SandForce's Technology: Very Low Write Amplification
According to SandForce, SSD manufacturers can tweak firmware in a number of different ways. Naturally, then, we were curious to see whether Intel altered the way SandForce's compression technology worked.
Gauging this requires us to calculate write amplification. Usually, we'd need to endure days of testing in order to generate the numbers used for this calculation. Fortunately, all SandForce-based SSDs come with SMART counters for host writes (E9) and NAND writes (F1). Intel's SSD 520 features the same counters, so it's really only a matter of setting up Iometer to write a compressible sequential workload. Once you look them up, it's pretty easy to calculate write amplification: just divide host writes by NAND writes.
| 128 KB 100% Compressible Sequential Write 1 Hour, QD=1 | Intel SSD 520 60 GB | OCZ Vertex 3 60 GB |
|---|---|---|
| Host Writes | 1258 GB | 1301 GB |
| NAND Writes | 176 GB | 182 GB |
| Write Amplification | 0.13990 | .139892 |
Intel doesn't appear to be changing the behavior of DuraWrite, which is perfectly fine. Though the company says the SSD 520's firmware is completely its own, this particular aspect of the controller is supplied by SandForce in perfect working order, necessitating nothing in the way of tuning.
Now, at a queue depth of one, an SSD with a non-SandForce controller in it always incurs write amplification greater than or equal to one, meaning flash cells wear faster than on a SandForce-based drive. By compressing data, the SSD 520 and its contemporaries are able to write less data and extend overall endurance.
Endurance: Even Better With SandForce's Compression Technology
By minimizing write amplification, endurance is positively impacted. We can't really understate this effect. Don't believe us? Fortunately, you don't have to take our word alone. All of Intel's latest SSDs come with workload counts that allow you to estimate the life-span of your SSD.
| Intel S.M.A.R.T. Workload Counters | Purpose |
|---|---|
| E2 | Percentage of Media Wear-out Indicator (MWI) used |
| E3 | Percentage of workload that is read operations |
| E4 | Time counter in minutes |
Think of Intel's workload counters similar to a car's trip counter. Instead of distance, they measure endurance over time. We apply a three-hour workload to the drive in order to generate enough data to be meaningful.
Before we dive in, we want to clarify a few things so that you don't misinterpret what we're saying here.
First, the media wear indicator on an SSD counts down from 100 to 1. Because the number of program-erase cycles a NAND cell can withstand is limited, the MWI is designed to facilitate a rough estimate of endurance. In theory, once you reach the end of the counter, the memory's rated P/E cycles have been exhausted, though that's not to say anything bad will happen immediately after.
Second, using workload counters to estimate endurance is still a tenuous measurement (and without running any of our drives down, we're presenting this information academically, rather than practically). Iometer runs so fast and writes so much that we're essentially condensing months worth of activity into hours. Both Micron and Intel estimate that the average desktop user writes between 7-10 GB worth of information per day. So, we're basing our real-world estimates on at least 7 GB of writes by the host.
Finally, P/E-cycle ratings apply to each flash cell. But because larger SSDs employ more NAND, it takes longer to write across all cells. So, they consequently enjoy a higher endurance rating. The numbers below apply to Intel's 60 GB SSD 520, specifically.
Now, we're able to look at the following data without freaking out about SSD longevity. This is really about SandForce's technology and its effect on write amplification, and, in turn, endurance.
| Workload Ratio: 35% 128 KB Sequential, 65% 4 KB Random 128 KB Sequential: 66% Reads, 34% Writes 4 KB Random: 66% Reads, 34% Writes Full Span, QD=1, ~3 Hours | Intel SSD 520 60 GB Incompressible | Intel SSD 520 60 GB Compressible |
|---|---|---|
| Total Host Writes | 211 GB | 583 GB |
| NAND Writes | 616 GB | 100 GB |
| Write Amplification | 2.9x | 0.17x |
| Percent MWI used (E2) | 0.078% | 0.037% |
| Endurance Rating For Workload | 0.170 Years | 0.905 Years |
| Real-World Endurance Rating Estimate (7 GB Written Per Day) | 5.12 Years | 75.37 Years |
Presented with completely compressible data, Intel's 60 GB SSD 520 is told to write 583 GB of data, and actually writes 100 GB to flash. This translates into a write amplification of 0.17x. That's downright incredible considering non-SandForce will generally end up with write amplification that looks more like our incompressible workload, where 211 GB of data is written as 616 GB of to the NAND (yielding amplification of 2.9x).
In reality, you'll probably never see either of the extremes presented here. We're taxing the heck out of these SSDs, allowing no idle time for background garbage collection to affect the drive. That's an important process, responsible for further minimizing write amplification. As a result, in normal use, endurance really isn't your indicator of reliability. Again, we're really just trying to illustrate how SandForce's compression technology, which is sometimes maligned for its variable impact on performance, might also help extend endurance for SSD vendors who use lower-binned NAND.
- Intel’s SSD 520: Enthusiast Storage By SandForce?
- Test Setup And Benchmarks
- Breaking Out New Benchmarks
- 4 KB Random Performance: Raw, Windows, And Mac
- 128 KB Sequential Performance: Raw, Windows, And Mac
- Incompressible Performance: SandForce's Weakness
- Tom's Hardware Storage Bench And PCMark 7
- Power Consumption: Idle And 4 KB Random (Windows 7/Mac OS X)
- Power Consumption: 128 KB Sequential (Windows 7/Mac OS X)
- Power Consumption: Incompressible Sequential (Windows 7/Mac OS X)
- Endurance Testing: Write Amplification And Estimated Lifespan
- Steady State Performance (Worst Case) And TRIM
- Real World Performance: File Copy And Backup
- Real-World Performance: Windows And Mac Boot Times
- Not Just Another SandForce SSD
Just need more SSD's to compare, I'd like to see similar tests done with 120GB...180GB...256GB and several more brands. Further, as I mentioned before in the other article please list the exact model numbers and OEM specs including their 4KB IOPS; otherwise folks don't understand the results and if relying on this a purchasing will have in many cases a 4 in 5 chance of selecting the wrong SSD.
Prior article - http://www.tomshardware.com/reviews/sata-6gbps-performance-sata-3gbps,3110.html
We didn't have the Octane on hand in the 256 GB capacity, but we'll be sure to make that side by side comparison down the road.
Excellent point. Price is always a fickle thing.
Thanks Les.
We'll keep that mind for future reviews. However, we already list model and firmware on the test page.
Cheers,
Andrew Ku
TomsHardware.com
And with an SSD, your computer comes out of standby faster than your monitors do. Not kidding.
Yes, I'm with you on that one. I've had an Intel 320 128Gb SSD for quite some time now and have nothing but the greatest things to say about my particular experience with it. I purposely held off from buying the Marvell controller 510 in hopes that the next refresh would have a new Intel made 6Gb controller. One thing I am curious about, does the new 520 still have the Intel Toolbox software with it? I have gotten alot of use out of it with my current drive and would really hate to not have it on a new one.
You can use the Toolbox software with the SSD 520. It will however not work with other SF drives.
Cheers,
Andrew Ku
TomsHardware.com
Your SSD investigations/reviews/history lessons are Tier 1.
But I always get curious when Intel starts on and on about how it has the best NAND around. It's not that I even doubt them when they say this, but AFAIK Intel/Micron/IMFT are made and binned in the same place.
IMFT is supposed to be 49% Intel and 51% Micron-owned. Now both companies' own drives are stocked with what I presume is the best available NAND at that price point, but how did Intel get the reputation of having better NAND? (and if I'm honest, there is at least some evidence that it does). Micron doesn't run around talking up their NAND as much as they should, and this makes me think that the details of the IMFT arrangement are probably pretty strange. In a blind taste test, they taste pretty similar.
But not all Micron NAND is created equal, nor is Intel's (for example, does Kingston really get Intel's top shelf shtuff?). The IMFT NAND used in so many drives today runs the gamut from fantastic down to mediocre, only one step above Hynix's too-dirty-for-television flash. I'm probably the one person on this planet that wants to know more about Intel, Micron, and their bastard love child, IMFT.
Or it is already fixed?
Or they have replaced it with a new version?
As far as Anand can tell... no.
http://www.anandtech.com/show/5508/intel-ssd-520-review-cherryville-brings-reliability-to-sandforce
Intel's SF2281 implementation seems to be stable, and BSOD free.