Welcome to enterprise-class storage. The stakes are officially higher here. Although many large businesses continue to use conventional SAS-based hard drives, which are battle-tested in the most demanding environments, adoption of solid-state technology is picking up as the performance gains are just too significant to ignore. Transitioning to solid-state storage may seem like a daunting investment, but in applications where the random I/O of one or two SSDs can replace entire JBODs worth of short-stroked disks, they're often cheaper to buy and keep powered up.
In the desktop world, you have vendors bickering about who's using the best flash memory in their SSDs, and who's using the stuff scraped off of the foundry floor. But when you're talking about mission-critical servers, there is no room to compromise on reliability in the name of cheaper prices. Writing hundreds, if not thousands of terabytes of data necessitates eMLC- or SLC-based SSDs. And the number of vendors selling drives based on those classes of NAND can be counted on one hand.
Of course, just because there are fewer companies selling enterprise-oriented SSDs doesn't mean the competition isn't stiff. Big businesses buy drives in the thousands and are willing to pay a premium for hardware able to deliver high performance and reliability. For its part, Toshiba brings lots of experience in hard drives and NAND manufacturing to the table, giving it a unique perspective on what an enterprise-oriented SSD should be able to do.

That perspective is manifest in the company's flagship MKx001GRZB family of SSDs. Available in 100, 200, and 400 GB capacities, Toshiba arms its very high-end line-up with a couple of specifications you don't see every day (or hardly ever, really) from our desktop SSD reviews: 6 Gb/s SAS connectivity and SLC NAND.
| Toshiba MKx001GRZB Specifications | MK1001GRZB | MK2001GRZB | MK4001GRZB |
|---|---|---|---|
| RAW NAND | 128 GB | 256 GB | 512 GB |
| User Capacity | 100 GB | 200 GB | 400 GB |
| Interface | SAS 6Gb/s | ||
| Sector Size | 512, 520, 528 | ||
| Sequential Read | 500 MB/s | ||
| Sequential Write | 250 MB/s | ||
| 4 KB Random Read | 90 000 IOPS | ||
| 4 KB Random Write | 16 000 IOPS | ||
| Power Consumption (Active) | 6.5 Watts | ||
| Warranty | 5 Years | ||
Compared to what we're used to seeing from today's fastest desktop SSDs, Toshiba's MKx001GRZB line-up doesn't necessarily impress with its specified write performance. However, read speeds are roughly on par with the fastest SATA 6Gb/s MLC-based drives (that is to say, both interfaces are close to maxing out already). With regard to random read performance, specifically, it's rare to find an SSD that claims in excess of 80 000 IOPS. That Toshiba cites 90 000 is a downright impressive achievement.
In addition to its higher-end specs, the MKx001GRZB family doesn't look like your typical 2.5” SSD, either. It employs a 15 mm z-height, to begin, making it clear that the company is aiming to fit within the same form factor as current-generation 10 000 and 15 000 RPM 2.5" hard drives. That makes perfect sense, since businesses are increasingly shifting to that size in an effort to maximize density in the enterprise space.
Inside the larger enclosure, Toshiba sandwiches two PCBs together using a proprietary connector. On one board, you see Marvell's 88SS9032-BLN2 eight-channel SAS controller, on-board cache, and six SLC NAND devices; the other has ten SLC NAND devices along with four ultra-capacitors. In order to combat the thermal output of a more dense and complex design than what we're used to seeing in an SSD, each component is covered by a thermal pad able to shunt heat out toward the metal casing.


We’re told that the whole product line features NAND manufactured using Toshiba’s 32 nm fab process, but since we’re testing the 400 GB model specifically, each NAND package on our sample presents 32 GB of raw capacity. Given 16 total packages, that gives us a total of 512 GB, translating to the standard 28% of overprovisioning commonly used for enterprise-class devices.
Endurance is a term thrown around a lot in discussions of solid-state storage because we all worry about that point where an SSD is no longer able to reliably store our data. If you have an SSD in your notebook or mainstream desktop, endurance shouldn't be much of a concern. It's unlikely that you'll ever write enough data per day, every day, to exhaust the useable life of the NAND flash cells that make up your drive. Far more likely is a firmware-related issue that results in problematic operation. But even those are fairly rare.
Endurance is a much more important discussion in the enterprise world, though. Demanding workloads force many machines to read or write data continuously, day in and day out. On a conventional hard drive, other issues contribute to eventual failures. But when it comes to SSDs, those business-oriented tasks gradually chip away at the rated number of program/erase cycles that each NAND vendor affixes to its memory products. Because eMLC and SLC flash offer the highest endurance ratings, they're particularly attractive for enterprise-oriented products.
That's not to say multi-level cell NAND is out of place in professional applications. Based on our discussions with data center managers, we know there are plenty of original X25-M and SSD 320s used in mission-critical environments. They are used in such a way that a failure won't result in data loss, though, and they aren't bombarded with writes in the same way one of these Toshiba drives might be.
Evaluating SSD Endurance
Before we take a stab at quantifying the endurance of different flash technologies, we want to discuss our methodology. Our estimates come from monitoring each drive's media wear indicator (referred to as the MWI), which counts down from 100 to 1. Because the number of program-erase cycles a NAND cell can withstand is finite, the MWI is designed to facilitate a rough estimate of endurance.
In theory, once you reach the end of the counter, all of the memory's rated P/E cycles are exhausted. That's not to say something bad happens when you hit the bottom, but nobody wants to entrust irreplaceable data to a drive living on borrowed time, either. Naturally, enterprise customers place a lot of importance into the MWI, then, because it represents “the safe zone.”
| Endurance Rating (Sequential Workload, QD=1, 2 MB) | Intel SSD 320 | Intel SSD 710 | Toshiba MK4001GRZB |
|---|---|---|---|
| NAND Type | Intel 25 nm MLC | Intel 25 nm eMLC (HET) | Toshiba 32 nm SLC |
| RAW NAND Capacity | 320 GB | 320 GB | 512 GB |
| IDEMA Capacity (User Accessible) | 300 GB | 200 GB | 400 GB |
| Overprovisioning | 7% | 60% | 28% |
| P/E Cycles Observed (IDEMA) | 5460 | 36 600 | 225 064 |
| P/E Cycles Observed (Raw) | 5119 | 22 875 | 175 831 |
| Host Writes per 1% of MWI | 16.38 TB | 73.20 TB | 900.2 TB |
According to Toshiba's spec sheet, the 100 GB MK100GRZB comes with an endurance rating of 8.2 PB. Each vendor uses its own method of estimating longevity, which is why it’s difficult to compare endurance across different SSD brands and models. Our numbers assume a purely sequential workload, which means we’re ignoring random access. However, this allows us to take a step back and look at SSD and NAND endurance academically.
Look at the numbers. It’s really clear to see why SLC flash remains the crème of the crop. While it continues to fetch a high premium, SLC is also capable of withstanding many more writes than MLC technology. If you remove the effects of overprovisioning, the Toshiba’s SLC NAND has a rating close to 175 000 P/E cycles. That’s 58 times higher than Intel’s 25 nm MLC NAND, which clocks in at ~5000 P/E cycles.
Remember that P/E-cycle ratings apply to each flash cell. But because larger SSDs employ more NAND (and consequently, a lot more flash cells), it takes longer to write across all of them. As a result, larger drives enjoy a higher endurance rating. If we do the math, our 400 GB MK4001GRZB should be capable of writing 88 PB of data sequentially. That’s insanely high. And perhaps it explains why Toshiba doesn’t provide endurance ratings on its higher-capacity SSDs. Instead, the 200 GB and 400 GB models come with a guarantee that you won’t have to worry about endurance during the company's five-year warranty period (a telling promise, indeed).
| Test Hardware | |
|---|---|
| Processor | Intel Core i5-2400 (Sandy Bridge), 32 nm, 3.1 GHz, LGA 1155, 6 MB Shared L3, Turbo Boost Enabled |
| Motherboard | Gigabyte GA-Z68X-UD7-B3 |
| Memory | Kingston Hyper-X 8 GB (2 x 4 GB) DDR3-1333 @ DDR3-1333, 1.5 V |
| System Drive | OCZ Vertex 3 240 GB SATA 6Gb/s |
| Tested Drives | Intel SSD 710 200 GB SATA 3Gb/s, Firmware: - |
| Intel SSD 320 300 GB SATA 3Gb/s, Firmware: - | |
| Intel SSD 520 240 GB SATA 6Gb/s, Firmware: - | |
| Toshiba MK4001GRZB 200 GB SAS 6Gb/s, Firmware: - SAS Controller: LSI SAS 9211-8i | |
| Graphics | Palit GeForce GTX 460 1 GB |
| Power Supply | Seasonic 760 W, 80 PLUS Gold |
| System Software and Drivers | |
| Operating System | Windows 7 x64 Ultimate |
| DirectX | DirectX 11 |
| Driver | Graphics: Nvidia 270.61 RST: 10.6.0.1002 Virtu: 1.1.101 |
| Benchmarks | |
|---|---|
| Iometer 1.1.0 | # Workers = 4, 4 KB Random: LBA= Full Span varying QDs, 128 KB & 2 MB Sequential |
| 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% |
We used LSI's SAS 9211-8i HBA for testing Toshiba's drive. Without it, we wouldn't have been able to generate the long-term endurance numbers for SLC NAND. We do have other SAS cards in the lab, but they're hardware-assisted RAID controllers, which usually means SMART monitoring is disabled when drives are accessed individually. In addition, Toshiba and others recommend an LSI-based solution for the purposes of benchmarking, as that's most common to enterprise environments.

Just as server- and workstation-oriented processors have to be tested differently than desktop CPUs, so too does enterprise-oriented storage need to be evaluated in a unique way.
We got the above slide from last year's Flash memory Summit in Santa Clara, CA. The assumption for enterprise storage is that it's generally full, and often being accessed around the clock. As a result, fresh-out-of-box/burst performance is largely irrelevant. There's little to no idle time for the background garbage collection and TRIM commands to recover performance, which means an SSD in such a taxing environment is going to hit its steady-state and stay there.

Ultimately, we have to use a different methodology to get to and then test steady-state performance. The goal is to benchmark at a point where an SSD's performance no longer changes over time, necessitating constant writes in order to determine sustained performance. The chart above illustrates how, after some period of use, an SSD drops from its out-of-box performance level to a more sustainable steady-state level.
In order to attain that second point, we precondition our SSDs before running our enterprise benchmarks. But because every drive's steady-state point is different (and because there are multiple steady states, depending on the workload you run), we specifically subject our SSDs to two types of conditioning:
- For our 4 KB random, database, file server, and Web server tests, we write 3x full capacity of the drive using random writes.
- For our 128 KB sequential tests, we write 3x full capacity of the drive sequentially.

Toshiba's MK4001GRZB is able to take full advantage of its SAS 6 Gb/s interface in this test, exceeding the performance of its SATA-based competition running at the same data rate. Presented with a queue depth of 32, performance starts to plateau at roughly 95 000 IOPS. The nearest competition is Micron's SLC-based P300, which starts higher than Toshiba's drive at lower queue depths, but falls just short of 60 000 IOPS with 16 outstanding commands or more.
Intel, the indisputable favorite amongst IT professionals shopping for enterprise-class SSDs, has a very driven focus on reliably. So much so, in fact, that its client-oriented drives are being used in certain server-based environments. So, we thought it'd be fitting to include SSD 320 and SSD 520 into our benchmarks for the sake of comparison. Interestingly, the company's SandForce controller-based drive achieves up to 50 000 IOPS using Intel's binned MLC flash. That falls shy of the P300, but it's a substantial improvement over the SSD 320 and more business-class SSD 710. And when we hit the drive with incompressible data, indicated by the light blue line marked Random, performance remains similar. The SandForce controller's reliance on compression for exceptional performance doesn't become a factor until we analyze write speed.

The MK4001GRZB has a rated random write speed of 16 000 IOPS at a queue depth of 16. According to our Iometer testing, that figure applies to all queue depths once the drive hits its steady state. In comparison, Micron's P300 performs much better, as it hits speeds just over 20 000 IOPS.
As an aside, Intel's SSD 520 really struts its stuff in this test thanks to SandForce's second-gen controller. At a queue depth of 64, the desktop-class SSD plateaus at speeds just over 50 000 IOPS.

Once we hit the steady state for 4 KB random I/O, the P300's average response time appears just slightly lower than Toshiba's MK4001GRZB, though the delta is only 25%.
Yet, in the same environment, peak response time measurements actually favor the SLC-based SSD. Going by the numbers, the max response time for the MK4001GRZB is 53.7 ms, which is almost 8x lower than the P300.
Although we don't mean to keep taking the emphasis off enterprise-class storage, we're again surprised by Intel's relatively new SSD 520. Depending on data type, the drive's maximum response time falls between ~125-155 ms. This puts the SSD 520 on par with Intel's SSD 710, and slightly better than its client-oriented SSD 320.

128 KB Sequential

At a queue depth of one, Toshiba's SSD offers sequential read speeds just north of 200 MB/s, putting it on par with the SSD 320 and 710. As you scale up, however, the MK4001GRZB's performance peaks and plateaus at 510 MB/s. That's substantially better than Micron's P300, which is only able to reach a top speed of 450 MB/s.
In read-heavy enterprise workloads, Intel's SSD 520 looks like an attractive option. It's able to nearly match the much more expensive MK4001GRZB when there are more than eight outstanding I/O commands.

Although performance plateaus at a queue depth of two for all of our tested SSDs in this 128 KB sequential write test, there are substantial differences between the various models. Toshiba's MK4001GRZB falls just shy of 300 MB/s, while Micron's P300 pushes closer to 350 MB/s.
Interestingly, the SandForce-based SSD 520 hits speeds just over 500 MB/s when it's presented with compressible data. At the other end of the spectrum, when you hammer it with incompressible information, the SSD 520 barely outperforms the SATA 3Gb/s-capable SSD 320 and 710.

Moving to a larger block size makes the effect of queue depth less important. Using 2 MB transfers, Intel's SSD 520 leads the pack with a sequential write speed close to 550 MB/s (so long as you're working with compressible data, that is). The MK4001GRZB falls right behind at 520 MB/s, which roughly matches the performance of the SSD 520 as it operates on incompressible data.
Although Toshiba's offering doesn't top this chart, it still outperforms the Micron P300, which plateaus at sequential read speeds of 450 MB/s.

Sequential 2 MB writes look a lot like the 128 KB chart without the impact of queue depth weighing on performance. The SandForce-based SSD 520 still reigns king when it comes to compressible data, though switching to incompressible information knocks Intel's newest desktop drive closer to the bottom of the chart.
Amongst the more purpose-built enterprise SSDs, Micron's P300 delivers the best performance at 350 MB/s. In comparison, Toshiba's MK4001GRZB falls a ways behind with speeds just shy of 300 MB/s.

Because SSDs in an enterprise environment are assumed to be active 24x7, idle power consumption doesn't receive the emphasis that it might in a desktop or notebook. Even so, it's interesting that Toshiba's enterprise SSD is the only drive that draws more than 1 W without doing anything at all.

When it's busy crunching 4 KB random access, Toshiba's MK4001GRZB uses more than two times the power of Micron's P300. Both drives employ SLC NAND, so the difference isn't necessarily attributable to Toshiba's choice in memory technology.


Switching to sequential accesses, the MK4001GRZB consumes substantially more power than all of the other tested SSDs.
If you flip back through the performance analysis, you find that, at a queue depth of eight, Intel's SSD 520 is roughly as fast as the Toshiba drive in this very same workload. Compare that to our power numbers and you find that the SandForce-based desktop drive uses a lot less power to achieve similar performance, making it a more efficient solution in read-heavy workloads.
We see a similar situation evolve in sequential write testing. The P300 outperforms the MK4001GRZB by roughly 19%, yet Toshiba's drive consumes almost 42% more power.


Larger block sizes don’t affect power consumption by much. At a queue depth of eight, 128 KB and 2 MB performance is roughly the same, which means efficiency is, too.

Our last batch of synthetic tests subjects each SSD to the standard database, file server, and Web server profiles in Iometer.
Our Iometer 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.

Despite the Toshiba drive's strong random performance, Intel's SSD 520 manages to jump into a compelling lead thanks to the compressibility of our data. Even if you hit it with incompressible information instead (the blue bar labeled Random), it still matches pace with Micron's P300, though. Interestingly, the MK4001GRZB falls behind, stymied by the mixed workload.

The file server test, which is also dominated by random I/O, is even more biased toward read operations. However, the gap separating Intel's SSD 520, Micron's P300, and Toshiba's MK4001GRZB is smaller.

The Web server profile consists completely of random reads. If you remember back to the first chart on page five, where Toshiba's drive absolutely dominated, we quickly come to understand why the MK4001GRZB jumps in front here as well. This is clearly an environment where Toshiba's enterprise SSD operates at its peak potential.
The cadre of IT managers who make purchasing decisions for big enterprises don't just read reviews and buy pallets of storage devices. Rather, they spend weeks and months with new technologies in isolated servers, testing their mettle before deploying into production. In many performance-sensitive workloads, SSDs make a lot of sense. They can even save money when they replace a much larger quantity of disk drives. But reliability is of the utmost importance and, by extension, endurance receives attention as well.
As a result, it's difficult to render final judgement on an enterprise-class SSDs. Performance isn't the headliner that it is in desktop environments. Rather, it shares the spotlight with data security, and that's a very difficult variable to quantify.
We can determine performance over the course of a weekend (even taking steady states into account). If the only thing you care about is raw throughput, you could conclude that Toshiba's MK4001GRZB delivers excellent read speed, but is generally matched by desktop-class drives like Intel's SSD 520 that cost a lot less and facilitate better efficiency. The MK4001GRZB sells for more than $7000, while a 480 GB SSD 520 is available under $1000. A 200 GB P300 goes for somewhere around $2000, which can't stand up to the Intel drive, but it also employs SLC NAND, too.
Reliability is the real challenge. Statistically, a handful of reviews is insufficient for crowning one company or another the best for keeping data secure. To really gauge reliability, you'd need to watch the failure rate of a large population of SSDs subjected to the same workload. Why? Because solid-state storage changes its behavior based on activity, which cannot be said for hard drives. Right now, vendors only seem willing to cite the return rates from distributors, which generally involve fairly small sample sizes. Invariably, it'll take more time and a study like Google's own independent analysis of failure trends to shed more light on how SSDs compare.
Endurance is related to reliability, but certainly not the only (or most important, even) determinant of it. It's possible to test and estimate the rated longevity of an SSD using SMART (Self-Monitoring, Analysis and Reporting Technology) tables and a bit of math. Unfortunately, it is a very time-consuming process. In order to give you an idea of what it took for us to present endurance figures for Toshiba's drive, we had to write to the MK4001GRZB for 41 days, 24 hours a day, to get the MWI to drop 1%. In the process, we wrote approximately 900 TB worth of data. And that figure only applies to a purely sequential workload. Estimating endurance for random access would require a separate test, as write amplification is higher. We're not sure how much longer testing would have taken, but it could have been as long as three or four months.
As a result, our conclusions are and will always be admittedly less complete than a review conducted over the course of months or even years. But that's the trade-off for also publishing something in a timely manner. This could have been a great candidate for a long-term story where we put the rubber to the road and go ahead with calculating random I/O endurance as well. Alas, with a $7000+ price tag, it's understandable that Toshiba wanted to get the drive back sooner than later.
At the end of the day, based on our testing, we can say that Toshiba's MK4001GRZB offers very fast reads, slower write performance, and amazing endurance in sequential workloads. The last point can't be understated. In an environment pushing sequential writes all day, every day, it'd take more than 11 years to use up the 400 GB model's rated P/E cycles. That's well beyond Toshiba's five-year warranty. And so, when it comes to enterprise storage, the MK4001GRZB shows us why SLC flash is still top-of-the-line.


