Sign in with
Sign up | Sign in
The SSD DC S3500 Review: Intel's 6 Gb/s Controller And 20 nm NAND
By ,
1. Intel SSD DC S3500: Focusing On Read Performance

This may come as a surprise to enthusiasts focused on cutting-edge consumer drives, but the 3 Gb/ss Intel SSD 320 family is still incredibly popular in the enterprise. Even though it's only two years old, though, the architecture's performance has not aged gracefully. A quick rundown of its four-corner specifications tells a sad story:

  • Sequential Reads: 270 MB/s
  • Sequential Writes: 200 MB/s
  • Random Reads (100% Span): 39,500 IOPS
  • Random Writes (100% Span): 400 IOPS

Wait, what? Yeah, you read that correctly. Four hundred IOPS for 4 KB random writes across all LBAs at a queue depth of 32. So why in the world are IT professionals not only buying these drives still, but buying them in the thousands of units? The answer isn't as straightforward. Even though the 320's performance isn't particularly impressive, the series covers the rest of its bases fairly well. Once Intel worked out its firmware issues, the SSD 320s became solid and reliable workhorses, and we've heard many anecdotal stories from large corporations about their reliability.

The SSD 320s clearly suffered an unfortunate identity disorder, too. Was it an enterprise drive or something intended for consumers? It had power loss protection and full-disk encryption, so it must be business-class hardware, right? At the same time, it replaced the X25-M, so surely it was intended for enthusiasts. In reality, it was a bit of both. You just had to do some reading in order to figure that out.

Intel spent the last two years trying to sort out its product channels. It's telling a clearer story now than even a year ago. And now enterprise customers are getting a true replacement for the SSD 320s in its SSD DC S3500. There is no confusing the issue on this one; it's business-oriented through and through. The DC stands for data center, so it sort of has to be.

The SSD DC S3500 is targeted mainly at read-intensive and mixed-workload applications. Anything more write-heavy is kicked up to the SSD DC S3700 (Intel SSD DC S3700 Review: Benchmarking Consistency). A few short months ago, when a big business wanted storage for the sort of role the S3500 is designed to fill, they were limited to consumer drives. However, since the start of the year, we've seen Seagate launch the 600 Pro (Seagate 600 Pro-Series 200 GB SSD Review: For The Enterprise) and Samsung introduce its 843. Along with the SSD DC S3500, we see those drives nosing out the desktop-oriented SSDs from enterprise rotation. 

Intel's latest entry comes with all of the bells and whistles expected from a pricier enterprise drive. You get end-to-end data protection, power loss protection, 256-bit AES encryption, ECC memory, a 2 million hour MTBF, and a five-year warranty. It's good to see Intel integrate all of that reliability-boosting technology, considering this is still an entry-level offering wish pricing not much higher than the desktop-class stuff we typically review.

Most SSD manufacturers give you a handful of options when it comes to configuring solid-state storage. With Intel, that's an understatement. In the 7 mm, 2.5" form factor, you can pick between 80, 120, 160, 240, 300, 480, 600, and 800 GB capacities. In the 5 mm, 1.8" form factor, there are 80, 240, 400, and 800 GB models. This wide range of choices lets Intel target applications ranging from industrial embedded to data centers to blade servers.

Unlike many enterprise SSD manufacturers, Intel always discloses pricing information up-front. While we don't have MSRPs for every capacity point, we do know that the 480 GB model we're reviewing should run around $579. At ~$1.20/GB, Intel is quite competitive next to the other read-focused enterprise SSDs. When you take into account the warranty and reliability-focused features, you might even be tempted to snag one for your next desktop build. Before we go down that path, though, let's look at the specs.

Intel SSD DC S3500 Line-Up
User Capacity (GB)
80
120
160
240
300
480
600
800
Interface2.5"  6 Gb/s SATA
Sequential Read (MB/s)
340
445
475
500
500
500500
500
Sequential Write (MB/s)
100
135
175
260
315
380
410
450
4K Random Read (IOPS)
70,000
75,000
4K Random Write (IOPS)
7000
4600
7500
9000
11,000
11,500
Power Consumption (Active)1.8 W
2.0 W
2.3 W
2.9 W
3.5 W
4.3 W
4.5 W
5 W
Power Consumption (Idle)0.6 W
Write Endurance (TBW)
45
70
100
140
170
275
330
450
2. Inside Intel's SSD DC S3500

If you've already read Intel SSD DC S3700 Review: Benchmarking Consistency, then you might want to skip ahead. The SSD DC S3500 and S3700 don't just share a few similarities; they share almost every single component. Starting on the exterior, they employ the same aluminum enclosure, right down to the part number.  Can you tell which is which?

Intel SSD DC S3700 (left), Intel SSD DC S3500 (right)Intel SSD DC S3700 (left), Intel SSD DC S3500 (right)

As with the S3700, we see two through-hole 35 V 47 uF capacitors notched into the edges of the PCB.

On the inside, we see three black, plastic stand-offs covering each of the screw holes. It's easy to observe that the PCB is identical to the one found on Intel's SSD DC S3700. Even the reference designators on the silk screen match. Both SSD DC S3000 families utilize the same PC29AS21CA0 controller, too. This Intel-developed, eight-channel, 6 Gb/s processors performed well in our S3700 review, exhibiting excellent consistency.

Next, we take note of two DDR3-1600 DRAM packages from Micron (MT41K512M8RA-125). Each FBGA module hosts 512 MB of memory, totaling 1 GB of cache on the SSD.

Up until now, the only difference between both drive families was the DRAM they use for cache. But with the SSD DC S3500, Intel is replacing the 25 nm HET-MLC found in the S3700 with 20 nm MLC. This is what gives Intel the ability to hit lower price points with its SSD DC S3500. And as we'll see shortly, it's also the reason why write endurance is so much lower.

As with the SSD DC S3700, some capacities of the S3500 have an odd assortment of NAND packages. The 480 GB version we have in the lab leverages fourteen 32 GB modules, one 64 GB module, and one 16 GB module. That adds up to 528 GB, yielding 9% over-provisioning. And that's substantially less than the SSD DC S3700 at ~22%.

3. Test Setup, Benchmarks, And Methodology
Test Hardware
ProcessorIntel 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 S3500, 480 GB
Graphics
AMD FirePro V4800 1 GB
Power Supply
OCZ ModXStream Pro 700 W
System Software and Drivers
Operating SystemWindows 7 x64 Ultimate
DirectXDirectX 11
DriverGraphics: ATI 8.883
Benchmark Suite
Iometer v1.1.0
4 Workers, 4 KB Random: LBA=Full, Span Varying Queue Depths 
ATTO
v2.4.7, 2 GB, QD=4
Custom
C++, 8 MB Sequential, QD=4
Enterprise Testing: Iometer Workloads
Read
Write
512 Bytes
1 KB
2 KB
4 KB
8 KB
16 KB
32 KB
64 KB
128 KB
512 KB
Database
67%
100%
n/a
n/a
n/a
n/a
100%
n/a
n/a
n/a
n/a
n/a
File Server
80%
100%
10%
5%
5%
60%
2%
4%
4%
10%
n/a
n/a
Web Server
100%
100%
22%
15%
8%
23%
15%
2%
6%
7%
1%
1%

The Storage Networking Industry Association (SNIA), a working group made up of SSD, flash, and controller vendors, has 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:

  1. Purge: Purging puts the drive at a known starting point. For SSDs, this normally means Secure Erase.
  2. Workload-Independent Preconditioning: A prescribed workload that is unrelated to the test workload.
  3. Workload-Based Preconditioning: The actual test workload (4 KB random, 128 KB sequential, and so on), which pushes the drive towards a steady state.
  4. 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’s incredibly easy to not fully condition the drive and still observe out-of-box behavior, which may lead one to think that it’s 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 S3500 does not perform any data compression prior to writing, so there is no difference in performance-based data patterns.

4. Results: Write Endurance

We typically spend a lot of time looking at write endurance when we review enterprise-class SSDs. Write endurance is one of the major differentiators separating enterprise and client-oriented drives, after all. As MLC-based storage continues pushing its way into spaces previously filled by SLC NAND, we have to keep a close eye on this difficult-to-benchmark, but still very important variable involved in evaluating solid-state storage.

The rise of read-focused enterprise drives makes the testing we're doing even more important, since you really want to know what writes will do to storage hardware designed for read-heavy workloads. Naturally, we have to appreciate the companies that treat write endurance as a first-class specification, and much of the credit for that goes to JEDEC for its JESD218A write endurance testing standard. Instead of issuing vague ratings, we now see most companies specifying their drives to JESD218A, which uses the JESD219A enterprise workload to quantify endurance. This closely matches the types of workloads we use in our Enterprise Workload Performance tests, employing sequential write patterns and large block sizes. The result is minimal write amplification and wear leveling, yielding a better indication of actual P/E cycles for the NAND.

Endurance Rating
Sequential Workload, QD=1, 8 MB
Intel SSD DC S3500Seagate 600 Pro
Intel SSD DC S3700
NAND Type
Intel 20 nm MLC
Toshiba 19 nm MLC
Intel 25 nm HET-MLC
RAW NAND Capacity
528 GB
256 GB264 GB
IDEMA Capacity (User Accessible)
480 GB200 GB200 GB
Over-Provisioning
9%28%32%
P/E Cycles Observed (IDEMA)
3,326
6,24536,343
P/E Cycles Observed (RAW)
3,024
4,87927,532
Host Writes per 1% of MWI
15.97 TB
12.49 TB72.69 TB
$/PB-Written
$362.55
$228.18$64.66

As we saw in Seagate 600 Pro-Series 200 GB SSD Review: For The Enterprise, read-focused SSDs are ill suited for write-intensive applications. Compared to the SSD DC S3700, you get roughly one-tenth of the endurance for half of the price. We don't consider that a good deal. Fortunately, you don't buy this drive for its great write endurance. You simply want to know what it'd do if you were to tax it with writes.

The bad news is that, up against Seagate's 600 Pro, write endurance looks even worse for the SSD DC S3500. Seagate's drive holds a commanding 60% lead in P/E cycles observed.

Now, we'll step back and consider what the two companies say about endurance. They both present specifications using the JESD218A standard, and they both sell 480 GB SSD. Intel states its write endurance at 275 terabytes written (TBW), while Seagate states 350 TBW. That's not the 60% we observed, but it's still a large 27% difference.

It's more difficult to compare write endurance between capacities. With Intel, the odd NAND configurations change over-provisioning at each capacity point. This means that the 800 GB model has exactly twice the endurance rating as the 400 GB version, but the 480 GB drive is slightly less than double that of the 240 GB SSD. Seagate, meanwhile, uses a complex endurance rating system based on the amount of factory over-provisioning. The 200 GB 600 Pro that we tested is rated at 520 TBW, while the 240 GB version, with the exact same amount of NAND, is rated at only 134 TBW. Even this ratio of TBW varies across capacities, from 3x to nearly 10x.

What does all of this information tell us about write endurance? Like all tests that involve random, small-block workloads (JESD218A), write endurance is heavily affected by over-provisioning. This is one of the reasons we test with large-block sequential workloads. The low write amplification helps us see through any configuration FUD. We weren't able to put that theory to the test by over-provisioning the SSD DC S3500 to 600 Pro levels in this story, but expect that analysis in the near future.

Bringing it all back to Intel's SSD DC S3500, by all measurable and stated results, Seagate's 600 Pro provides higher write endurance. With that in mind, how important is write endurance on a drive that was not made for write-intensive applications? Now more than ever, this SSD encourages you to be educated about the workloads you apply to it, else you spend a lot of time and money on replacements.

5. Results: 4 KB Random Performance And Latency

We expect an SSD that exists for the sole purpose of serving up reads to do this job well in a random 4 KB read test. Intel's SSD DC S3500 does not disappoint. It matches the S3700 at every queue depth, eventually hitting 77,000 IOPS. That's not enough to knock Seagate's 600 Pro off its perch at queue depths above 16. Peaking at 84,000 IOPS, the 600 Pro still dominates this test.

Random 4 KB writes are much harder to interpret. The 600 Pro we recently reviewed was of the 200 GB variety, and its extra factory over-provisioning boosts random write performance by 3x. That also pushes price per gigabyte up to ~$1.60/GB, which is 33% higher than Intel's SSD DC S3500. We still don't have any of the non-over-provisioned 600 Pros to review, but those drives are more in line with the S 3500's pricing and are rated at 11,000 IOPS. At the end of the day, you get what you pay for. The 600 Pro family has an advantage in that you have an option to pay a little more per gigabyte and get a big boost in random write performance.

This made us wonder what would happen if we over-provisioned the SSD DC S3500 by an additional 20%? Would we see the additional gains that the 600 Pro achieves? The short answer is no. No matter what we did, we couldn't get much above 11,000 IOPS.

The average response time lines up perfectly with what we recorded for random 4 KB IOPS. And as with our IOPS measurement, the SSD DC S3500 trails the field by a fairly wide margin. A little more troubling was a maximum response time almost double that of the 600 Pro. Before we get too concerned, lets take a look at performance consistency and see if we have something to worry about.

6. Results: Performance Consistency

Increasingly, we pay close attention to the performance consistency of enterprise-class SSDs. This is what separates a good drive from a great one when all of the corner case testing seems equal. Over the past year, we measured this in terms of large-block transfers in our Enterprise Video Streaming section. Armed with that data and our exclusive analysis, the peaks, valleys, and frequency of each became clear. If you look at the information for long enough, you start to see fingerprints for each drive.

We started with large-block transfers because, in enterprise video applications, if you don't buffer or write data fast enough, you can lose it completely. Random 4 KB transfers are slightly more academic, but they also emulate database transfers pretty well. With this sort of workload, you might not actually lose anything, but your system will certainly slow down.

In the following tests, we subjected our enterprise SSDs to 25 hours of continuous random 4 KB writes across each drive. We recorded the IOPS every second, giving us 90,000 data points. We then zoomed in to the last 60 minutes to more coherently visualize the results. 

As the graph clearly shows, Intel and Seagate take completely different approaches to latency. At one end, you have the SSD DC S3700, which delivers rock-solid performance with very little variation. On the other end, you have the S3500 that has higher latency and variance. When you zoom in on the data, though, the those "fingerprints" I mentioned earlier look almost identical. Even the histograms look similar. This comes as no surprise, since both Intel SSDs use the same controller.

Then there is Seagate's 600 Pro. Even though its average latency and IOPS are closer to the SSD DC S3700, the variance is so large that it approaches the maximum observed from Intel's SSD DC S3500. 

A look at the histogram makes it easy to notice the two-level distribution of latency. Two peaks around 2.7 ms correlate to ~12,000 IOPS, while the peaks around 3.25 ms represent slightly less than 10,000 IOPS.

Intel makes it a point to specify the performance consistency of its enterprise SSDs. With the 480 GB SSD DC S3500, IOPS in the 99.9th slowest one-second interval should be within 75% of the overall average. In our testing, we recorded results closer to 80%.

7. Results: Enterprise Workload Performance

Our next set of tests simulates different enterprise-oriented workloads, including database, file server, Web server, and workstation configurations.

The 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.

Compared to our random performance benchmarks a couple of pages back, the SSD DC S3500 comes a lot closer to Seagate's 600 Pro. But it's clearly the slowest drive on our list still.

In the file server workload, which consists of 80% random reads of varying transfer sizes, Intel's latest trails the other SSDs by a fairly wide margin.

The Web server workload (100% read, varying transfer size) doesn't do much to differentiate these enterprise-class SSDs. Both Intel drives yield nearly identical performance, but also trail the 600 Pro.

The workstation benchmark (80% reads, 80% random), proves no match for the SSD DC S3700, while the S3500 and 600 Pro are more evenly matched. Intel's S3500 actually pulls out a rare win against the 600 Pro.

8. Results: Sequential Performance

As with our random read numbers, the SSD DC S3500 and S3700 demonstrate almost identical sequential read performance. And perhaps that's the point; an SSD DC S3500 gives you comparable read performance at a lower price. The thing is, neither Intel SSD matches Seagate's 600 Pro, especially at low transfer sizes.

The performance curves come much closer to each other on sequential writes, though the 600 Pro is still roughly 10% better at larger transfer sizes.

The real problem Intel is going to have with its SSD DC S3500 won't be the 480, 600, or 800 GB capacity points. Rather, it'll be the smaller drives. Even though Intel offers a comprehensive line-up, the performance of lower-capacity models trails off fast. Let's compare the 128 GB Seagate 600 Pro with the 120 GB Intel SSD DC S3500.   


Intel SSD DC S3500 120 GB
Seagate 600 Pro 128 GB
Sequential Read
445 MB/s
520 MB/s
Sequential Write
135 MB/s
300 MB/s

That's a hard sell when the 600 Pro is only slightly more expensive. Seagate does a great job of enabling outstanding sequential performance at the bottom of its 600 Pro family.

9. Results: Enterprise Video Streaming Performance

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.

Once the drive 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.

When we first saw the results of our streaming performance tests, we were shocked. In fact, we performed this test on multiple machines, behind HBAs, on different drives, and with different test software. No matter what we did, we always saw this oscillation.

Even though our overall average was ~430 MB/s, we had some wild swings between 400 and 475 MB/s. The problem with this behavior is that a good portion of the writes are at or below the rated speed of 410 MB/s. How does this translate to sustained performance?

Threshold
Best-Case Buffer Size
Worst-Case Buffer Size
375 MB/s
11 MB
17 MB
400 MB/s23 MB31 MB
410 MB/s162 MB172 MB
420 MB/s492 MB771 MB
430 MB/s1239 MB1614 MB

Even though there are prolonged periods of performance well above this drive's specification, the dips below gate the SSD DC S3500's performance. But, even minimal amounts of software buffering should allow you achieve its stated performance. Just don't go much above, or you will pay the price very quickly.

10. SSD DC S3500: Not Quite An S3700 Or 600 Pro

On paper and in the lab, Intel's SSD DC S3500 is, hands down, superior to the SSD 320 family it replaces. It's so much better, in fact, that we didn't even include the SSD 320 in our comparison. There are two reasons for this. First, IT professionals buying the SSD 320 today will happily switch to SSD DC 3500s. You get a better SSD from Intel at a 20% discount. That should be an incredibly easy sell. Second, there is just so much competition in the enterprise SSD space today. Nearly every vendor we talk to is coming out with some sort of MLC-based read-oriented drive in 2013. The landscape is changing so quickly that it's difficult to draw a line in the sand and declare a winner (which is what review sites love to do).

It's even hard to make a direct comparison to Seagate's 600 Pro. Seagate ships multiple versions at each capacity point with different amounts of over-provisioning. The 600 Pro we reviewed had three times the over-provisioning as Intel's SSD DC S3500. In random write workloads, the difference is pronounced, even though we know that's not what these drives are meant for.  

When you focus on random writes, at the same $/GB, the SSD DC S3500 and 600 Pro are pretty evenly matched. What Seagate gives you is the opportunity to pay a little more for even higher performance. That is something that you don't often see in the SSD space. Typically, performance scales with capacity, not over-provisioning. For a 33% premium, the 600 Pro can deliver two to three times the random write performance of Intel's SSD DC S3500.

Sequential writes also favor the 600 Pro, though in this case it doesn't matter which version  you pick. It takes an 800 GB SSD DC S3500 to match the sequential write performance of Seagate's 200 GB 600 Pro. As you go down the line, at matching capacity points, the 600 Pro delivers anywhere from 10 to 100% more sequential write performance. That's impressive.

Enterprise customers shopping for a read-focused SSD probably won't care much about our write benchmarks. In fact, both Intel and Seagate emphasize the read capabilities of their drives. So, let's turn our attention there. In read operations, you get roughly 10% more performance across the board from Seagate's 600 Pro than Intel's SSD DC S3500. There are certain scenarios where the results come closer than that, but, at worst, there is still a discernible difference between both drives.

As much as we wanted to love Intel's new SSD DC S3500, the S3700 set a mighty high bar, and this drive just doesn't measure up. It's not that there's anything inherently wrong with the S3500, but it doesn't out-muscle Seagate's new 600 Pro. That's not to say Intel's latest won't sell well; believe us, it will. So, if you need a very specific capacity or form factor, you'll probably find an SSD DC S3500 able to meet your specifications. For everyone else, there is Seagate's 600 Pro.