system memory can be used by RAPID, and compression makes the most of that capacity. A corresponding amount of SSD storage is required to back up the contents of the cache between reboots. That cached data isn’t copied back into the RAPID cache until after Windows loads, so there’s no way for the scheme to accelerate the boot process.

Unfortunately, we didn’t have any luck getting RAPID to improve load times in our usual gaming tests. The caching scheme had virtually no impact on the 840 EVO’s Duke Nukem Forever and Portal 2 load times even after we ran our tests five times in a row. Frustrated, we fired up the CrystalDiskMark benchmark that Samsung used to demonstrate RAPID caching at its Global SSD Summit in Seoul, Korea last week.

The 840 EVO is no slouch without RAPID mode enabled. However, the secondary caching layer dramatically improves the drive’s scores in six of CrystalDiskMark’s eight standard tests. At the very least, it’s a prolific benchmark accelerator.

Unfortunately, our time with the 840 EVO has been too limited to test RAPID more fully. We didn’t receive our sample until Monday evening, and I’ve barely slept trying to expedite the drive’s journey through our standard test suite. My apologies for the impact that lack of sleep has had on the writing. I’m sustained by little more than caffeine at this point.

In addition to exploring RAPID mode in greater depth, we plan to test the 250GB version of the 840 EVO. That capacity is popular among enthusiasts, but we couldn’t get our hands on one in time for this initial review. In the meantime, let’s see how the 500GB model compares to its rivals.

Our testing methods
If you’re familiar with our testing methods and hardware, the rest of this page is filled with nerdy details you already know; feel free to skip ahead to the benchmark results. For the rest of you, we’ve summarized the essential characteristics of all the drives we’ve tested in the table below. Our collection of SSDs includes the most popular SSD configurations on the market.

	Cache	Flash controller	NAND
Corsair Force Series 3 240GB	NA	SandForce SF-2281	25nm Micron async MLC
Corsair Force Series GT 240GB	NA	SandForce SF-2281	25nm Intel sync MLC
Corsair Neutron 240GB	256MB	LAMD LM87800	25nm Micron sync MLC
Corsair Neutron GTX 240GB	256MB	LAMD LM87800	26nm Toshiba Toggle MLC
Crucial m4 256GB	256MB	Marvell 88SS9174	25nm Micron sync MLC
Crucial M500 240GB	256MB	Marvell 88SS9187	20nm Micron sync MLC
Crucial M500 480GB	512MB	Marvell 88SS9187	20nm Micron sync MLC
Intel 335 Series 240GB	NA	SandForce SF-2281	20nm Intel sync MLC
Intel 520 Series 240GB	NA	SandForce SF-2281	25nm Intel sync MLC
OCZ Agility 4 256GB	512MB	Indilinx Everest 2	25nm Micron async MLC
OCZ Vector 256GB	512MB	Indilinx Barefoot 3	25nm Intel sync MLC
OCZ Vertex 4 256GB	512MB	Indilinx Everest 2	25nm Micron sync MLC
OCZ Vertex 450 256GB	512MB	Indilinx Barefoot 3 M10	20nm Intel sync MLC
SanDisk Extreme II 240GB	256MB	Marvell 88SS9187	19nm SanDisk Toggle SLC/MLC
Samsung 830 Series 256GB	256MB	Samsung MCX	27nm Samsung Toggle MLC
Samsung 840 Series 250GB	512MB	Samsung MDX	21nm Samsung Toggle TLC
Samsung 840 EVO 250GB	512MB	Samsung MEX	19nm Samsung Toggle TLC
Samsung 840 Pro 256GB	512MB	Samsung MDX	21nm Samsung Toggle MLC
Seagate 600 SSD 240GB	256MB	LAMD LM87800	19nm Toshiba Toggle MLC
WD Caviar Black 1TB	64MB	NA	NA

Our usual assortment of 240-256GB contenders represents the highest-performance models for each drive family. These SSDs are a good match for the Samsung 840 EVO 500GB despite offering roughly half the capacity. The only exception is the Crucial M500, which like the EVO, uses 128Gb NAND that requires a higher-capacity configuration to deliver full performance. We’ve included the 240 and 480GB flavors of the M500 to provide additional perspective.

To illustrate how this crop of SSDs stacks up against mechanical storage, we’ve thrown an older Western Digital Caviar Black 1TB into the mix. Don’t expect this 7,200-RPM hard drive to keep up; it’s included for reference only.

We used the following system configuration for testing:

Thanks to Asus for providing the systems’ motherboards and graphics cards, Intel for the CPUs, Corsair for the memory and PSUs, Thermaltake for the CPU coolers, and Western Digital for the Caviar Black 1TB system drives.

We used the following versions of our test applications:

• Intel IOMeter 1.1.0 RC1
• HD Tune 4.61
• TR DriveBench 1.0
• TR DriveBench 2.0
• TR FileBench 0.2
• Qt SDK 2010.05
• MinGW GCC 4.4.0
• Duke Nukem Forever
• Portal 2

Some further notes on our test methods:

• To ensure consistent and repeatable results, the SSDs were secure-erased before almost every component of our test suite. Some of our tests then put the SSDs into a used state before the workload begins, which better exposes each drive’s long-term performance characteristics. In other tests, like DriveBench and FileBench, we induce a used state before testing. In all cases, the SSDs were in the same state before each test, ensuring an even playing field. The performance of mechanical hard drives is much more consistent between factory fresh and used states, so we skipped wiping the HDDs before each test—mechanical drives take forever to secure erase.

• We run all our tests at least three times and report the median of the results. We’ve found IOMeter performance can fall off with SSDs after the first couple of runs, so we use five runs for solid-state drives and throw out the first two.

• Steps have been taken to ensure that Sandy Bridge’s power-saving features don’t taint any of our results. All of the CPU’s low-power states have been disabled, effectively pegging the 2500K at 3.3GHz. Transitioning in and out of different power states can affect the performance of storage benchmarks, especially when dealing with short burst transfers.

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at a 75Hz screen refresh rate. Most of the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

HD Tune — Transfer rates
HD Tune lets us present transfer rates in a couple of different ways. Using the benchmark’s “full test” setting gives us a good look at performance across the entire drive rather than extrapolating based on a handful of sample points. The full test gives us fodder for line graphs, which we’ve split up by drive maker. You can click the buttons below each line graph to see how the Samsung 840 EVO compares to its rivals.

The Samsung 840 EVO largely matches the sequential read speeds of rest the 840 family. Although the SanDisk Extreme II comes out on top, the 840 posse isn’t far behind. Neither is the Crucial M500, for that matter.

The field is more spread out in HD Tune’s sequential write speed test. Again, the 840 EVO hangs close to the 840 Series. This time, however, the 840 Pro has a solid lead over its entry-level siblings. The Crucial M500 isn’t even close.

If you look at the line graphs, you’ll notice that the 840 EVO’s transfer rate is slightly faster at the very beginning of the test. Performance levels off pretty quickly, and I suspect TurboWrite may be responsible for that initial burst.

HD Tune runs on unpartitioned drives, with no file system in place, which might explain the write-rate spikes exhibited by some of the SSDs. For another take on sequential speed, we’ll turn to CrystalDiskMark, which runs on partitioned drives. We used the benchmark’s sequential test with the default 1GB transfer size and randomized data.

CrystalDiskMark produces higher sequential transfer rates overall, and the Samsung 840 EVO remains among the leaders. The default 1GB test size is small enough to fit inside the TurboWrite cache, which appears to be responsible for a massive performance improvement versus the previous generation. The 840 EVO more than doubles the sequential write speed of the old 840 Series.

HD Tune — Random access times
In addition to letting us test transfer rates, HD Tune can measure random access times. We’ve tested with four transfer sizes and presented all the results in a couple of line graphs. We’ve also busted out the 4KB and 1MB transfers sizes into bar graphs that should be easier to read without the presence of the mechanical drive.

Don’t try to differentiate between the SSDs in the line graphs. Those graphs are meant to illustrate the massive gap in access times between solid-state and mechanical storage.

The gulf between the SSDs and our lone hard drive works out to multiple orders of magnitude. This vast difference in access times is what makes solid-state drives feel so much more responsive than their mechanical counterparts.

Among the SSDs, the results are very close. Most of the drives are separated by tiny fractions of a millisecond. The Samsung 840 EVO scores better than its siblings in both tests, but you’d be hard-pressed to feel the difference in the real world.

HD Tune’s random write speed test produces similar results. The SSDs are evenly matched, and the mechanical drive is nowhere close.

The Samsung 840 EVO has the quickest 1MB random write response time of the lot, likely thanks to its TurboWrite cache. Not even the high-end 840 Pro can match the EVO in that test. At least the Pro comes close, though. The standard 840 Series lags well behind the top contenders and is nearly 2X slower than its replacement.

TR FileBench — Real-world copy speeds
Concocted by resident developer Bruno “morphine” Ferreira, FileBench runs through a series of file copy operations using Windows 7’s xcopy command. Using xcopy produces nearly identical copy speeds to dragging and dropping files using the Windows GUI, so our results should be representative of typical real-world performance. We tested using the following five file sets—note the differences in average file sizes and their compressibility. We evaluated the compressibility of each file set by comparing its size before and after being run through 7-Zip’s “ultra” compression scheme.

The names of most of the file sets are self-explanatory. The Mozilla set is made up of all the files necessary to compile the browser, while the TR set includes years worth of the images, HTML files, and spreadsheets behind my reviews. Those two sets contain much larger numbers of smaller files than the other three. They’re also the most amenable to compression.

To get a sense of how aggressively each SSD reclaims flash pages tagged by the TRIM command, we run FileBench with the solid-state drives in two states. We first test the SSDs in a fresh state after a secure erase. They’re then subjected to a 30-minute IOMeter workload, generating a tortured used state ahead of another batch of copy tests. We haven’t found a substantial difference in the performance of mechanical drives between these two states. Let’s start with the fresh-state results.

Each of our file sets is small enough to fit within the Samsung 840 EVO 500GB’s TurboWrite cache, and the buffering scheme is probably responsible for the drive’s strong performance. The EVO delivers the fastest copy speeds in two of three tests, and it’s just a smidgen off the lead in a third.

Although the EVO falls further down the standings when copying the smaller files in the Mozilla and TR sets, it still manages to beat everything but its SandForce-based rivals. Impressively, the EVO stays one step ahead of the 840 Pro across the board.

Some of Samsung’s older SSDs and firmware revisions have exhibited much lower FileBench performance after the drives have been put into a used state. So, what about the 840 EVO?

Samsung’s new hotness is tempered somewhat by our torture test, but its copy speeds drop by just 5-6%, and the slowdown is only evident in the movie and RAW tests. The 840 EVO continues to deliver the best all-around performance in FileBench.

TR DriveBench 1.0 — Disk-intensive multitasking
TR DriveBench allows us to record the individual IO requests associated with a Windows session and then play those results back as fast as possible on different drives. We’ve used this app to create a set of multitasking workloads that combine common desktop tasks with disk-intensive background operations like compiling code, copying files, downloading via BitTorrent, transcoding video, and scanning for viruses. The individual workloads are explained in more detail here.

Below, you’ll find an overall average followed by scores for each of our individual workloads. The overall score is an average of the mean performance score for each multitasking workload.

Our DriveBench workloads vary in size from 10-40GB, and the test eliminates idle time to push the SSDs as hard as possible. That formula isn’t ideal for TurboWrite acceleration. However, the 840 EVO still outperforms its siblings in the 840 family. Its advantage over the standard 840 Series is particularly notable, and I suspect the EVO’s higher controller clock frequency deserves some of the credit.

Although the 840 EVO has a higher I/O rate than the other Samsung SSDs, it’s not the fastest overall. That honor belongs to the Indilinx-based OCZ Vector and Vertex 450 SSDs. Let’s see what we can learn from the EVO’s performance in the individual DriveBench tests.

With strong scores in each multitasking workload, the 840 EVO shows no signs of weakness. The 840 Pro sneaks ahead in the file copy test, which is a bit of a surprise given our FileBench results. That said, the Pro’s advantage over the EVO is relatively small. The EVO enjoys a much larger edge over the Pro in the compiling test.

TR DriveBench 2.0 — More disk-intensive multitasking
As much as we like DriveBench 1.0’s individual workloads, the traces cover only slices of disk activity. Because we fire the recorded I/Os at the disks as fast as possible, solid-state drives also have no downtime during which to engage background garbage collection or other optimization algorithms. DriveBench 2.0 addresses both of those issues with a much larger trace that spans two weeks of typical desktop activity peppered with multitasking loads similar to those in DriveBench 1.0. We’ve also adjusted our testing methods to give solid-state drives enough idle time to tidy up after themselves. More details on DriveBench 2.0 are available on this page of our last major SSD round-up.

Instead of looking at a raw IOps rate, we’re going to switch gears and explore service times—the amount of time it takes drives to complete an I/O request. We’ll start with an overall mean service time before slicing and dicing the results.

The Samsung 840 EVO isn’t among the leaders in DriveBench 2.0, at least when we look at the mean service time for the entire trace. The EVO falls almost exactly between the 840 Series and the 840 Pro. It’s more responsive overall than the Crucial M500, but only just.

We can sort DriveBench 2.0 service times into reads and writes to learn more about the EVO’s performance characteristics.

The EVO’s read service times are comparatively better than its writes. In both cases, though, Samsung’s latest SSD is sandwiched between its siblings. The 840 Series is a little bit slower, while the 840 Pro is a little bit faster.

If TurboWrite makes a difference here, its impact appears to be minimal. DriveBench 2.0 may compress the trace’s idle time too much for the cache to empty itself. We’re waiting to hear back from Samsung about how long the EVO must idle before buffered data is transferred to the TLC NAND.

There are millions of I/O requests in this trace, so we can’t easily graph service times to look at the variance. However, our analysis tools do report the standard deviation, which can give us a sense of how much service times vary from the mean.

Although the 840 EVO exhibits more service time variance than some of its peers, neither of its standard deviation scores is alarmingly high. With reads, the mechanical drive has way more variance than the SSDs. With writes, the Crucial drives are the ones that suffer.

We can’t easily graph all the service times recorded by DriveBench 2.0, but we can sort them. The graphs below plot the percentage of service times that fall below various thresholds. You can click the buttons below the graphs to see how the Samsung 840 EVO compares to the other drives.

The service time distributions tell us a few interesting things. The 840 EVO has more read service times under each threshold than most of the competition. It’s hard to differentiate between the SSDs as the read threshold exceeds 0.7 milliseconds, though.

Things are even more bunched up in the write distributions, which show the SSDs largely stacked on top of each other. Note that the 840 EVO is a little behind the curve at the quickest thresholds.

All the distributions approach 100% as we hit the right side of the plots. That means the vast majority of requests are completed in one millisecond or less—pretty quickly, even in modern PC terms. The requests that take 100 milliseconds or more can potentially compromise system responsiveness, so we’ve singled them out in a final set of graphs. Although the actual percentages are low, remember that DriveBench 2.0 comprises tens of millions of I/O requests.

The 840 EVO has barely any extremely long write service times. It doesn’t fare quite as well with reads, but the percentage of long access times isn’t high enough to cause concern. There are certainly much worse offenders.

IOMeter
Our IOMeter workloads feature a ramping number of concurrent I/O requests. Most desktop systems will only have a few requests in flight at any given time (87% of DriveBench 2.0 requests have a queue depth of four or less). We’ve extended our scaling up to 32 concurrent requests to reach the depth of the Native Command Queuing pipeline associated with the Serial ATA specification. Ramping up the number of requests also gives us a sense of how the drives might perform in more demanding enterprise environments.

We run our IOMeter tests using the fully randomized data pattern, which presents a particular challenge for SandForce’s write compression scheme. We’d rather measure SSD performance in this worst-case scenario than using easily compressible data.

There’s too much data to show clearly on a single graph for each access pattern, so we’ve once again split the results by drive maker. You can compare the performance of the Samsung 840 EVO to that of the competition by clicking the buttons below each graph.

The web server access pattern is all about read performance—there isn’t a single write request in the entire workload. The Samsung 840 EVO is an absolute beast here, topping the transaction rate of just about every one of its peers. Only the 840 Pro offers higher performance across the full extent of our scaling load.

Next up: a trio of workloads that mix read and write requests.

So much for TurboWrite. The SLC caching scheme appears to be of no help in IOMeter, leaving the Samsung 840 EVO defenseless against even its 840 Series predecessor.

To be fair, our IOMeter tests are a bit extreme for the EVO’s caching mojo. Each loop hammers drives with 72 minutes of continuous I/O at an increasing queue depth. This onslaught should saturate the SLC cache in short order, and there’s little idle time to flush the buffer. These IOMeter tests are really geared toward evaluating performance with demanding server workloads. TurboWrite, and indeed the 840 EVO as a whole, has been tuned for client environments, where disk idle time abounds and I/O tends to occur in short bursts.

Boot duration
Before timing a couple of real-world applications, we first have to load the OS. We can measure how long that takes by checking the Windows 7 boot duration using the operating system’s performance-monitoring tools. This is actually the first test in which we’re booting Windows off each drive; up until this point, our testing has been hosted by an OS housed on a separate system drive.

Level load times
Modern games lack built-in timing tests to measure level loads, so we busted out a stopwatch with a couple of reasonably recent titles.

As UFC promoter Dana White would say, the Samsung 840 EVO is “in the mix” in our load time tests. The trouble is, so is pretty much every other SSD. No more than a second separates most of the drives, and you simply won’t notice the differences between them.

You could practically doze off in the time it takes the mechanical drive to complete the same tests, though. That may be a bit of an exaggeration, but I’m really tired. For everyday tasks like booting the OS and loading applications, the benefits of solid-state storage are clearly evident.

Power consumption
We tested power consumption under load with IOMeter’s workstation access pattern chewing through 32 concurrent I/O requests. Idle power consumption was probed one minute after processing Windows 7’s idle tasks on an empty desktop.

The 840 EVO’s idle power consumption isn’t anything to write home about. I have some doubts about whether this test scenario leaves enough time for some of the SSDs to perform background flash maintenance before dropping into lower-power states, though. We may have to start logging wattage over time to get a better sense of real-world power draw.

Our IOMeter load hits the SSDs hard enough that their power consumption tends to be consistent, at least over time. In this test, the 840 EVO pulls fewer watts than most of the competition.

The value perspective
Welcome to our famous value analysis, which adds capacity and pricing to the performance data we’ve explored over the preceding pages. We used Newegg prices for most of the SSDs, and we didn’t take mail-in rebates into account when performing our calculations. Since the Samsung 840 EVO isn’t available for sale as we write this, we’ve used the 500GB model’s $370 MSRP.

First, we’ll look at the all-important cost per gigabyte, which we’ve obtained using the amount of storage capacity accessible to users in Windows.

The EVO delivers on TLC NAND’s promise of more storage at lower prices. Only Samsung’s own 840 Series has a lower cost per gigabyte, and that’s another TLC-based drive. The 840 Series is on sale right now, too, like it is pretty much every other week. I suspect the EVO will enjoy similar discounts when it fully replaces the standard model.

Our remaining value calculation uses a single performance score that we’ve derived by comparing how each drive stacks up against a common baseline provided by the Momentus 5400.4, a 2.5″ notebook drive with a painfully slow 5,400-RPM spindle speed. This index uses a subset of our performance data described on this page of our last SSD round-up.

Despite being priced like a budget model, the 840 EVO sits near the top of our overall performance standings. It’s only a little slower overall than the 840 Pro, which is good enough for fourth place ahead of the SanDisk Extreme II. The old 840 Series is considerably slower overall, and so is the new Crucial M500.

Now for the real magic. We can plot this overall score on one axis and each drive’s cost per gigabyte on the other to create a scatter plot of performance per dollar per gigabyte. The best place on the plot is the upper-left corner, which combines high performance with a low price.

Excuse the cropped axes; you’ll see why they were necessary in a moment. For now, focus on the 840 EVO and its prime position toward the upper left corner of the plot. This region signifies high performance and low pricing, which is a pretty sweet combo. Only a handful of SSDs boast higher overall performance than the EVO, and they all have higher per-gigabyte prices. The EVO is clearly a big step up from the 840 Series.

We can shed more light on the EVO’s value proposition by expanding the plot to include our lone mechanical drive.

See why I trimmed the axes? The SSDs are virtually impossible to tell apart here, in part because of the hoops I had to jump through to avoid label overlap.

The differences between the SSDs are much smaller than the gap between the solid-state field and our lone mechanical drive. From this vantage point, most of the SSDs look fairly evenly matched on performance. There’s more variance in their pricing, which easily favors the EVO.