Samsung’s 850 Pro solid-state drive reviewed
Here we go again. Another Serial ATA SSD is ready for prime time. Like all the others that have been cropping up recently, it combines a familiar controller with next-gen flash memory.
But Samsung’s 850 Pro isn’t like the others at all.
In most new SSDs, the next-gen flash is just a die shrink of the previous generation. The nanoscale features are smaller, enabling higher bit densities, but the technology is fundamentally the same. The flash memory in the 850 Pro is on a whole other level—32 of them, actually. While traditional NAND sticks to a planar layout, the 850 Pro’s V-NAND extends into three dimensions by stacking multiple flash layers on top of one another.
V-NAND is designed to avoid some of the constraints associated with shrinking NAND lithography to ever-smaller process nodes. Samsung claims V-NAND offers higher performance and longer endurance than typical flash, too. Naturally, the 850 Pro follows suit. This baby is billed as not just the fastest SATA drive around, but also the most durable. To underscore that claim, Samsung has given it a 10-year warranty.
So, yeah, the 850 Pro isn’t just another Serial ATA SSD. Let’s see what makes it tick.
Flash in three dimensions
Before discussing the drive, we should probably start with the flash and the fact that this isn’t Samsung’s first 3D rodeo. The first generation of V-NAND debuted last August with 24 layers and, eventually, found its way into an accompanying server drive aimed at datacenters and other enterprise applications. The 850 Pro is based on second-gen V-NAND with 32 layers. Samsung tells us “essentially the same equipment” was used to manufacture the new chips, which like their predecessors, are fundamentally different than the planar NAND found in pretty much every other SSD.
Most flash memory is laid out on a single layer. Storage densities are increased by shrinking the lithography, allowing more cells to be squeezed into the same area. More cells per unit area translates to more gigabytes per wafer, effectively lowering the per-gigabyte cost. Those savings are passed along to consumers in the form of lower SSD prices.
For years, flash makers have pursued finer fabrication techniques. They’ve been very successful, but planar technology is approaching its limits. Patterning becomes more difficult as the feature size shrinks, and packing cells closer together increases the potential for interference between them. 3D NAND skirts both of those challenges by stacking memory cells vertically, a path to higher densities that doesn’t involve shrinkage.
V-NAND’s structural changes run even deeper than its multi-tiered layout. In a traditional planar NAND cell (pictured on the left in the image above), data is stored by trapping electrons inside a floating gate. The gate is conductive, but it’s suspended in an oxide insulator that keeps electrons from escaping. The electrons come from the underlying substrate, which sits below a tunnel oxide layer that acts as an insulator. Voltage applied at the control gate causes electrons to tunnel through the oxide layer.
Electrons tunnel into the cell when data is written and back to the substrate when it’s erased. As write and erase cycles accumulate, the tunneling process breaks down the oxide layer and leaves stray electrons stranded in it. Eventually, a short can form between the floating gate and the substrate, draining the gate of electrons and eliminating its ability to hold a charge. The cell is effectively dead at this point and must be retired.
Floating gate tech has been around since the 70s, but it’s not the only option. Another approach, called charge trap flash, swaps the floating gate for a trapping layer. The same tunneling process is used to move electrons in and out of the cell, but the trapping layer is an insulator, making the cell less vulnerable to erosion of the tunnel oxide. Shorts don’t drain the entire cell, just the electrons in the immediate vicinity of the breach.
The inherent short-circuit tolerance of charge trap flash allows for a thinner tunnel oxide layer. Write and erase speeds can improve as a result. The thinner layer also enables lower programming voltages, which can reduce power consumption and slow the breakdown of the oxide layer.
Samsung introduced planar NAND based on charge trap tech way back in 2006. The chip was built on a 40-nm process, and the approach apparently didn’t stick, because Samsung’s last few generations of 2D NAND have been based on floating gates. But the trapping layer was reborn in V-NAND, where it wraps around a vertical electron channel that spans multiple layers.
Despite what my crude frame capture from Samsung’s V-NAND promo video shows, the control gate wraps all the way around the cylindrical trap. (The full video is located at the bottom of this page.) V-NAND is three-dimensional right down to its component cells. Those cells are arranged vertically, with “much width of space” between each layer, according to the video. Samsung’s V-NAND presentation (PDF) from last year’s Flash Memory Summit claims this arrangement is “almost free” of cell-to-cell interference along its vertical word lines. The horizontal bit lines are “interference free,” the slides add.
Samsung provided additional details on its first-gen V-NAND during the International Solid-State Circuits Conference earlier this year. Nikkei Technology’s coverage reveals several interesting tidbits, including the fact that the chips are fabbed on a 40-nm process. Sound familiar? Samsung apparently intends to keep that feature size as it adds layers, which should prevent patterning issues from standing in the way of higher densities. There’s room to scale “more than 5 generations,” the firm says, and it eventually expects V-NAND to squeeze 128GB onto a single die.
The initial V-NAND chips used 24 layers to reach 16GB, so Samsung has a long way to go before it reaches the terabit threshold (1Tb = 128GB). The second-gen chips have 32 layers, an increase of 33%, but we’re still waiting for confirmation on their capacity per die. Those extra layers could have been used to increase the capacity, but it’s also possible Samsung opted to create a 16GB part with a smaller planar footprint. Either way, it looks like the second-gen chips retain the two-bit MLC configuration of their forebears.
Details on the 850 Pro’s second-gen V-NAND are a little scarce as I write this. The drive makes its debut at Samsung’s SSD Global Summit in Seoul, South Korea today, and I’m at the event to learn more. Hopefully, I’ll have additional V-NAND details to share with you soon.
In the meantime, it’s worth highlighting a few more attributes of the first-gen chips. Samsung says they write at twice the speed of conventional MLC NAND, consume half the power, and offer “two to ten times” the endurance. Like I said, this is next-level stuff.
The endurance appears to be contingent on the write speed. According to the ISSCC slides, V-NAND can survive 3,000 write cycles at 50MB/s or 35,000 cycles at 36MB/s. That’s an order-of-magnitude increase in endurance for only a 28% decrease in performance. Samsung wouldn’t elaborate when I asked if V-NAND offers the freedom to choose between higher performance and longer endurance, but it sounds like the company will have more to say on the matter later this year.
Now that we have a better understanding of V-NAND, let’s look at how it’s implemented in the 850 Pro.
The first consumer SSD with 3D NAND
Despite the complexity of its underlying flash, the 850 Pro is relatively straightforward. It’s a direct replacement for Samsung’s previous desktop flagship, the 840 Pro, which has been around since 2012. That drive had a good run, but it lacks a few key features and tops out at 512GB. The 850 Pro goes all the way up to a terabyte, and it has all the perks one might expect from a modern SSD.
Although the case conforms to the 2.5″ form factor, the diminutive circuit board looks more like a 1.8″ drive. Our 512GB sample has four flash packages on the back, plus four more on the top. The other side of the board also houses the controller chip and its accompanying DRAM. Like the flash, these components are manufactured by Samsung. Precious few SSD makers can match the firm’s level of vertical integration.
The 850 Pro has a Samsung MEX controller with eight parallel NAND channels and a 6Gbps Serial ATA interface. That configuration is pretty typical for consumer SSDs, and the MEX chip is especially familiar. Samsung’s value-oriented 840 EVO uses the same chip. The controller has three ARM-based cores, just like the MDX chip behind the 840 Pro, but Samsung claims there’s more “hardware automation” built in. The clock frequency is higher, as well. While the MDX is clocked at 300MHz, the MEX runs at 400MHz—or it does in the EVO, anyway. We’re still waiting for specifics on the 850 Pro’s controller frequency.
Whatever the clock speed, the frequency is likely dynamic to some degree. The 850 Pro has a thermal throttling mechanism that dials back performance if the temperature gets too high. Overheating shouldn’t be a problem in desktops, but it could be an issue in mobile systems.
In another nod to notebooks, the 850 Pro supports the ultra-low-power DevSleep mode defined by the SATA spec. Samsung claims the drive’s 2-mW DevSleep draw is lower than that of any other 2.5″ SSD. The 840 Pro draws over 10 mW while slumbering, the company says.
Hardware-based encryption is also on the menu. The 850 Pro can scramble data with a 256-bit AES algorithm, and it supports the requisite Windows eDrive (IEEE 1667) and TCG Opal 2.0 standards.
|Capacity||Max sequential (MB/s)||Max 4KB random (IOps)||Price||$/GB|
The family hits the most popular capacity points between 128GB and 1TB. Surprisingly, the base model’s performance ratings are almost identical to those of its bigger brothers. Lower-capacity SSDs are typically slower because they lack sufficient NAND dies to exploit the controller’s internal parallelism. Samsung says V-NAND is fast enough to make up the difference, allowing the 128GB drive to largely match its siblings. According to the official specs, the 128GB unit only lags behind in sequential writes—and by only about 10%.
If V-NAND helps the lower-capacity drives keep up, it should theoretically make the higher-capacity ones even faster. The 850 Pro’s performance appears to be bound by the Serial ATA interface, though. A faster PCI Express interface will likely be required to fully exploit V-NAND’s potential. That said, Samsung claims the 850 Pro still offers a speed boost at the lower queue depths associated with typical client workloads. Performance consistency is supposed to be improved, as well. We’ll see how the drive stacks up against a wide range of competitors in a moment.
First, we should discuss endurance, which we haven’t had the time to test ourselves. The 850 Pro has the highest endurance spec we’ve encountered on a consumer SSD: 150TB of total writes, or 80GB per day over five years. That’s a big step up from the 840 Pro, which is rated for only 73TB of total writes, but don’t read too much into those numbers. Our ongoing SSD Endurance Experiment has demonstrated that modern drives can take a lot more than their specifications suggest. All six of our subjects, including the 840 Pro and Samsung’s TLC-based 840 Series, wrote hundreds of terabytes without issue. The 840 Pro has written over 1.1 petabytes to date, and it’s still going strong.
Most users will struggle to write more than 150TB during the 850 Pro’s useful life, making its class-leading endurance rating somewhat academic. The 10-year warranty is in the same boat. While I admire Samsung’s willingness to stand behind the product for that long, the Serial ATA interface will probably be scarce a decade from now. By then, even a terabyte SSD may be too small to be useful.
At least the longer warranty and higher endurance rating help to offset the sticker shock. The 850 Pro is one of the most expensive consumer SSDs around, with per-gigabyte prices ranging from $0.68-$1.02. It’s possible street prices could end up being lower than Samsung’s MSRPs, but the 840 Pro maintained similar pricing for most of its life. I wouldn’t expect its successor to participate in a price war anytime soon. Samsung can race to the bottom with its value-oriented 840 EVO SSD.
The 850 Pro has a couple of other extras worth mentioning. Data migration software is included for folks upgrading existing systems. Then there’s Samsung’s Magician utility, which covers firmware updates, health monitoring, and OS optimization, among other functions. The utility can even allocate a portion of user-accessible storage as additional overprovisioned area.
The Magician app also serves as a gateway to RAPID mode, otherwise known as Real-time Accelerated Processing of I/O Data. Disabled by default, this optional caching scheme commandeers a slice of system memory to serve as a repository for frequently accessed data. Windows’ SuperFetch mechanism behaves similarly, but it’s limited to caching application data to speed read requests. RAPID covers all data types and even incoming writes. It only discriminates against larger media files that won’t benefit from quicker access times.
Caching writes in volatile DRAM is risky, but RAPID mode moves those writes to the SSD whenever the Windows write cache is flushed, so the chance of data loss should be relatively low. The cache’s contents are written to the drive when Windows shuts down, and they’re loaded back into RAM when the OS boots up. There’s no need to re-train the system after a reboot.
RAPID mode debuted alongside the 840 EVO before migrating to the 840 Pro. The cache size has been capped at 1GB, but the latest revision introduces a 4GB option for systems with 16GB of RAM. Our test rigs have 8GB of RAM, so we had to make do with a 1GB cache. We’ve tested the 850 Pro with and without RAPID mode enabled to gauge its performance impact. Speaking of which, let’s move on to the benchmark results on the next page.
CrystalDiskMark — transfer rates
TR regulars will notice that we’ve trimmed a few tests from our usual suite of storage results. The drives were all benchmarked in the same way, but we’ve excluded the results for tests that have grown problematic or less relevant over time. This abbreviated format should be a little easier to digest until our next-gen storage suite is ready.
First, we’ll tackle sequential performance with CrystalDiskMark. This test runs on partitioned drives with the benchmark’s default 1GB transfer size and randomized data. We’ve color-coded the results to make the 850 Pro and other Samsung SSDs easier to spot.
Modern solid-state drives are fast, but they’re no match for system memory. The 850 Pro’s RAPID config absolutely dominates the field in these tests. Samsung’s caching scheme has gotten faster, too. The original implementation managed only about 1100MB/s when we tested it with the 840 EVO last year.
Since the RAPID results throw off the scale for the other SSDs, let’s look at those separately.
The 850 Pro comes out on top even without its caching sidekick. That said, the fastest drives are all within a few MB/s of each other, and the 850 Pro actually ties for the lead in the read speed test. It’s not head-and-shoulders above the competition here.
HD Tune — random access times
Next, we’ll turn our attention to random access times. We used HD Tune to measure access times across multiple transfer sizes. SSDs have near-instantaneous seek times, so it’s hard to graph the results on the same scale as mechanical drives. The WD Black and Seagate SSHD will sit out this round to focus our attention on the SSDs.
RAPID mode once again elevates the 850 Pro above the fray. The cache-aided config has quicker random write access times for all the transfer sizes we tested, and 4KB random reads get a boost, too. But the caching scheme doesn’t improve performance with 1MB random reads. RAPID doesn’t speed read access times HD Tune’s 512-byte and 64KB tests, either.
With RAPID disabled, the 850 Pro looks much like any other SSD. I wouldn’t read too much into its relative standing when the drives are separated by tiny fractions of a millisecond. Only the 1MB random write test produces larger gaps, and even then, the top contenders are evenly matched.
TR FileBench — Real-world copy speeds
FileBench, which was concocted by TR’s resident developer Bruno “morphine” Ferreira, 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.
|Number of files||Average file size||Total size||Compressibility|
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, the SSDs are tested in a simulated used state after crunching IOMeter’s workstation access pattern for 30 minutes. The drives are also tested in a factory fresh state, right after a secure erase, to see if there is any discrepancy between the two states. There wasn’t much of one with the 850 Pro, so we’re only presenting the used-state scores.
The 850 Pro doesn’t need RAPID mode to turn in a strong performance in FileBench. Samsung’s latest is among the leaders in all five tests, and it’s the fastest SSD overall.
RAPID caching accelerates copy speeds in the movie, MP3, and RAW tests. The speedups aren’t as big as those in CrystalDiskMark and HD Tune, though, and the Mozilla and TR tests actually run slower with RAPID enabled.
The RAPID config’s Mozilla and TR copy speeds barely increased over the first three runs, suggesting the performance hit would persist even with more repetition. Additional runs probably wouldn’t improve performance in the other tests, where copy speeds only increased substantially between the first and second runs.
TR DriveBench 2.0 — Disk-intensive multitasking
DriveBench 2.0 is a trace-based test comprised of nearly two weeks of typical desktop activity peppered with intense multitasking loads. More details on are available on this page of our last major SSD round-up.
We measure DriveBench performance by analyzing service times—the amount of time it takes drives to complete I/O requests. Those results are split into reads and writes.
Even without RAPID mode engaged, the 850 Pro has a lower mean read service time than all of its peers—and by a relatively wide margin, considering the tight gaps between the rest of the contenders. The 850 Pro doesn’t take the top spot with writes, but it’s barely behind the leaders.
Adding DRAM caching to the equation improves the 850 Pro’s read performance slightly. However, RAPID mode appears to touch every write request, resulting in a much lower mean service time in that category.
Mean scores are only one part of the picture, of course. DriveBench also quantifies the variance in service times by reporting the standard deviation for both reads and writes. As one might expect, the 850 Pro has less read variance than any other SSD. It’s tied with the Vertex 460 for the lowest standard deviation with writes (excluding the RAPID config).
All the SSDs execute the vast majority of DriveBench requests in one millisecond or less—too little time for end users to perceive. We can also sort out the number of service times longer than 100 milliseconds, which is far more interesting data. These extremely long service times make up only a fraction of the overall total, but they’re much more likely to be noticeable.
Impressively, the 850 Pro’s write service times all clock in at less than 100 milliseconds. A few of the other SSDs are likewise devoid of sluggish writes, but none of them avoid longer reads. Heck, even the RAPID-boosted 850 Pro logs a few hundred of those. The 850 Pro suffers more long read service times on its own, pushing the drive just off the podium.
Our IOMeter workload features 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 test 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, so we’ve split the results. You can compare the performance of the Samsung 850 Pro to that of the competition by clicking the buttons below each graph.
Instead of presenting the results of multiple access patterns, we’re concentrating on IOMeter’s database test. This access pattern has a mix of read and write requests, and it’s similar to the file server and workstation tests. The results for these three access patterns are usually pretty similar. We also run IOMeter’s web server access pattern as part of our standard suite of tests, but it’s made up exclusively of read requests, so the results aren’t as applicable to real-world scenarios. Our own web servers log a fair amount of writes, for example.
Huh. RAPID caching hinders the 850 Pro’s IOMeter throughput with one concurrent request, improves performance with two concurrent requests, and is powerless to overcome whatever bottleneck caps performance at higher queue depths.
That bottleneck prevents the 850 Pro from matching the fastest drives at higher queue depths. Typical PC workloads have lower queue depths, though. The 850 Pro excels with those lighter loads, and it performs much better overall than most of its competition, including all the other Samsung SSDs.
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.
Only about a second separates the fastest SSD from the slowest one in our boot duration test. The 850 Pro sits in the middle of the pack, and RAPID mode actually slows it down slightly, perhaps due to the additional time required to repopulate the cache.
Level load times
Modern games lack built-in timing tests to measure level loads, so we busted out a stopwatch with a couple of titles.
Although the 850 Pro technically comes out on top here, its margins of victory are incredibly slim. As in the boot duration test, all the SSDs are within about a second of each other. Good luck telling the difference between them in the real world.
RAPID mode has little impact here, even with additional test runs. That seems a little counter-intuitive, but remember that Windows’ SuperFetch mechanism already caches application data speculatively. There may be limited room for auxiliary caching systems to improve game load times further.
We’re working on an updated batch of load-time tests for our next-gen storage suite. Shoot me an email if you have any suggestions.
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 850 Pro ties its cousin for the lowest idle power draw we’ve ever measured. Despite V-NAND’s supposedly lower power consumption, the 850 Pro sits in the middle of the pack under load. Our IOMeter workload mixes read and write requests, which may negate the flash’s claimed efficiency advantage with writes.
That’s it for our performance results. Information on our test methods and system configurations can be found on the next page, but that section is more reference than reading material. Unless you’re into nerdy technical details, feel free to skip ahead to the conclusion.
Test notes and methods
Here’s a full rundown of the SSDs we tested, along with their essential characteristics.
|Adata SP920 512GB||512MB||Marvell 88SS9189||20nm Micron sync 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 M500 240GB||256MB||Marvell 88SS9187||20nm Micron sync MLC|
|Crucial M500 480GB||512MB||Marvell 88SS9187||20nm Micron sync MLC|
|Crucial M500 960GB||1GB||Marvell 88SS9187||20nm Micron sync MLC|
|Crucial M550 256GB||256MB||Marvell 88SS9189||20nm Micron sync MLC|
|Crucial M550 512GB||512MB||Marvell 88SS9189||20nm Micron sync MLC|
|Crucial M550 1TB||1GB||Marvell 88SS9189||20nm Micron sync MLC|
|Crucial MX100 256GB||256MB||Marvell 88SS9189||16nm Micron sync MLC|
|Crucial MX100 512GB||512MB||Marvell 88SS9189||16nm 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|
|Intel 730 Series 480GB||1GB||Intel PC29AS21CA0||20nm 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 Micron sync MLC|
|OCZ Vertex 460 240GB||512MB||Indilinx Barefoot 3 M10||19nm Toshiba Toggle MLC|
|SanDisk Extreme II 240GB||256MB||Marvell 88SS9187||19nm SanDisk Toggle SLC/MLC|
|Samsung 840 Series 250GB||512MB||Samsung MDX||21nm Samsung Toggle TLC|
|Samsung 840 EVO 250GB||256MB||Samsung MEX||19nm Samsung Toggle TLC|
|Samsung 840 EVO 500GB||512MB||Samsung MEX||19nm Samsung Toggle TLC|
|Samsung 840 EVO 1TB||1GB||Samsung MEX||19nm Samsung Toggle TLC|
|Samsung 840 Pro 256GB||512MB||Samsung MDX||21nm Samsung Toggle MLC|
|Samsung 850 Pro 512GB||512MB||Samsung MEX||32-layer Samsung V-NAND|
|Seagate 600 SSD 240GB||256MB||LAMD LM87800||19nm Toshiba Toggle MLC|
|Seagate Desktop SSHD 2TB||64MB||NA||24nm Toshiba Toggle SLC/MLC|
|WD Caviar Black 1TB||64MB||NA||NA|
Our main body of results contains some of the most popular SSDs around. The bulk of the field is in the 240-256GB range, and most of those drives have 32-die configurations with no performance handicaps. For the Crucial M500, M550, MX100, and Samsung 840 EVO, whose lower-capacity flavors are tagged with slower specs, we have results for multiple capacities, including the fastest models. You can find full reviews of most of the drives in our storage section.
The solid-state crowd is augmented by a couple of mechanical drives. WD’s Caviar Black 1TB represents the old-school hard drive camp. Seagate’s Desktop SSHD 2TB is along for the ride, as well. The SSHD combines mechanical platters with 8GB of flash cache, but like the Caviar Black, it’s really not a direct competitor to the SSDs. The mechanical and hybrid drives are meant to provide additional context for our SSD results.
If you made it this far, you’re probably the sort of person who likes to gawk at naked circuit board shots. Enjoy:
We used the following system configuration for testing:
|Processor||Intel Core i5-2500K 3.3GHz|
|CPU cooler||Thermaltake Frio|
|Motherboard||Asus P8P67 Deluxe|
|Platform hub||Intel P67 Express|
|Platform drivers||INF update 188.8.131.520
|Memory size||8GB (2 DIMMs)|
|Memory type||Corsair Vengeance DDR3 SDRAM at 1333MHz|
|Audio||Realtek ALC892 with 2.62 drivers|
|Graphics||Asus EAH6670/DIS/1GD5 1GB with Catalyst 11.7 drivers|
|Hard drives||Seagate Desktop SSHD 2TB with CC43 firmware
WD Caviar Black 1TB with 05.01D05 firmware
Corsair Force Series GT 240GB with 1.3.2 firmware
Corsair Neutron 240GB with M206 firmware
Corsair Neutron GTX 240GB with M206 firmware
Crucial MX100 256GB with MU01 firmware
Crucial MX100 512GB with MU01 firmware
Crucial M500 240GB with MU03 firmware
Crucial M500 480GB with MU03 firmware
Crucial M500 960GB with MU03 firmware
Crucial M550 256GB with MU01 firmware
Crucial M550 1TB with MU01 firmware
Intel 335 Series 240GB with 335s firmware
Intel 520 Series 240GB with 400i firmware
Intel 730 Series 480GB with XXX firmware
OCZ Vector 150 256GB with 1.1 firmware
OCZ Vertex 450 256GB with 1.0 firmware
OCZ Vertex 460 240GB with 1.0 firmware
SanDisk Extreme II 240GB with R1131
Samsung 830 Series 256GB with CXM03B1Q firmware
Samsung 840 Series 250GB with DXT07B0Q firmware
Samsung 840 EVO 250GB with EXT0AB0Q firmware
Samsung 840 EVO 500GB with EXT0AB0Q firmware
Samsung 840 EVO 1TB with EXT0AB0Q firmware
Samsung 840 Pro Series 256GB with DXM04B0Q firmware
Samsung 850 Pro 512GB with EXM01B6Q firmware
Seagate 600 SSD 240GB with B660 firmware
|Power supply||Corsair Professional Series Gold AX650W|
|OS||Windows 7 Ultimate x64|
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.
Our scatter plots use an overall performance score derived by comparing how each drive stacks up against a common baseline. This score is based on a subset of the performance data from our full suite, but with CrystalDiskMark’s sequential transfer rates substituted for older HD Tune scores. (More details about how we calculate overall performance are available here.) We’ve mashed up the overall scores with per-gigabyte prices from Newegg. (Samsung’s suggested retail price was used for the 850 Pro, since it’s not available yet.)
Note that we’ve excluded the 850 Pro RAPID config from the plots. It scores a fair bit higher overall (1215%), which pushes the other results even closer together. Also, we’re leery of even momentarily caching writes in volatile DRAM, so we wouldn’t recommend RAPID mode for everyday computing. The caching scheme is an interesting extra that can boost performance in certain scenarios, but it’s not really central to the value proposition.
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.
Before we weigh in with our final verdict, we’ll bust out a few of our famous value scatter plots. The best solutions will gravitate toward the upper left corner of the plot, which signifies high performance and low prices. Click the buttons below the plot to switch between all the drives and a cropped look at just the SSDs—and keep in mind that we’ve trimmed the axes for the SSD-only plot.
The 850 Pro is the fastest SATA SSD we’ve ever tested. Period. Although it’s not miles ahead of the other top-tier alternatives overall, it has a definite edge in our catch-all metric. More importantly, the 850 Pro excels in some of our most demanding tests. I’m particularly impressed with the drive’s quick service times in our real-world DriveBench simulation and its high throughput at lower IOMeter queue depths. Both of those performances suggest Samsung’s new hotness is well-suited to demanding consumer workloads.
Class-leading performance doesn’t come cheap. The 850 Pro costs roughly twice as much per gig as the best value-oriented SSDs. That’s a tough pill to swallow, especially since the drive doesn’t outshine its budget brethren in each and every test. The higher endurance rating and 10-year warranty soften the blow a little, but not enough to make the 850 Pro a good value.
The thing is, value isn’t the primary concern in premium circles. Having the best is a higher priority, and the 850 Pro fits the bill in pretty much every other category. Bragging rights? Check.
Then there’s the fact that this is the first consumer SSD with 3D V-NAND. Stacking multiple layers seems to be the future of flash memory, and the 850 Pro delivers the future today. That makes the drive rather special, I think; it’s a truly unique offering in a world filled with increasingly similar designs. Cutting-edge? Definitely.
I can’t make a rational case for the 850 Pro. Heck, if it were my money, I’d probably get a much larger SSD and live happily with slightly slower access to a lot more data. But I still want the 850 Pro—or better yet, a version with a PCI Express interface. Another V-NAND SSD is due before the end of the year, so perhaps I’ll get my wish. Fingers crossed.