Intel’s X25-M solid-state drive

Depending on who you believe, the solid-state storage revolution is on the horizon, just over it, a few years away, or already upon us. Analysts, journalists, and forum fanboys can’t seem to come to an agreement on exactly where we are in the process. One thing seems certain, though: At least for some markets, it’s not a question of if solid-state storage will take over from traditional mechanical hard drives, but when.

Of course, the mechanical hard drive is by no means dead. The torrid pace of platter development is still yielding impressive increases in performance and capacity. But as the price of solid-state alternatives tumbles, SSDs are sure to carve out a niche in the mobile space, where their superior shock tolerance and low power consumption are more important than storage capacity. Solid-state drives won’t just make their way into notebooks, though. The near-instantaneous seek time of flash memory offers tantalizing performance potential for any workload dominated by random I/O requests. SSDs aren’t nearly as impressive when it comes to sequential transfers, but they’re getting faster on that front, too, making them interesting options for high-end desktops, workstations, and servers.

So solid-state drives are going to be big. Very big. And everyone wants a slice of the pie. It’s no wonder, then, that industry giant Intel is eager to get in on the action. The company is in a rather unique position to serve the SSD market, too, with expertise not only in the chip manufacturing capability essential to populating a drive with memory, but also in the storage controller design that largely dictates performance.

Intel formally detailed its solid-state drive plans at IDF just a couple of weeks ago, revealing not just a single drive, but an entire collection of products that will trickle out in the coming months. The first of these SSDs to hit the market will be the X25-M, which boasts an impressive 250MB/s sustained read rate, a 70MB/s sustained write rate, 80GB of storage capacity in a 2.5″ form factor, and support for Native Command Queuing. Read on to see how this drive stacks up against a mix of solid-state and mechanical alternatives through our exhaustive suite of performance, noise level, and power consumption tests.

Flash inside

The X25-M is based on multi-level cell (MLC) NAND flash memory—a cheaper alternative to the single-level cell (SLC) memory used in some flash drives. Cheaper is relative, of course; we’re still talking about SSDs.

MLC memory is characterized by relatively slow write speeds, and the X25-M’s 70MB/s sustained write rate lives up to MLC’s reptuation. Calling 70MB/s relatively slow seems almost comical given that few 7,200-RPM notebook hard drives can match that speed. However, the Intel drive’s whopping 250MB/s sustained read rate quickly puts things into perspective. That’s more than twice the sustained read throughput of the latest desktop WD VelociRaptor and leagues ahead of the fastest mobile competition. (For comparison, the SLC-based 64GB Samsung FlashSSD we reviewed not long ago is rated for 100MB/s sustained reads and 80MB/s writes.) Finally, we have a drive capable of exploiting its 300MB/s Serial ATA interface.

While we’re embarassing the VelociRaptor, we should note that the X25-M has a read latency of just 85 microseconds. That’s 0.085 milliseconds, if we convert to units more commonly associated with hard drives. For comparison, the random access time of the VelociRaptor is 7.4 milliseconds—a difference of two orders of magnitude.

The X25-M naked, with its memory chips and storage controller exposed

The X25-M owes its low latency to the zippy access times inherent to flash memory. 20 flash chips can be found on the X25-M: 10 on the top of the drive’s circuit board and 10 on the back, for a total of 80GB. Intel fabricates these chips itself using a 55nm process, although rival flash giant Samsung amusingly makes an appearance with a K4S281632I-UC60 memory chip that weighs in at 16MB. Way to throw Samsung a few dollars (or more likely, pennies) for every SSD sale you beat them to, Intel.

As one might expect from a company with a long history of developing core-logic chipsets, the X25-M’s storage controller is an Intel design—and a smart one at that, with support for, ahem, SMART monitoring. More interestingly, the controller supports Native Command Queuing (NCQ)—a new trick for SSDs. NCQ was developed to reduce the performance impact of mechanical latency found in traditional hard drives, so it’s might seem like an odd choice for a solid-state drive with no mechanical parts. According to Intel, its SSDs are so fast that NCQ helps to compensate for latency encountered in the host PC. Even today’s fastest systems take some time (time is relative in the microsecond world of the SSD) between when a request is completed and another one is issued. Queuing up multiple requests can keep a solid-state drive busy during this downtime, and the X25-M is capable of stacking requests 32 deep.

More memory chips on the back

So the X25-M shouldn’t be short on performance, but what about longevity? MLC-based flash memory cells are limited to 10,000 write-erase cycles, giving solid-state drives a finite lifespan. When estimating the operating life of their drives, other SSD makers generally rely on a basic formula to calculate the number of cycles used:

Cycles = (Host writes) / (Drive capacity)

Intel says this formula oversimplifies the issue, and that two other factors must be considered. The first of these variables is write amplification, which refers to the amount of data actually written to a drive for a given write request. Intel gives an example in which a host system generates a 4KB write request that, thanks to a drive’s 128KB erase block size, actually incurs a 128KB NAND write. Dividing the NAND write size by the request size yields the amplification factor, which is 32 in this case. Intel says the X25-M’s write-amplification factor is extremely low at 1.1, while “traditional” SSDs have much higher amplification factor of 20.

The efficiency of wear-leveling algorithms also has a hand in determining an SSD’s lifespan. If a drive is going to shuffle bits around to avoid bad cells and more efficiently use those available, it must do so without wasting precious write-erase cycles. Intel estimates the X25-M’s wear-leveling efficiency factor at less than 1.1, claiming that traditional SSDs have an efficiency factor of 3.

Taking write-amplification and wear-leveling efficiency into account, Intel says the correct formula for cycling is as follows:

Cycles = (Host writes) * (Write amplification factor) * (Wear leveling factor) / (Drive capacity)

Using a write-amplification factor of 1.1 and a wear-leveling efficiency factor of 1.1, 20GB of write-erase per day for five years should consume only about 550 cycles on an 80GB X25-M. Using “traditional” SSD technology with an amplification factor of 20 and an efficiency factor of 3, the same write-erase load would use over 27,000 cycles. That’s a huge difference, and to be fair, it’s one that relies on values provided by Intel that aren’t entirely consistent. Another Intel presentation from IDF estimates that “mediocre” SSDs have a write-amplification factor of 10 and a wear-leveling efficiency factor of 5, resulting in just under 23,000 cycles for our 20GB of write-erase per day example. That presentation also pegs the X25-M’s efficiency factor at 1.04 rather than 1.1. We can’t easily test a drive’s lifespan ourselves, but we did ask Samsung for the write-amplification and wear-leveling efficiency factor values for its SSDs. Samsung hasn’t responded yet, though.

If you don’t want to crunch through the math, Intel estimates that the 80GB X25-M will last for five years with “much greater than” 100GB of write-erase per day. That’s a relatively long time for much more data than most folks are likely to write or erase on a daily basis.

Actual drive lifespans aside, Intel rates the X25-M’s Mean Time Between Failures (MTBF) at 1.2 million hours. That’s competitive with the MTBF rating of other MLC-based flash drives and equivalent to common MTBF ratings for enterprise-class mechanical hard drives.

A simple black casing holds everything together

Rather than banking on a single SSD, Intel has prepared a full lineup of solid-state drives with multiple form factors, capacities, and memory types. The X25-M will come first, with 80GB models in mass production this month, followed by 160GB derivatives early next year. Intel will also offer 1.8″ versions of this drive destined for thin-and-light notebooks and, dare we dream, netbooks as well.

Lest one market segment go without unwarranted Mountain Dew branding, Intel has an Extreme SSD in the works, too. The X25-E will only be available in 2.5″ form factors with capacity points at 32 and 64GB. This SLC-based drive maintains the X25-M’s 250MB/s sustained read rate, but boosts writes up to 170MB/s and drops read latency to 75 microseconds. The Extreme also features an MTBF rating of two million hours and enjoys SLC memory’s 100,000 write-erase cycle tolerance, making it even more attractive for enterprise environments. 32GB versions of the X25-E should enter production in the next three months, followed by 64GB flavors next year.

The X25-E certainly won’t be cheap—SLC-based drives rarely are—but we don’t yet know exactly how much it will cost. Intel has, however, confirmed that the 80GB X25-M will sell for $595 in 1,000-unit quantities. Drives will ship to the channel and major PC builders this week, and will either be sold on their own or installed in pre-built systems. We’ve also heard word that Intel partners may sell the company’s SSDs under their own names. That wouldn’t be surprising considering that we’ve already seen OCZ repackaging Samsung’s FlashSSD drives.

Test notes
We’ll be comparing the performance of the X25-M with that of a slew of direct and indirect competitors, including solid-state drives from Samsung and Super Talent, a handful of 2.5” mobile drives from Seagate and Western Digital, and a collection of the fastest 3.5” desktop drives on the market. These drives can differ when it comes to external transfer rates, capacities, cache sizes, and (for the mechanical ones) spindle speeds, and platter densities, all of which can have an impact on performance. Keep in mind the following differences as we move through our benchmarks:


Max external
transfer rate

Spindle speed

Cache size

Platter size

Capacity
Form factor

Barracuda 7200.11
300MB/s 7,200-RPM 32MB 250GB 1TB 3.5″

Caviar Black
300MB/s 7,200-RPM 32MB 334GB 1TB 3.5″


Caviar SE16 (640GB)
300MB/s 7,200-RPM 16MB 320GB 640GB 3.5″

Deskstar 7K1000
300MB/s 7,200-RPM 32MB 200GB 1TB 3.5″

FlashSSD
300MB/s NA NA NA 64GB 2.5″

MasterDrive MX
300MB/s NA NA NA 60GB 2.5″

Momentus 7200.3
300MB/s 7,200-RPM 16MB 160GB 320GB 2.5″

Scorpio Black
300MB/s 7,200-RPM 16MB 160GB 320GB 2.5″

Scorpio Blue
300MB/s 5,400-RPM 16MB 160GB 320GB 2.5″

SpinPoint F1
300MB/s 7,200-RPM 32MB 334GB 1TB 3.5″

VelociRaptor VR150
300MB/s 10,000-RPM 16MB 150GB 300GB 3.5″

X25-M
300MB/s NA NA NA 80GB 2.5″

On the solid-state front, the X25-M is joined by Samsung’s FlashSSD and Super Talent’s MasterDrive MX. The FlashSSD is an SLC-based drive rated for 100MB/s sustained reads and 80MB/s writes. The MasterDrive uses Super Talent’s original firmware and is rated for 120MB/s reads and 40MB/s writes. Pay particular attention to how these three drives stack up against each other.

To give the X25-M some additional 2.5″ competition, we’ve included the latest 7,200-RPM mobile drives from Seagate and Western Digital. Western Digital’s 5,400-RPM Scorpio Blue is also in the mix to illustrate how the SSDs look against a slower spindle speed.

Our 3.5″ drives can’t squeeze into systems that can easily accommodate the 2.5″ X25-M, but we’ve included a collection of the latest desktop models because Intel is eager to push solid-state drives for servers, workstations, and high-end desktops. It will be interesting to see how the X25-M fares against the VelociRaptor, which is the fastest mechanical SATA drive on the market.

Performance data from such a daunting collection of drives can make our bar graphs a little hard to read, so we’ve colored the bars by manufacturer, with the X25-M appearing in bright blue.

Our testing methods
All tests were run three times, and their results were averaged, using the following test system.

Processor Pentium 4 Extreme Edition 3.4GHz
System bus 800MHz (200MHz quad-pumped)
Motherboard Asus P5WD2 Premium
Bios revision 0422
North bridge Intel 955X MCH
South bridge Intel ICH7R
Chipset drivers Chipset 7.2.1.1003
AHCI/RAID 5.1.0.1022
Memory size 1GB (2 DIMMs)
Memory type Micron DDR2 SDRAM at 533MHz
CAS latency (CL) 3
RAS to CAS delay (tRCD) 3
RAS precharge (tRP) 3
Cycle time (tRAS) 8
Audio codec ALC882D
Graphics Radeon X700 Pro 256MB with CATALYST 5.7 drivers
Hard drives Seagate Barracuda 7200.11 1TB SATA
Samsung Spinpoint F1 1TB SATA
Hitachi Deskstar 7K1000 1TB SATA

Western Digital VelociRaptor 300GB
SATA


Western Digital Caviar Black 1TB
SATA

Samsung FlashSSD 64GB
SATA

Super Talent MasterDrive MX SATA 60GB
Intel X25-M SATA 80GB

Western Digital Scorpio Black 320GB
SATA

Western Digital Scorpio Blue 320GB
SATA

Seagate Momentus 7200.3 320GB
SATA

OS Windows XP Professional
OS updates Service Pack 2

Thanks to NCIX for getting us the Deskstar 7K1000 and Spinpoint F1.

Our test system was powered by OCZ PowerStream power supply units.

We used the following versions of our test applications:

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

All 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.

WorldBench overall performance
WorldBench uses scripting to step through a series of tasks in common Windows applications. It then produces an overall score. WorldBench also spits out individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results.

The X25-M gets off to a good start, scoring higher in WorldBench than any other hard drive we’ve tested. Samsung’s FlashSSD is only one point off the pace, followed by a collection of 3.5″ desktop drives.

Multimedia editing and encoding

MusicMatch Jukebox

Windows Media Encoder

Adobe Premiere

VideoWave Movie Creator

In three of four WorldBench multimedia editing and encoding tests, the X25-M turns in the quickest completion times of the lot. With the exception of the MasterDrive, however, all of the contenders we’ve assembled are pretty close.

Image processing

Adobe Photoshop

ACDSee PowerPack

The X25-M ties for the lead in WorldBench’s Photoshop test and is right in the thick of things in the ACDSee test. Scores are still relatively close across the board, though.

Multitasking and office applications

Microsoft Office

Mozilla

Mozilla and Windows Media Encoder

Completion times for WorldBench’s office and multitasking tests don’t vary much from one drive to the next, but the Intel SSD still finds itself at, or close to, the front of the pack.

Other applications

WinZip

Nero

Of all the WorldBench application tests, Nero and WinZip are the most demanding of the storage subsystem. The X25-M comes out on top in both tests, just edging out the Samsung FlashSSD, ahead of the fastest 3.5″ SATA drives on the market.

Boot and load times
To test system boot and game level load times, we busted out our trusty stopwatch.

The X25-M’s system boot time isn’t particularly impressive—it’s actually slower than that of our 5,400-RPM Scorpio Blue. However, Intel’s first solid-state drive comes back in a big way in our level load tests, which have it leading the field, a hair ahead of the FlashSSD. Note just how big a lead the X25-M has over not only our 3.5″ contenders, but also the fastest 2.5″ mechanical hard drive of the lot, the Scorpio Black.

File Copy Test
File Copy Test is a pseudo-real-world benchmark that times how long it takes to create, read, and copy files in various test patterns. File copying is tested twice: once with the source and target on the same partition, and once with the target on a separate partition. Scores are presented in MB/s.

To make things easier to read, we’ve separated our FC-Test results into individual graphs for each test pattern. We’ll tackle file creation performance first.

MLC-based flash drives are notorious for their relatively slow write performance, and the X25-M isn’t an exception to that rule. The Intel SSD isn’t nearly as slow as the MasterDrive here, and it’s surprisingly faster than the SLC-based FlashSSD with one workload. Overall, however, the X25-M’s file creation speeds can’t keep up with even the 5,400-RPM Scorpio Blue.

What the X25-M lacks in sustained write performance it more than makes up when it comes time to read. The Intel drive easily outguns the rest of the pack here, enjoying a commanding lead with all five test patterns. Not even the desktop VelicoRaptor comes close.

FC-Test – continued
Next, File Copy Test combines read and write tasks in some, er, copy tests.

Copy tests combine read and write operations, and it’s the X25-M’s slow write speeds that likely hinder it here. The Intel SSD actually fares pretty well with the Windows and Programs test patterns, which are made up of large numbers of small files. However, it’s not nearly as competitive with the ISO, MP3, and Install test patterns.

FC-Test’s second wave of copy tests involves copying files from one partition to another on the same drive.

Our partition copy results closely track with those of the standard copy test. The X25-M offers reasonable performance when copying large groups of small files, but it’s much slower with test patterns dominated by smaller numbers of larger files.

iPEAK multitasking
We’ve developed a series of disk-intensive multitasking tests to highlight the impact of seek times and command queuing on hard drive performance. You can get the low-down on these iPEAK-based tests here. The mean service time of each drive is reported in milliseconds, with lower values representing better performance.

The X25-M leads the field with iPEAK workloads that include a read-intensive Virtualdub import operation as a secondary task. Otherwise, the drive’s performance is mixed, and not nearly as consistent as that of the Samsung FlashSSD.

If we average all our iPEAK scores, the FlashSSD comes out ahead of the field with a mean service time of 0.81 milliseconds. The X25-M scores a second place finish with a mean service time average of 1.05 milliseconds—0.26 milliseconds ahead of the VelociRaptor, which isn’t bad at all.

IOMeter – Transaction rate
IOMeter presents a good test case for both seek times and command queuing.

This, ladies and gentlemen, is absolute domination. The X25-M thoroughly outclasses the competition here, wiping the floor with not only every mechanical hard drive in the field, but the other SSDs, as well.

IOMeter’s file server, workstation, and database test patterns are made up of a mix of read and write requests, so the X25-M’s ability to stay so far ahead of the competition is especially impressive there. The web server test pattern is made up exclusively of read requests, allowing the X25-M to achieve even higher transaction rates. Note that with the web server test pattern, the Intel SSD’s performance levels off after we hit 32 concurrent I/O requests—the max depth of the drive’s Native Command Queueing implementation.

IOMeter – Response time

As one might expect, the X25-M is the class of the field when we look at IOMeter response times. All our SSDs have a huge advantage with the web server test pattern, but among them, the Intel drive is still the quickest.

IOMeter – CPU utilization

We don’t usually see IOMeter CPU utilization scores higher than a few percent, but particularly with the web server test pattern, the X25-M consumes more cycles than its rivals. This is to be expected, of course, because the Intel SSD is pushing much higher transaction rates.

HD Tach
We tested HD Tach with the benchmark’s full variable zone size setting.

Although it doesn’t quite hit the 250MB/s sustained read rate Intel promises, the X25-M manages to post a blistering 229MB/s in HD Tach’s average read speed test. That gives the drive a huge lead over the field—right up until we test its sustained write performance. Somewhat surprisingly, the X25-M’s average write speed eclipses its rated 70MB/s. However, that’s only fast enough to stay ahead of our 2.5″ mechanical drives, not quite quick enough to catch Samsung’s FlashSSD or the cream of the 3.5″ desktop crop.

The Intel SSD’s burst performance leads the field, but what’s more interesting is the fact that the MasterDrive and FlashSSD both fail to muster much more than 100MB/s in this test.

HD Tach doesn’t measure seek times down to the microsecond, so the X25-M will have to make do with tying the Samsung drive with a 0.1-millisecond random access time.

The X25-M’s CPU utilization is a little higher than the rest here, but keep in mind that HD Tach’s margin of error in this test is +/- 2%.

Noise levels
Noise levels were measured with an Extech 407727 Digital Sound Level meter 1″ from the side of the drives at idle and under an HD Tach seek load. Drives were run with the PCB facing up.

Without mechanical parts, our SSDs are entirely silent. Their 42.6-decibel reading here represents the noise generated by the rest of our test system.

Power consumption
For our power-consumption tests, we measured the voltage drop across a 0.1-ohm resistor placed in line with the 5V and 12V lines connected to each drive. Through the magic of Ohm’s Law, we were able to calculate the power draw from each voltage rail and add them together for the total power draw of the drive.

Those eager to pop the X25-M into a notebook will be happy to learn that the drive sips less than one watt at idle and only about one-and-a-half watts when seeking. While those results are impressive, Samsung’s FlashSSD consumes even less power. To be fair, though, the X25-M is an 80GB drive while the FlashSSD is only 64GB.

Conclusions

We had high hopes for Intel’s first solid-state drive. As a master fabricator, the company has the chip-making chops to churn out fast, power-efficient memory cells. More than a decade of core-logic chipset design also gives Intel the storage controller mojo necessary to craft a wicked-fast SSD architecture. The X25-M delivers on both fronts, with low power consumption that should make notebook users swoon and truly inspiring performance with some workloads.

Solid-state drives have an inherent power consumption advantage over their mechanical counterparts, so the energy efficiency isn’t much of a feat. What’s more impressive is the X25-M’s performance. Thanks to a 250MB/s sustained read rate and a smart Native Command Queuing implementation, Intel’s first SSD sets a new standard for MLC-based solid-state drives. Unfortunately, though, Intel can’t escape the relatively slow write speeds that plague MLC drives, and that results in a performance profile that’s decidedly mixed.

When the X25-M is good, it’s exceptional. The drive absolutely dominated our IOMeter workloads and ran away from the field in our sustained-read-speed drag race and in our real-world file read tests. The X25-M also posted speedy game level load times and a higher WorldBench overall score than any other drive—solid-state or mechanical.

Start to stress the Intel SSD’s relatively slow write rate, however, and things don’t look nearly as impressive. The X25-M excels with iPEAK multitasking workloads that are heavy on read requests, but not those that favor writes. Its real-world write speeds aren’t all that hot, either, with the Intel drive turning in particularly poor file creation speeds in FC-Test. Flash can be very fast indeed, but the slower write speed of MLC memory is still a weak link.

Price is another problem for solid-state drives, and with the 80GB X25-M slated to sell for just under $600 in 1,000-unit quantites, Intel’s first entry in the market won’t be cheap. At that price, the X-25M sits between budget MLC-based models and their more expensive SLC-based cousins, which seems about right to me. After all, the X25-M was often faster than Samsung’s SLC-based FlashSSD, which costs nearly $800 for only 64GB.

Obviously, SSDs still have a lousy cost per gigabyte when compared to their mechanical rivals. But there’s usually a premium to be paid for performance leadership, and if you measure value in IOMeter transactions per dollar, the X25-M is an absolute bargain. This SSD’s power efficiency, shock tolerance, and silent acoustic profile will probably appeal to other markets, as well. I have a feeling boutique PC builders are going to snap these up for their uber-high-end notebooks and even desktops.

Perhaps more impressive than what the X25-M offers today is what the drive fortells for the future. Intel has a very good SSD architecture on its hands right now, and we know it plans to follow up with more products based on both SLC and MLC designs. Meanwhile, the company intends to transition its flash memory to a 32nm process node. Taken together, these steps could well cement its position as a leader in solid-state storage, just as SSDs are set to carve out a larger share of the market.

Comments closed
    • indeego
    • 11 years ago

    The price on the X25-M are already $200 less than the article statesg{<.<}g

    • d111
    • 11 years ago

    Legacy OS like Windows Vista, XP, and Applications like Microsoft Office 2003, 2007, etc. have built in, inherent flaws with regard to SSDs.

    Specifically, optimizations of these OS for mechanical hard drives like superfetch, prefetch, etc. tend to slow down, rather than help performance and is unnecessary to speed up reads in an SSD, but slow it down with unnecessary writes of small files, which SSDs are slower than a regular hard drive.

    Things like automatic drive defragmentation with Vista does nothing for SSDs except to slow them down.

    Properly optimized, even low cost 2007 generation SSDs test out as equivalent to a 7200 rpm consumer grade drive, and typical SSDs made in 2008 or later tend to outperform mechanical hard drives.

    The tests done here have done nothing to “tweak” the OS to remove design hindrances to SSD performance, and thus, have no validity or technical merit.

    What your test demonstrate is that SSDs outperform Hard Drives IN SPITE OF the deliberate slowdowns done by the OS.

    The test, as presented, would be similar to installing a 19th century steam engine on a sailing ship, and observing that it is rather slow —- without mentioning the drag and performance hits caused by the unused sail rigging, masts, etc.

    See the discussion here for a detailed discussion of SSD performance tweaks and what it takes to make them perform well with legacy OS and Applications.

    §[<http://www.ocztechnologyforum.com/forum…display.php?s=&daysprune=&f=88<]§

    • Spurenleser
    • 11 years ago

    I find the difference between Intels official values for power consumption (0.15 W max!!) and TR’s values really confusing. Did you measure power consumption with this “DIPM” feature or without? ( §[<http://www.intel.com/design/flash/nand/mainstream/index.htm<]§ )

    • srg86
    • 11 years ago

    It seems to me that unless your usage of this drive is very read intensive, then overall performance is pretty poor. Until there is less of a yawning gap between read and write performance, it’s really not worth the money imho.

      • gaspard.leon
      • 10 years ago

      as usual things are more complex then just sequential MB/s

      The issues are not sequential read/write speed… those have been pretty good even with harddrives… that only applies when you’re copying or moving large files!

      With small files (normal windows usage) the most time lost is when you seek to a file, so if your seek time is 12 ms (mechanical hard drive) then the shortest time to read 100 files is 1.2 seconds… but if your seek time is about 0.1ms then you can read all those 100 files in 10ms… while the mechanical drive is still seeking to the first file…

      that’s what you’re buying here… not capacity, not sequential write/read speed, you’re buying seek time! NO MOVING PARTS

    • Pax-UX
    • 11 years ago

    While it’s great to see these solid-state drive in action it will be at least 4 years before I would use one in my main laptop. I still don’t trust them and the price is sooo high per GB that it makes more sense to go with a 7200rpm, yeah you loss on the battery but this isn’t an issue for me.

    • Perezoso
    • 11 years ago

    On the negative side, you had to use an old ICH7R in your test rig because newer Intel southbridges are completely b0rked when it comes to SSD compatibility and performance. Hot Hardware had to use NVIDIA’s nForce 790i SLI Ultra chipset for their own review. Since we’re talking about the biggest core-logic provider in the world, I think that’s a little sad.

    • Krogoth
    • 11 years ago

    Intel has one nice SSD drive, but it is still only useful for servers and workstations.

    Mainstream 7200RPM HDDs still do the job for general users and enthusiast for less $$$$$.

    BTW, game load time is mostly I/O random seek time and CPU-bounded.

      • UberGerbil
      • 11 years ago

      Yeah, and SSDs are great at random seeks, which is probably why they dominate the level load tests in this review (which also suggests for those two games at least level loads are not yet CPU bound).

      You’re right, of course: it’s impossible to make a rational case for SSDs, at their current price points, for anything except certain niches like web servers and certain workstation loads. Fortunately, just as we saw with the original Raptors, enthusiasts don’t have to concern themselves with rationality and will buy them anyway (helping drive down prices for everyone else).

      • Meadows
      • 11 years ago

      Captain Obvious to the rescue.

        • Krogoth
        • 11 years ago

        Geez, when will you cease to do the needless jabbing?

        You got one hell of a chip on those shoulders.

        • ludi
        • 11 years ago

        With his trusty sidekick, Soda Mew!

      • swaaye
      • 11 years ago

      That’s one thing that I have noticed about even the crap SSDs in my EeePC: their random access performance and read speed allow games to load very fast. You definitely become CPU bound, especially when your CPU is a 900-1000MHz Celeron M. 🙂

      Funny thing is that when I want to save a game, it will pause for a moment as the horribly slow write speed comes into play! Insta-level-load, 5 second save game pause.

      It really is a nice little machine for playing those Quake3 and Unreal engine shooters from years ago. Along with everything else from before 2002 or so. Pairing GMA 900 up with a Celeron 900 isn’t a bad combo; it’s kinda like a Athlon XP + GF2/3.

    • axeman
    • 11 years ago

    MLC drives = *yawn*
    At current prices, I can only see these being attractive for ruggedness. I mean, the write performance isn’t always, or even usually, going to hinder performance that much, but the cost per GB? Ick. 80GB ? Windows 7 probably won’t even install on that </tongue in cheek>

    • Traumflug
    • 11 years ago

    You mentioned the drive isn’t as expensive as competing SLC drives. But you forgot to mention you have to pay rougly double the price of competing MLC drives.

    • Grigory
    • 11 years ago

    Can we get some Bigfoot sized SDDs now? Or maybe 3.5″ double height? I would love to see a 2 TB+ capacity. Then again I was declared insane just last week. Oh well.

    • shank15217
    • 11 years ago

    Damage, NCQ is limited to a queue size of 32 by protocol design. The NCQ controller has to allow a queue size of up to 32 to be compliant. In fact its limited to 31 because of a design flaw. I bet Intel would have made a larger queue size if it could. This also shows that current i/o busses don’t take flash storage into account. Expect a larger queue size and other flash based storage optimizations in newer versions of sata.

    from §[<http://linux-ata.org/faq.html#ncq<]§ "Both the host controller and device constrain the number of NCQ commands that can be outstanding. The SATA specification maximum is 32 tags. Most devices support 32 tags, but the standard permits devices to support less. Similarly, most host controllers support 32 tags. However, the ATA standard has a design flaw. The NCQ tag is presumed to be a 32-bit bitmap (32-bit dword). If all 32 tags are asserted, this produces a value (0xffffffff) that is the same value returned by reading a hardware register after the hardware has been hot-unplugged, or suffers a major failure. Thus, to distinguish this condition, libata artificially limits all NCQ configurations to 31 tags rather than 32"

      • Damage
      • 11 years ago

      Thanks, but Geoff wrote this, not me. 😉

        • shank15217
        • 11 years ago

        Oh ok for Geoff then 🙂

    • da sponge
    • 11 years ago

    Nice review. I’d really like to see it compared to 15k RPM SAS drives (both 3.5″ and 2.5″). Maybe not as important for the consumer class drives, but definitely for the enterprise drives that’ll debut later. Still, it’d be interesting to put these in perspective with SAS.

    • Hattig
    • 11 years ago

    I guess Samsung are going to have to work on their SSD controllers a little! Good review, a new baseline has been drawn. Let’s see where things go from here.

      • UberGerbil
      • 11 years ago

      Yes, the control electronics is where all the cool innovation is happening. The move to smaller process nodes makes for cheaper/larger drives, and that can drive performance (if it allows you to run more chips in parallel or make writes quicker) but all the real excitement is in figuring out ways to make controller better hide flash’s limitations. It’s a bit like improving IPC through microarchitecture innovations vs moving to a smaller node in CPU design.

    • Jon
    • 11 years ago

    I love that this technology is improving with every generation but it hasn’t proven itself to the consumer yet. 2-5 years and perhaps a couple SSD generations later I would be much more comfortable adopting the technology and using it for day to day use. Bugs need to be worked out, life expectancy needs to improve and general write performance needs to increase.

    • Nitrodist
    • 11 years ago

    Damage, you should test out the Fusion-IO’s ioDrive.

    §[<http://www.fusionio.com/Products.aspx<]§

      • jwb
      • 11 years ago

      The Windows losers won’t know what to do with it, because it only works on Linux.

        • pot
        • 11 years ago

        Your wrong, it works on x64 version of XP and Vista.

      • shank15217
      • 11 years ago

      This product has higher performance because its not limited by the sata i/o bus. It has a direct connection via pci-e 4x to main memory. Its not that surprising really.

    • liquidsquid
    • 11 years ago

    These are disks you will never want or need to de-fragment. With seek times that fast, and the risk of wear, why would you bother?

    Aside from that, these still make an excellent option for backup servers and even web servers.

    For a home user, the price premium vs. performance is still not worth it. 10 seconds of lost time in total every day would have a tough time coming up with the cost difference between 1TB of HD and 1TB of SSD.

    -LS

      • UberGerbil
      • 11 years ago

      But I don’t need 1TB. I certainly don’t need 1TB of fast SSD storage.

      A well-defined chunk of my storage is read-only, or seldom-written: Program files and Windows itself (aside from patches). The rest, the bulk of my storage consumption, isn’t speed-sensitive for the most part: media files, for example.

      I’m seriously considering putting an ~80GB SSD system/program drive into my next system alongside a larger HD for data. Some data that is mostly read-only and frequently-accessed, like my email archives and some test/work dbs, would probably live on the SSD as well.

      I don’t expect this will become a mainstream setup (unless much smarter, much better hybrid drives arrive) but I like the idea of using both storage techs and tuning the contents appropriately for their performance characteristics. At least for the next couple of years until SSDs get /[

        • Nitrodist
        • 11 years ago

        Who needs 1MB of storage? I’ll only need 640k!

          • Jon
          • 11 years ago

          I think there is a world market for maybe five computers.

          • UberGerbil
          • 11 years ago

          I said *[

        • ew
        • 11 years ago

        I agree with what your saying. An SSD really only needs to be big enough for the OS and programs.

        What I’d really like to see is some sort of hybrid drive that combines flash memory and hard drive platters and, automatically stores data where it would give the best performance.

          • UberGerbil
          • 11 years ago

          Hybrid drives exist; Intel and MS were pushing them prior to the Vista introduction (along with Readyboost and Readydrive). HP investigated them and decided they didn’t have any real benefit. I haven’t seen their research, but my impression is that these didn’t go far enough: they only had a GB or so of flash, and they weren’t especially smart about migrating frequently-read data into the flash. Essentially they were just HDs with built-in Readyboost, rather than a combination of a full SSD+HD in the same package. Of course that wouldn’t be cheap, but you probably could get away with less flash (32GB say). With all the advancements that are going on I expect the concept will get revisited.

        • d0g_p00p
        • 11 years ago

        That is pretty much my setup. I have a 64GB OCZ Core SSD drive as my main OS/apps/game drive and a nice fat external NAS setup for media.

        It’s a perfect setup. Only one disk in my main PC and big storage for all my extra PC’s. Sure my main PC does not have that much storage but for the OS and the few games installed it’s totally worth it.

    • indeego
    • 11 years ago

    Semirelated, but does everyone else hate the new Intel website? grrrrrrrrrrrrrrrg{<...<}g

      • eitje
      • 11 years ago

      yes.

    • donkeycrock
    • 11 years ago

    why didnt they ship you 2, so we could see the raid scaling?

    • computron9000
    • 11 years ago

    Good review.

    I’ll buy one when the price drops by 80%, the size is 10x is big, and the write performance is the same as the read performance. The uber-drive. Then since they use little power and (thus) make little heat, I’ll get like 10 of them and have a massive RAID10 or something.

    mmmmmmm… SSD RAID

      • moritzgedig
      • 11 years ago

      “the write performance is the same as the read”
      Then you will never buy a SSD exept a producer makes it read slower.

        • computron9000
        • 11 years ago

        Hardly. DRAM-based and other techs are evolving to make the write-issues moot.

          • srg86
          • 11 years ago

          They may do in the future, but as this review shows, it’s not moot at the moment.

    • ew
    • 11 years ago

    The CPU utilization graphs would be much more useful if they showed CPU/MB/sec or CPU/IO/sec instead of just CPU. Can’t easily tell if a drive is more CPU efficient then another or not with the current graphs.

      • UberGerbil
      • 11 years ago

      I’ve mentioned that in the context of other drive reviews. Some kind of normalization would certainly make those graphs more useful.

      • Nitrodist
      • 11 years ago

      If you go back and forth between the CPU utilization page and the transaction rate pages for the IOMeter tests you’ll notice that the X25’s results are definitely related.

    • adisor19
    • 11 years ago

    Awesome review ! I’ve been waiting for this ever since Intel mentioned it for the first time. This drive has a LOT of potential especially if Intel switches production to 32 nano.

    Any info on when will the 1.8″ form factor drives be out ?

    Adi

    • UberGerbil
    • 11 years ago

    Something I’ve been wondering, but haven’t had an opportunity to test, is how much impact filesystem metadata updates have on SSDs where write performance can be the gating factor. Specifically, or at least most acutely, the fact that Windows by default updates the LastAccessDate whenever a file is read means that you get writes even in a read-only test. This is controlled by the
    NtfsDisableLastAccessUpdate
    regkey in SYSTEM\CurrentControlSet\Control\FileSystem

    Most of the time it’s not going to make any difference (particularly on a drive that does a good job of handling concurrent IO), but I wonder if it would show up in some of the more intensive tests involving large numbers of files.

    • mwaschkowski
    • 11 years ago

    Thanks very much for the review, was anxiously awaiting.

    Could you give us some idea of real world usage, subjectively? Does the system ‘feel’ faster/snappier? I have heard rumblings of hitching or few second lock ups with other SSD drives, does Intel’s offering suffer any of that?

      • mwaschkowski
      • 11 years ago

      OK, found the answer to my question at Anands:
      §[<http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=7<]§

        • indeego
        • 11 years ago

        That is a different drive. We have MLC SSD drives installed with no hitching in any system we use them in. In “real world” use there is little difference from a HDD other than noiseg{<.<}g

          • mwaschkowski
          • 11 years ago

          Actually, no. If you read the article, you will see that it is about the X25-M, the page I was referring to was discussing some of the other drives that DO have issues I was asking about, and then the article further goes on to state that those issues do not exist with the X-25M.

    • ssidbroadcast
    • 11 years ago

    After I read the part about disk longevity I was confused. How long would this thing last?

      • Dissonance
      • 11 years ago

      Intel estimates the drive’s lifespan at 5 years with >>100GB of write-erase per day.

        • Pachyuromys
        • 11 years ago

        That figure of “much greater than (for those who dont know what ‘>>’ means) 100GB per day” is deceptively reassuring, or at best uninformative, without some additional commentary.

        People may read that and think, “Hmm, that means I can download X number of Netflix movies per day for five years…,” but (depending on how much RAM is in the system, the Operating System, other usage patterns, etc.) the computer is really writing things to the drive all the time. Windows swapfile paging back and forth from RAM to disk tops the list of course, but also intermediate file saves, file and directory datestamps, registry updates, NTFS stream data, etc., etc., etc. IINM, even every user file download in Internet Explorer writes every file twice (once to the temp directory, then to the destination directory).

        I don’t know how much Windows writes on its own to a drive on a daily basis for the “average user” (whoever that is), but I’m certain it’s a significant enough amount to affect the expected longevity of a drive with a hard limit on the number of writes it can sustain before irrevocably failing.

        It seems almost fraudulent that SSD drive makers would not disclose this information, at least as a footnote, when advertising how much data can be written to a Solid State Drive before it turns into a paperweight.

        And don’t even get me started (again) on how possible or likely it is to lose 100% of your data to a lightning strike, dying UPS battery, or other power surge as opposed to just a small fraction on a mechanical hard drive.

        Imesho, as tantalizing as they may look, SSDs are Not Ready For Prime Time. Yet.

          • IntelMole
          • 11 years ago

          /[

            • Pachyuromys
            • 11 years ago

            I said it was possible, not cheap. You may fry the drive electronics, a section of the platter and even the drive head, but you can still get most of your data back if you’re willing to pay enough for it. The same can’t be said for smoked NAND chips.

          • UberGerbil
          • 11 years ago

          Incidental writes (like page file and updated filesystem metadata) are transparently redistributed across the entire drive by the wear-levelling mechanism of the drive electronics. This is taken into account when the mfrs quote their longevity figures. You can’t kill an SSD by rewriting the same cluster over and over any quicker than writing the same number of clusters spread across many large files.

          Anything like a lightning strike that could fry the drive electronics in an SSD is just as likely to fry the drive electronics in an HD. SSDs are no more prone to loss in those cases (and are more immune to shock, dust/air pollution, and high magnetic fields). An SSD does not turn into “a paperweight” as blocks fail. At first reservoir blocks are swapped in; once those are exhausted, the drive simply shrinks. Throughout this the data on the drive remains available. An SSD that has reached the end of its life remains perfectly usable as a read-only store of data, which is a far less brittle and far more benign failure mode than typically found in mechanical drives.

          Unlike HDs, SSDs don’t have to contend with the possibility of a head crash rendering /[

            • Pachyuromys
            • 11 years ago

            Well at least you’ve got a sense of humor despite your site sycophancy. 🙂

            1) q[

            • StuV
            • 11 years ago

            Anyone who spends $10,000+ to recover a fried HD is a fool. For data that important, there is a new-fangled solution called “backup”.

            • GTVic
            • 11 years ago

            I think I understand, you were not wrong, you just misspoke. Did you compose that Hillary Clinton speech about her ordeal under sniper fire?

            Most of your complaints are speculation. Intel obviously has access to the amount of data a typical PC/server OS writes on a daily basis above and beyond the simple file saving initiated by the user. If they “misspoke” on the MTBF they would get sued out of business by the enterprise server customers who would end up on the short end.

            Very few people try to recover a damaged hard drive so that is not a selling point in your favour. Plus, you are assuming that every chip on the board would fry. That is not necessarily the case and removing each chip individually and mounting it in a recovery apparatus is feasible. You could probably even unpackage the chip and use a probe device to access individual cells if you really wanted to spend some money.

            • ew
            • 11 years ago

            l[

            • UberGerbil
            • 11 years ago

            1. They don’t need to know what OS you’re using because the amount of incidental metadata writes should be a trivial percentage of the total regardless, and pagefile usage tends to be similar among the popular OSes. No widely-deployed filesystem has so much metadata that it will make a dent in the lifespan of these drives. The pagefile is potentially more worrisome (and I’ve pointed this out in previous threads about SSDs) since it may be written to almost continuously in a worst-case scenario where the system is thrashing (lots of physical memory that is nevertheless vastly overcommitted). You’re probably not going to run a system in such a state for very long (it’s not exactly responsive), but it can happen. Intel’s best numbers (20GB/day over 5 years yields 550 cycles, or 5.5% of its 10,000 write lifetime) are rather astonishing, and suggest this isn’t a huge worry — if accurate, this represents a far bigger accomplishment in SSDs by Intel than anything related to performance — but of course those are Intel’s numbers. We certainly can quibble with them until we get independent verification, but even if Intel is inflating them several times we should still expect the drive to withstand much more than 20GB/day while reaching a 5 year lifespan (which, after all, is the longest any HD mfr will warranty their drives). From times when I’ve monitored the relevant performance counters such as page writes and total disk writes (you can use perfmon to give you instantaneous and average values for these, but it’s a little more work to accumulate totals) I haven’t seen anything like those kinds of numbers. It is of course possible, particularly on a system that is seeing a lot of other write activity as well. We might conclude such a system isn’t well suited to SSDs (or we might spec much larger SSDs, since the more available blocks the longer it takes the same amount of writes to wear them out). Not every technology is suited for every scenario. I don’t think many potential customers fall into this case; I guess we’ll see as real world experience accumulates. But I don’t think most users have a sense of how much total data they write over the course of a year anyway, so knowing how much additional “overhead” is added by the OS probably wouldn’t help.

            2. §[<http://www.google.com/search?q=forensic+flash+recovery<]§ 3. And "turning into DVD" is a bad failure mode how? Nobody has claimed these things last forever. Hard drives don't either. But when hard drives fail, there's generally no cheap way to recover the data from them. They certainly aren't helpful enough to fail their way into a ROM equivalent. You seem to be faulting SSDs for not living up to a completely unrealistic standard, and yet also disparaging them for them for failing in a more user-friendly way than their competition. ("Sycophancy"? “Stray pit bull?? My, my. I’ll cheerfully admit I sometimes defend certain technologies with gusto (in some cases, perhaps, with more gusto than they deserve). But I don’t recall spending much time off my chain chewing on the legs of passers-by who happen to be impugning the good name of Tech Report. Typically I leave that to Damage, who usually is on the scene ahead of me and has already done a far better job than I would have anyway. I try to stick to my areas of expertise, and the inner workings of Techreport is one I know nothing about. Though I do recall making a critical remark when Scott was trying to figure out whether he should retire his Widenote due to battery issues and you constructed a rather scurrilous and delusional-sounding scenario involving threats of litigation and bribes. )

          • indeego
          • 11 years ago

          “SSDs are Not Ready For Prime Time. Yet.”

          The people that use them every day disagreeg{<.<}g Those people using Worldbench sample apps that bill at hundreds an hour easily see the benefit to migrating. Office and Mozilla, zip and game load times are astounding. The majority of the people I work with would gain much more out of this than a quad core upgradeg{<,<}g not to mention power and noise improvements.

            • Pachyuromys
            • 11 years ago

            A curious conundrum of tech sites like this is that 99% of the attention is paid to the latest whiz-bang gadget, and far less than 1% is paid to failures in same said gadgets somewhere down the road (like the IBM Deskstar and more recently alledged Nvidia GPU failures).

            At some point in the future these people that are enjoying and extolling the present short-term benefits of SSDs will be lamenting their shortcomings, but when that happens I seriously doubt you’ll be seeing a full-length front-page article about that here.

            • StuV
            • 11 years ago

            Holy Hatorade Batman. Did the Techreport run over your puppy?

            • Damage
            • 11 years ago

            He’s just sore at being pw3nd by UberGerbil and somehow thinks insulting TR will annoy one of our readers.

            Good luck with that. 😉

            FWIW, we were out in front on the GXP failures story and have been watching the Nvidia GPU failures story, as well (though it doesn’t look to be the same magnitude of problem at all). Pachy seems to be confused about where he is.

            • Pachyuromys
            • 11 years ago

            To the contrary, Scott, I specifically cited those two examples /[

            • ssidbroadcast
            • 11 years ago

            q[< I, but his value as a contributor is compromised by his orientation as an unqualified TR defender. <]q Yeah the paperwork process takes FOREVER if you're looking for TR defense certification. We still have plenty of qualified TR /[

            • Pachyuromys
            • 11 years ago

            Since I can’t tell if you’re really that uneducated:

            §[<http://www.merriam-webster.com/dictionary/unqualified<]§ Definition 2.

            • ssidbroadcast
            • 11 years ago

            Wow. I got a definition for ya:

            §[<http://www.merriam-webster.com/dictionary/humor<]§ Number 3.

            • Spotpuff
            • 11 years ago

            Nice review keep up the good work.

            • Sargent Duck
            • 11 years ago

            /[<...when Sony just recalled half a million potentially flammable Vaios? Where was TR, for that matter?<]/ I remember reading about that on here, at least as a news post. Did you want them to do a review of exploding batteries? That got plenty of attention all around, TR didn't need to delve too deeply into it. ... /[

            • eitje
            • 11 years ago

            it’s a catch 22, sarge. you can’t come to the defense of TR, because if you do, you’re a TR defender.

    • MadManOriginal
    • 11 years ago

    Nice review.

    There’s something I’ve been wondering for a whlie with the storage tests though. At this point the storage testbed is just a bit dated. I understand the usefulness of using results for previous drive tests but at what point do you decide to update the storage testing platform?

      • Damage
      • 11 years ago

      Well, dated and outdated are different things. The upcoming transition to SATA 6Gbps would seem like a logical breaking point. Throwing out all of our comparative data now, with that on the horizon and no new spec eclipsing our current setup’s storage controller’s features would be unnecessarily painful.

      I’m sure a subset of our readers would choose very little history in order to have tests run on computer just like their new build, but I don’t like the tradeoff.

        • tfp
        • 11 years ago

        Are the results for the drives similar on new system then?

      • Forge
      • 11 years ago

      While the specs look dated compared to current rigs, you have to think that it’s the *storage* test rig. ICH7R has NCQ, SATA2, and any other storage features that matter. ICH8R, 9R, and 10R are largely die shrinks and very minor non-storage feature updates. ICH7R is the core tech that matters, and it’s current.

      Damage has it spot-on. The only thing that’s going to change any time soon and make the ICH7R outdated will be SATA3/SATA6Gb/s, and neither of them are here, so ICH7R is still as good as anything else.

      That said, it would be interesting to get a PCIe SATA controller or two into the picture, see how disks act on SATA controllers from other than Intel. I know it would be 100% identical for most disks, but I’ve heard mutterings that WD in particular tailors their firmwares to compliment Intel controllers. I doubt it, but it would be good to get a spot check now and then (I wouldn’t ask for the testing to be done on all the controllers, that would be a massive amount of work).

Pin It on Pinterest

Share This