Intel took a long and winding road to get to its SSD DC S3700. When the company launched its X25-E almost four years ago (oh yeah, we were there: Intel’s X25-E SSD Walks All Over The Competition), it outperformed pretty much everything on the market. A home-grown 3 Gb/s controller, 50 nm SLC NAND, unprecedented write endurance, and amazingly consistent performance were all cutting-edge at the time. To this day, the X25-E remains one of our favorite enterprise drives.
Available capacities were small, though, and the drive was extremely expensive throughout its life. Of course, that didn't matter to many of the companies that bought X25-Es, since performance and reliability compared to the competition were outstanding. That advantage was attributable to the fact that Intel not only wrote its own custom firmware, but also designed the controller itself.
The SSD 710's controller: We've seen that one before!
From that point on, though, Intel's storage team didn't come up with anything nearly as groundbreaking. On the NAND side, the company abruptly switched from SLC to HET-MLC in its enterprise-oriented drives. The first X25-E successor, SSD 710, was promising, but suffered a few major flaws. Its controller was simply a revised version of the 10-channel design the company had used for years.
Although its performance was close to that achieved by the venerable X25-E, the SSD 710 surfaced at a time when 6 Gb/s drives were the norm, so its results came across fairly underwhelming. It simply didn't have any other controller options to fall back on. Intel chose to partner with Marvell and then SandForce for its 6 Gb/s client-oriented SSDs. But once it leaned on those other vendors, no amount of firmware tweaking was enough to distinguish its product from other companies using the same ASICs.
And while the SSD 710's HET-MLC flash offered a good compromise between performance, capacity, and endurance, the NAND technology was too expensive. As a result, Intel's enterprise follow-up didn't really differentiate itself in a segment that had become increasingly competitive. The company needed to start over, go back, and do what it does best: innovate.
Meet The SSD DC S3700

It's amazing how much can change in one year. The introduction of Intel's SSD DC S3700 appears to right much of what was wrong with the SSD 710. Equipped with a new eight-channel controller of its own design and mature 25 nm HET-MLC NAND, the company has a much more compelling enterprise-oriented offering, at least on paper.
The SSD DC S3700 is available at four capacity points in a 2.5" form factor (100, 200, 400, and 800 GB), and two capacity points in a 1.8" design (200 and 400 GB). We'd expect 2.5" drives, but the 1.8" offerings are more of a surprise. We've seen vendors increasingly shy away from the smaller form factor. However, Intel claims there is a growing demand for them in the blade and micro-server markets.
| Intel SSD DC S3700 Series | ||||
|---|---|---|---|---|
| User Capacity | 100 GB | 200 GB | 400 GB | 800 GB |
| Interface | 2.5" 6 Gb/s SATA | 2.5"/1.8" 6 Gb/s SATA | 2.5" 6 Gb/s SATA | |
| Sequential Read | 500 MB/s | |||
| Sequential Write | 200 MB/s | 365 MB/s | 460 MB/s | 460 MB/s |
| 4K Random Read | 75,000 IOPS | |||
| 4K Random Write | 19,000 IOPS | 32,000/29,000 IOPS | 36,000 IOPS | 36,000 IOPS |
| Power Consumption (+5 V Active) | 2.8 W | 4.2 W | 5.2 W | 5.8 W |
| Power Consumption (+5 V Idle) | 0.6 W | |||
| Write Endurance | 1.83 PB | 3.65 PB | 7.3 PB | 14.6 PB |
| Encryption | AES-256 | |||
Although those vendor-supplied specifications aren't aggressive enough to knock Samsung's client-oriented 840 Pro from its perch, they do represent a significant improvement over the SSD 710. The SSD DC S3700 is expected to be 2x faster than its predecessor in random read operations, and 15x quicker in random writes. Sequential performance should also be more than twice as fast.
Those specs won't bowl over anyone current with SSD technology. More attention-grabbing was the pricing. When Intel introduced the SSD 710, it wanted something like $7/GB. Even the PCI Express-based SSD 910 launched a few months ago and equipped with the same 25 nm HET-MLC sells for ~$5.5/GB.
In contrast, the 2.5" SSD DC S3700 at 100, 200, 400, and 800 GB is priced at $235, $470, $940, and $1880, respectively. The 1.8" models are expected to command a roughly $25 premium at each capacity point. But that's only $2.35/GB. So, in the span of a year, what you pay per gigabyte of capacity dropped nearly 65%. Even though Intel is still asking almost two times what you'd spend on an 840 Pro, the SSD DC S3700 incorporates enterprise-oriented functionality you won't find anywhere else.
We peeled the cover off of our 200 and 800 GB samples to get a better sense of hardware Intel is using. On both drives, it's easy to spot the new 6 Gb/s PC29AS21CA06 Gb/s controller. Again, it employs an eight-channel architecture instead of the 10-channel design used previously.
On the 800 GB drive, the controller is flanked by a pair of 4 Gb Samsung DDR3-1600 FBGA RAM packages, totaling 1 GB of cache. Below the controller, you see eight NAND packages, each adding 64 GB of capacity. The other side of the drive features eight more, pushing the 800 GB model's total capacity up to 1 TB.

The 200 GB version is configured quite a bit differently. The layout and controller are almost identical, but as you can see, there's a single 2 Gb DDR3 cache chip from Micron, rather than the two 4 Gb Samsung packages.
You still find 16 total NAND packages on-board. However, their densities aren't uniform. Fourteen of them are 16 GB chips, one adds 32 GB, and one is an 8 GB part, totaling 264 GB.
The 200 GB drive.
The 800 GB sample.
One of the SSD DC S3700's more unique enterprise-oriented features is the use of capacitors to maintain data integrity during a power failure. Two 35 V, 47 uF capacitors store enough charge to commit all data in the write cache to NAND. The radial electrolytic capacitors are bent into a cutout in the PCB, allowing Intel to free up board space that would have been needed for surface-mount capacitors.
Intel also incorporates sensor logic that periodically checks capacitor health. Any outright failure, or even degraded performance, triggers a SMART event and disables write caching to counter the risk of data loss.

Another first for Intel is utilization of the SATA connector's 12 V rail. In fact, the SSD DC S3700 can operate on +5 V, +12 V, or both. Granted, using the +12 V rail does increase power draw, but we'll go into more detail on that in the power consumption analysis.
SSD DC S3700 Write Endurance
We typically spend a lot of time testing the write endurance of enterprise-oriented SSDs. Why? It's a big reason the highest-end SSDs cost so much more than the client-oriented stuff.
Intel uses the same 25 nm HET-MLC on its SSD DC S3700 that it put in the SSD 710 and 910. In fact, the company gives an identical write endurance spec for its 400 and 800 GB SSD DC S3700 and SSD 910 drives. There may be some differences in the way Intel's new controller handles TRIM and garbage collection, but they're likely very small.
| Endurance Rating Sequential Workload, QD=1, 8 MB, Random | Intel SSD DC S3700 | Intel SSD 710 | Intel SSD 910 |
|---|---|---|---|
| NAND Type | Intel 25 nm HET-MLC | Intel 25 nm HET-MLC | Intel 25 nm HET-MLC |
| RAW NAND Capacity | 264 GB | 320 GB | 896 GB |
| IDEMA Capacity (User Accessible) | 200 GB | 200 GB | 800 GB |
| Over-provisioning | 32% | 60% | 12% |
| P/E Cycles Observed (IDEMA) | 36,343 | 36,600 | 46,339 |
| P/E Cycles Observed (Raw) | 27,532 | 22,875 | 41,374 |
| Host Writes per 1% of MWI | 72.69 TB | 73.20 TB | 370.71 TB |
| $/PB-Written | $64.66 | $181.72 | $106.60 |
Based on its over-provisioning and lower cost, the SSD DC S3700's cost per petabyte written is extremely low for an MLC-based drive. Of course, it cannot compete with the extreme endurance rating of SLC flash. Drives like Micron's RealSSD P320h reign supreme at less than $40 per petabyte written.
Will that change in the near future? If Intel manages to push the price of HET-MLC drives down to $1.40/GB (a 40% reduction from current prices), it would match the Micron drive in dollars per petabyte written. Considering the recent price history of Intel's HET-MLC-based drives and the rather static pricing of SLC-based devices in general, it doesn't seem far-fetched to anticipate this in the next 12 months.
We've seen SSD vendors spend tons of R&D dollars over the past few years to improve every quantifiable aspect of solid-state storage. Sequential performance, random 4 KB writes, pricing, and overall quality are all way up. But there's one aspect of SSDs that still looks a lot like the Wild West: performance consistency.
As we've noted in previous drive reviews, averages are great on spec sheets, but they don't always tell the whole story (like frame rates in graphics card evaluations). When it comes to dealing with time-critical, deterministic systems like enterprise video, it's more important to design for worst-case scenarios. Averages just aren't good enough.
Nearly every SATA-based drive we've tested, even the ones from Intel, has exhibited consistency issues at some point. Normally, hiccups are caused by the controller firmware. In some cases (like garbage collection), you're looking at the drive's inherent behavior. In other cases, poor implementation of certain algoritms is to blame. Fortunately, Intel put a special emphasis on the consistency of its SSD DC S3700, completely redesigning the firmware to prioritize even performance over hitting peak throughput numbers. The company even adds consistency and QoS specifications and definitions to its datasheets.
First, lets take a look at the performance consistency spec, according to Intel.
| Performance Consistency Specification | 100 GB | 200/400/800 GB |
|---|---|---|
| Random 4/8 KB Read | 90% | 90% |
| Random 4/8 KB Write | 85% | 90% |
And here's the company's definition of performance consistency from its product specification:
"Performance consistency measured using Iometer based on Random 4 KB QD=32 workload, measured as the (IOPS in the 99.9th percentile slowest 1-second interval)/(average IOPS during the test). Measurements are performed on a full Logical Block Address (LBA) span of the drive once the workload has reached steady state but including all background activities required for normal operation and data reliability"
First, we have to credit Intel for going so far as to put out a spec that even attempts to quantify consistency. The company doesn't try to cherry-pick an easy test for its specification, either. It's using 4 KB random reads and writes across the entire LBA, with all background activities active during the measurement. Even with those parameters, the SSD DC S3700 is able to achieve 90% consistency across all capacities, other than random writes on the 100 GB model.
The other new specification is Quality of Service. While performance consistency takes a one-second average, QoS shows us the maximum latency for a given percentage of commands.
| QoS Specification | Queue Depth=1 | Queue Depth=32 | ||
|---|---|---|---|---|
| Capacity | 100 GB | 200/400/800 GB | 100 GB | 200/400/800 GB |
| QoS (99.9%) | ||||
| Reads | 0.5 ms | 0.5 ms | 1 ms | 1 ms |
| Writes | 0.5 ms | 0.5ms | 15 ms | 10 ms |
| QoS (99.9999%) | ||||
| Reads | 10 ms | 5 ms | 10 ms | 5 ms |
| Writes | 10 ms | 5 ms | 20 ms | 20 ms |
This specification is derived using 4 KB transfer sizes in Iometer, measuring the maximum time it takes for 99.9 or 99.9999% of commands to travel round-trip from host to drive and back to host. Once again, this is a great improvement over the normal average or typical latencies that most vendors specify.
There is one drawback to this type of devotion to consistency: maximum performance takes a hit. You won't see the extreme high-end numbers proffered by the quickest desktop SSDs. Intel is betting that the trade-off for consistency is worthwhile in most enterprise environments, though.
| Test Hardware | |
|---|---|
| Processor | Intel Core i7-3960X (Sandy Bridge-E), 32 nm, 3.3 GHz, LGA 2011, 15 MB Shared L3, Turbo Boost Enabled |
| Motherboard | Intel DX79SI, X79 Express |
| Memory | G.Skill Ripjaws Z-Series (4 x 4 GB) DDR3-1600 @ DDR3-1600, 1.5 V |
| System Drive | Intel SSD 320 160 GB SATA 3Gb/s |
| Tested Drives | Intel SSD DC S3700 200 and 800 GB, Firmware: 5DVA0138 |
| Graphics | AMD FirePro V4800 1 GB |
| Power Supply | OCZ ModXStream Pro 700 W |
| System Software and Drivers | |
| Operating System | Windows 7 x64 Ultimate |
| DirectX | DirectX 11 |
| Driver | Graphics: ATI 8.883 |
| Iometer 1.1.0 | # Workers = 4, 4 KB Random: LBA= Full Span varying Queue Depths | ||
|---|---|---|---|
| AS SSD | v1.6437.30508 | ||
| ATTO | v2.47, 2 GB, QD=4 | ||
| Custom | C++, 8 MB Sequential, QD=4 | ||
| Enterprise Testing: Iometer Workloads | Read | Random | Transfer Size |
| Database | 67% | 100% | 8 KB: 100% |
| File server | 80% | 100% | 512 Bytes: 10% |
| 1 KB: 5% | |||
| 2 KB: 5% | |||
| 4 KB: 60% | |||
| 8 KB: 2% | |||
| 16 KB: 4% | |||
| 32 KB: 4% | |||
| 64 KB: 10% | |||
| Web server | 100% | 100% | 512 Bytes: 22% |
| 1 KB: 15% | |||
| 2 KB: 8% | |||
| 4 KB: 23% | |||
| 8 KB: 15% | |||
| 16 KB: 2% | |||
| 32 KB: 6% | |||
| 64 KB: 7% | |||
| 128 KB: 1% | |||
| 512 KB: 1% |
The Storage Networking Industry Association (SNIA), a working group made up of SSD, flash, and controller vendors, has produced a testing procedure that attempts to control as many of the variables inherent to SSDs as possible. SNIA’s Solid State Storage Performance Test Specification (SSS PTS) is a great resource for enterprise SSD testing. The procedure does not define what tests should be run, but rather the way in which they are run. This workflow is broken down into four parts:
- Purge: Purging puts the drive at a known starting point. For SSDs, this normally means Secure Erase.
- Workload-Independent Preconditioning: A prescribed workload that is unrelated to the test workload.
- Workload-Based Preconditioning: The actual test workload (4 KB random, 128 KB sequential, and so on), which pushes the drive towards a steady state.
- Steady State: The point at which the drive’s performance is no longer changing for the variable being tracked.
These steps are critical when testing SSDs. It is incredibly easy to not fully condition the drive and still see fresh-out-of-box behavior and think it is steady-state. These steps are also important when going between random and sequential writes.
For all performance tests in this review, the SSS PTS was followed to ensure accurate and repeatable results.
All tests employ random data, when available. Intel's SSD DC S3700 does not perform any data compression prior to writing, so there is no difference in performance based on data patterns.


As you can see, the SSD DC S3700 offers good performance in random 4 KB reads and writes. Both capacities achieve their specifications or slightly exceed them at reasonably-low queue depths. You won't see the performance levels of a desktop drive based on a second-gen SandForce controller (a theme you'll see throughout our tests), but this still represents a significant step up from Intel's previous generation of enterprise-oriented SSDs.
Indeed, Intel claims to trade best-in-class throughput for consistency and low latency. Let's have a look at the latency tied to these results to gauge the true impact of the company's optimizations.


The SSD DC S3700 behaves more like a PCI Express-based drive, such as Micron's P320h or OCZ's Z-Drive R4, than its peers in the SATA-based space. In fact, we measure average and maximum response times that are lower than Intel's SSD 910 (1.62 and 41.77 ms, respectively).
Comparing these numbers to Intel's QoS specification, the 800 GB model is well within the 20 ms maximum time allowed for 99.9999% of round-trip writes. However, the 200 GB drive is slightly higher than the spec, at 26.21 ms. Is there a plausible explanation for this anomaly?
It's likely attributable to the passing of time. We make the drives that go through our enterprise storage review process suffer quite a bit as we generate results. In some cases, the tests take hours. Other times, they take days. Our random 4 KB write testing, which was used to generate our latency numbers, involved 24 hours of testing. We do this to more definitively address issues that might not surface during shorter tests. And if you're buying enterprise-class hardware, you want the guys reviewing those drives to be thorough. So, when we look back at our data and do a little bit of math, it's easy to see why the 99.9999th percentile still leaves a lot of commands that theoretically could take longer.
In this case, we have roughly 30,000 IOPS x 86,400 seconds, or roughly 2.5 billion operations. That leaves 2,500 operations outside of Intel's specification. We went back and reran the tests on the both the 200 and 800 GB drives and saw similar results, where just a handful (roughly 0.0001%) of commands were outside of the spec. More surprising was that nearly 99.995% of writes fell below 10 ms, which is significantly better than Intel's reference.
Our next set of tests simulates different enterprise-oriented workloads, including database, file server, Web server, and workstation configurations.
Notice that the results are pretty similar, regardless of whether you're looking at the 200 or 800 GB model. This is the case largely because high-end workloads are generally biased to read operations. Both capacities offer the same read performance, so it's no surprise to see them so close to each other.
At lower queue depths, the 800 GB SSD DC S3700 is consistently faster in each workload. It also exhibits an advantage in the file server workload. However, if your application mostly involves read operations, any of Intel's available capacities should be suitable. Our only warning would be that the 100 GB drive, which we don't have in-house, is rated for significantly lower write performance.
Our database workload (also categorized as transaction processing) involves purely random I/O. Its profile consists of 67% reads and 33% writes using 8 KB transfers.

The file server workload consists of 80% random reads of varying transfer sizes.


The Web server (100% read, varying transfer size) and workstation (80% reads, 80% random) workloads show the same basic trend.


Both Intel SSD DC S3700 drives perform well in our sequential write workloads. The 200 and 800 GB capacities slightly exceed their specifications, though neither solution yields the high-end performance we see from desktop-oriented SSDs.

Read performance, as we'd expect, is nearly identical. There isn't much to say about these numbers. Who would have thought, though, that we'd become so complacent with 460 MB/s writes and 500 MB/s reads after relying on mechanical storage for so long?
That's the thing about this drive, though. To truly appreciate what it was designed to do, you have to transcend its corner-case test results. Although the sequential performance averages are merely average, consistency is, once again, outstanding. Our exclusive Enterprise Video Streaming Performance benchmark on the next page puts this into perspective.
Enterprise video streaming is a demanding workload within the enterprise space. Companies want more HD streams with higher bit-rates and no stuttering. A storage solution well-suited for enterprise-class video delivery has completely different capabilities than something designed for databases. At the end of the day, you're basically looking for exceptional large-block sequential write performance. You also need a high level of consistency that traditionally isn't seen from consumer SSDs. For a more in-depth analysis, take a look at page 10 of Intel SSD 910 Review: PCI Express-Based Enterprise Storage.
Briefly, once the drive in question is in a steady state, we write its entire capacity 100 times. We use 8 MB transfer sizes and a queue depth of four, recording timestamps for each individual write. The graph below reflects 100-point averaging, so that you can better visualize the results.

Now this is what we're talking about. Normally, it's pretty easy to pick out the best- and worst-case runs (either by average, lowest dip, number of dips, and so on), but this drive is so incredibly consistent that's it's really hard to tell them apart. This is exactly the type of graph you want to see from an enterprise-oriented SSD. There are no major dips, all of the data points are packed pretty tightly, and performance is constant across the full drive capacity.
Even though Intel specifies performance consistency using random 4 KB writes, we're curious to see how it stacks up in an admittedly much easier workload. We take all of the 8 MB sequential writes from our worst-case run, and determine the overall average and worst-case one-second performance.
We averaged 465 MB/s across the entire drive in our worst-case run. The worst-case one-second performance was within 92.4% of the average. In fact, there were only two one-second averages below 95% of the average. Ninety percent of the one-second averages were within 99% of the average. We've been performing this type of testing for a while now, and this is the best performance we've ever seen from a SATA-based drive.
When we look at the required buffer sizes required to maintain a certain transfer speed, Intel's consistency story surfaces once more.
| Threshold | Best-Case Buffer Size | Worst-Case Buffer Size |
|---|---|---|
| 450 MB/s | 7 MB | 50 MB |
| 460 MB/s | 28 MB | 74 MB |
| 470 MB/s | 185 MB | 228 MB |
| 475 MB/s | 6,967 MB | 7,008 MB |
As you approach its average, the SSD DC S3700 requires a very small memory buffer. As soon as you exceed its average, memory requirements go up exponentially.
These are the types of tests we love doing; they separate very good drives from great drives, and you're only able to get a sense for a given product's potential when you dive in deep like this.
The SSD DC S3700 is a unique product for Intel. It's the first drive able to operate on both the +5 V and +12 V power rails. When you look at the PCB, you can clearly see the inductors for the power supply circuitry. This gives Intel a lot more flexibility in terms of deployment.

If there is one thing that hurts the SSD DC S3700, particularly when you consider how it's going to be used, power consumption would be it. As you can see in the table below, the maximum RMS (root mean squared) burst power draw is 8.2 W, while the average is 5.8 W (6 W on the +12 V rail). That's quite a bit more than Intel's SSD 710, which, for the 200 GB model, typically drew 3.5 W. It looks even worse compared tot he SSD 520, which has an active power draw of only 0.85 W.
When you multiply out that additional power use across dozens of drives or more, it may become necessary to keep a rack full of storage on a tighter power budget.
| Intel SSD DC S3700 | 100 GB | 200 GB | 400 GB | 800 GB |
|---|---|---|---|---|
| +5 V Supply (+12 V Supply) | ||||
| Active Write: RMS Average | 2.8 (2.9) W | 4.2 (4.4) W | 5.2 (5.4) W | 5.8 (6.0) W |
| Active Write: RMS Burst | 3.1 (3.3) W | 4.6 (4.8) W | 7.7 (7.6) W | 8.2 (8.2) W |
| Idle | 0.6 (0.8) W | |||
We took a look at the 800 GB model and confirmed that Intel's numbers are correct; this drive is closer to a Hummer than a Prius.
In the four years that passed since Intel launched its X25-E, the company appeared to either lose its way or lose interest in the enterprise segment. The only follow-up effort to emerge from its labs prior to now was the SSD 710, and that drive was simultaneously underwhelming and expensive. With the introduction of the SSD DC S3700, Intel fixed many of the reasons why its SSD 710 didn't succeed. And along the way, Intel's engineers helped redefine the way we evaluate the performance of solid-state storage.
The first major fix was a complete redesign of its controller. A new 6 Gb/s ASIC represents a major step forward for Intel. Not only does this help push performance far beyond what the SSD 710 could do, but it also establishes a new benchmark for consistency never-before seen from a SATA-based SSD.
We can't say enough about the SSD DC S3700's consistency. In every test, no matter how hard we tried, we kept getting reliable, repeatable results. Intel deserves major recognition for the effort that went into controller design and firmware implementation. It's only too bad that, as of now, it sounds like the company has no plans to apply its latest controller technology to a desktop-oriented SSD. We sure do hope that a consumer version surfaces at some point, though.
The last important fix was cost. Intel's SSD 710 was simply priced too close to competing SLC-based drives, particularly when you took lower endurance and higher latencies into consideration. While $2.35/GB still seems like a lot next to the fastest desktop drives out there, the SSD DC S3700's pricing structure is much better, given the 25 nm HET-MLC that Intel uses. The write endurance we've seen from this memory technology has remained consistent over the past year, and we've seen it hold up well in all but the most intensive write-heavy workloads.
There is one notable down-side to mention: power consumption is really high. This could cause issues in densely-configured racks or embedded systems. It's probably not a problem for most business users, particularly those coming from magnetic storage, where performance per watt is much lower, but you should keep it in mind anyway.
A recommendation favoring this drive cannot come from looking at its spec sheet or price tag. You have to actually use it and see how it affects the applications that matter to you. Intel's SSD DC S3700 redefines the way we look at an SSD's performance. And, in the enterprise space, there are plenty of workloads better-served by consistent performance than higher peaks (and lower valleys). As you saw in the enterprise video streaming tests, this can be a very welcome change.

