Given the firm’s expertise in both storage controller design and semiconductor fabrication, it was probably only a matter of time before Intel got into the solid-state storage business. The chip giant finally burst onto the SSD scene a little less than a year ago with the introduction of its X25-M line of 2.5″ multi-level cell (MLC) flash drives. A performance revelation at the time, the X25-M easily captured the SSD performance crown. In fact, one could argue that it’s still the SSD to beat, even today.
Much has changed in the 11 months since the X25-M’s initial launch, though. Samsung has a new SSD design that’s being sold by Corsair and OCZ, and it offers better sequential throughput than the X25-M in both synthetic benchmarks and with real-world file operations. A Korean firm by the name of Indilinx has also introduced a Barefoot SSD controller that’s being used by nearly half a dozen drive makers. The Indilinx has been quite popular in certain circles, and with strong random-write performance and the promise of TRIM support for Windows 7, it’s become an intriguing alternative to the X25-M.
Intel has been busy since the X25-M’s launch, too. In November of last year, it unleashed a single-level cell (SLC) version of the X25 geared toward enterprise environments. A line of 1.8″ X18-M SSDs also joined the family. More recently, in April, Intel released a firmware update to improve the long-term performance of its X25-M and X18-M models. All along the way, Intel has aggressively cut prices. The 80GB X25-M initially launched at just under $600. Today, it costs close to half that.
SSDs remain an expensive proposition, even for enthusiasts accustomed to paying top dollar for high-performance hardware. Prices continue to plummet, though, spurred most recently by Intel’s introduction of a new generation of X25-M drives based on 34nm fabrication technology. These new units are much cheaper than their forebears, with the latest 80GB flavor selling for just $225 in bulk quantities35% less than the street price of the old X25-M. And it gets better, because Intel says these latest models are even faster than the originals. One arrived at our doorstep this past Wednesday afternoon, and we’ve been testing it since. Let’s see what we’ve learned.
Subtract 16 nanometers
Because solid-state drives are essentially just arrays of flash memory chips, the best way to decrease their cost is to move to finer fabrication technologies that squeeze more gigabytes onto each silicon wafer. The original X25-M’s MLC NAND flash memory chips were fabbed using 50nm process technology, but the new models have flash chips built with 34nm tech. As a result, there’s more storage per chip in the new drives. The X25-M debuted at 80GB with 20 flash chips weighing in at 4GB each. This time around, the X25-M line is led by a 160GB model packing a whopping 16GB per chip.
The second-generation X25-M’s storage controller is also fresh silicon, but there’s been no die shrink here. Instead, Intel has removed any and all traces of halogen from the chip, which should appeal to laptop makers looking to boost their Greenpeace ratings. According to Intel, the controller’s architecture is on the same “technology node” as the original, so it’s still a 10-channel design. Interestingly, however, the company wouldn’t reveal whether the new chip runs at a higher internal clock speed than its forebear.
In fact, the new X25-M’s cache memory has actually gotten slower. The new drives have 32MB of 133MHz Micron DRAM, up 16MB but down 33MHz from the first-gen X25-M. The 32MB cache is still much smaller than what’s strapped to new SSDs based on Indilinx designs, which use 64MB DRAM chips, and the latest Samsung controllers, which come with 128MB of cache.
Intel does say that latencies have been reduced throughout the updated storage controller. As a result, the drive’s random read latency has been reduced from 85 to 65 microseconds, while its random write latency has dropped from 115 to 85 microseconds. That’s about a 25% improvement on both fronts, which suggests an internal clock speed bump.
|Intel PC29AS21AA0||Intel PC29AS21BA0|
Flash fabrication process
|80, 160GB||80, 160GB|
Max sequential reads
Max sequential writes
|85 µs||65 µs|
|115 µs||85 µs|
Max 4KB write IOPS
Max 4KB read IOPS
Active power consumption
|150 mW||150 mW|
Idle power consumption
|60 mW||75 mW|
In addition to reducing latencies, speeding random writes seems to have been another focus of Intel’s performance optimizations. The company claims that the new 80GB X25-M can push up to 6,600 random 4KB write IOPSdouble the original’s theoretical peak and 2,000 shy of the 8,600 IOPS claimed for the new 160GB model. The X25-M’s peak random read rating hasn’t increased, though; it’s still sitting pretty at a staggering 35,000 IOPS.
Although the X25-M’s random write performance has apparently been improved, it doesn’t appear that much has been done to hasten sequential transfers. The gen-two unit’s still saddled with its predecessor’s relatively sluggish 70MB/s sustained write speed rating. That rating wasn’t particularly impressive when the X25-M first launched, and with Indilinx- and Samsung-based drives boasting sustained write speed ratings in the 200MB/s range, it’s a glaring potential liability now. Intel says sustained write speeds haven’t increased because it was concentrating on making a quick transition to 34nm flash and driving down drive prices, instead. That may be the most prudent approach. After all, we’ve found that the original X25-M is no slouch when it comes to real-world write performance. We’ll see how this latest spin fares in a moment.
A few more tweaks at 34nm
More important than changes to the X25-M’s performance ratings are the steps Intel has taken to combat the scourge of long-term SSD performance: the block rewrite penalty. This penalty arises due to the nature of flash memory cells, which are made up of 4KB pages organized into 512KB blocks. It’s possible to write to pages directly, but only if they’re empty. When writing to an occupied page, the drive must rewrite the entire block. Doing so involves reading the much larger block into cache, modifying its contents, and then writing back the whole thingadditional steps that take time, slowing write performance.
But a solid-state drive with loads of free storage capacity should have plenty of vacant pages ripe for direct writing, right? Unfortunately, no. When Windows deletes a file, it doesn’t require that the associated data be wiped from the drive. The flash pages in question are marked as available, but their actual contents go untouched, so they’re technically still occupied. Thus, even in normal day-to-day use, one will eventually exhaust a flash-based drive’s supply of fresh pages and be forced to suffer through the block rewrite penalty for each and every subsequent write.
Intel uses a couple of approaches to combat performance degradation over time. The first is a series of garbage collection algorithms that are constantly working to cleanse the drive of internal fragmentation. Latency reductions inside the new X25-M storage controller apparently benefit these algorithms. Intel says the new design’s larger cache helps, too, as do improvements to the storage controller’s ability to perform concurrent operations and to the efficiency of its NAND array management.
The second stage of Intel’s block-rewrite defense comes in the fourth quarter of this year in the form of a firmware update with TRIM support for Windows 7. TRIM will address the block rewrite penalty by requiring that flash pages associated with deleted files be emptied rather than simply marked as available, keeping drives topped up with unoccupied pages.
Unfortunately, Intel has no plans to offer a TRIM-capable firmware update for its 50nm SSD family. The drives are nearly a year old now, and the TRIM specification wasn’t even close to being finalized back then, so Intel’s position is defensible. Still, it irks me a little that TRIM support perhaps could be added to the old drives with little effort.
Packaged with the 34nm TRIM firmware update will be an Intel SSD Toolbox application with a “TRIM-style” manual cleaning utility for both Windows XP and Vista. Indilinx offers a similar “wiper” utility for its drives, although that tool has compatibility problems with storage controller drivers from AMD, Intel, and Nvidia. Let’s hope Intel’s app is more robust.
The second-gen X25-M’s extensive under-the-hood changes are capped by a new skin. Gone is the black exterior of the old drive; in its place is a bare metal casing that’s gone unpainted in an attempt to further reduce costs. A black metal shim has been added to bring the drive up to a standard 9.5-mm thickness, though.
It’s a good thing that there’s a color contrast between the old and new models, because Intel intends to produce both as its customers make the transition to the new hotness. One may also differentiate between generations based on their model numbers: The old ones end in G1, while the new ones are tagged as G2.
Rumors preceding the X25-M G2’s official launch suggested that a 320GB flavor would join existing 80 and 160GB models. There’s even room on the barren back of the new 160GB drive’s PCB to add another 10 flash chips. Intel has yet to announce a 320GB drive, though, saying only that higher capacities won’t come until next year. We’ll also get a “high performance” 34nm drive in 2010. I suspect that will be an enterprise-class Extreme model based on single-level cell (SLC) flash.
At least we won’t have to wait long for Intel’s 34nm flash chips to migrate to its 1.8″ drives. An updated X18-M is coming by the end of this quarter, and it has the same performance ratings as the second-gen X25-M.
Our testing methods
Today we’ll be testing the new X25-M against not only its predecessor, but also rivals based on the latest SSD designs from Indilinx and Samsung. We’ve found that storage controller designs typically define SSD performance, so the OCZ Summit and Vertex drives we used for testing should be representative of what you can expect from other Samsung- and Indilinx-based drives, respectively.
Because the block-rewrite penalty can severely impact SSD performance, we’ve elected to test the drives in a simulated used state, with all their flash pages occupied. We don’t believe that testing SSDs in a factory-fresh state accurately represents their long-term performance, and we’re far more interested in seeing how drives handle a more typical scenario than chasing higher benchmark scores with SSDs that have been manually freshened with secure-erase tools that clear the contents of all flash pages.
We haven’t used Indilinx’s beta wiper utility to cleanse the Vertex, either. This application still has problems with common WHQL-certified storage drivers from AMD, Nvidia, and Intel, which is a huge problem in our eyes. We’re not inclined to perform additional time-consuming testing on the Vertex with a whole new set of storage controller drivers just because Indilinx hasn’t done sufficient compatibility testing on its end.
Intel Core 2 Duo E6700 2.66GHz
|System bus||1066MHz (266MHz quad-pumped)|
Intel P45 Express
|South bridge||Intel ICH10R|
OCZ PC2-6400 Platinum Edition at 800MHz
|CAS latency (CL)||
|RAS to CAS delay (tRCD)||4|
|RAS precharge (tRP)||4|
|Cycle time (tRAS)||15|
Realtek ALC889A with 2.24 drivers
Gigabyte GeForce 8600 GT 256MB with ForceWare 185.85 drivers
Intel X25-M 80GB with 8820 firmware
OCZ Summit with 18C1 firmware
OCZ Vertex 120GB with 1.3 firmware
Windows Vista Ultimate x64
|OS updates||Service Pack 2|
Our test system was powered by an OCZ GameXStream power supply unit.
We used the following versions of our test applications:
- WorldBench 6 Beta 2
- Intel IOMeter v2006.07.27
- Xbit Labs File Copy Test v0.3
- HD Tach v3.01
- Far Cry 2 v1.3
- Call of Duty 4 v1.4
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.
Quantifying the block-rewrite penalty
Before digging into our benchmark results, it’s worth taking a moment to quantify the magnitude of the block rewrite penalty associated with each drive. As we’ve seen in the past, some SSDs deal with this issue better than others.
First, we used an IOMeter workload consisting exclusively of 4KB random write requests to measure the response time of each drive in its factory-fresh state, with no occupied flash pages. We then subjected each SSD to several runs through HD Tach’s “full” disk benchmark, whose write speed test fills drives with a single, contiguous file. This test neatly occupies all available flash pages, forcing a block rewrite for every subsequent write request.
With our SSDs now in a simulated used state, we ran our IOMeter random writes test once more to gather response time data.
The gen-two X25-M’s response times are quicker then the original’s in both states. That’s good enough for the lead, at least in a used state. However, the Vertex remains the most responsive drive in a factory-fresh state.
Obviously, the Summit’s comparatively slow used-state response times are a cause for concern. I’ve actually been using a Summit in my primary desktop to see if this has a discernable impact on performance with day-to-day tasks. So far, I haven’t noticed the Summit to be any slower than the two-drive Caviar Black 640GB RAID 1 array that it replaced.
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.
Intel’s careful tweaking earns the new X25-M three more points in WorldBench. That puts it in the lead overall, but what about the individual application tests?
Of WorldBench’s multimedia editing and encoding tests, Photoshop is the most demanding of the storage subsystem. The Vertex still has the quickest completion time in that test. However, the G2 X25-M is much faster than the gen-one unit.
WorldBench’s office and multitasking tests aren’t particularly storage-bound. However, the new X25-5M is a few seconds slower than the rest in tests that involve Firefox.
The new X25-M finishes the Nero test quicker than all the other drives. It’s fast in WinZip, too, but is just edged out at the line by the Vertex.
Boot and load times
To test system boot and game level load times, we busted out our trusty stopwatch.
Our test system boots a little slower with the X25-M G2 than it does with the other SSDs. Less than a second and a half separates the fastest drive from the slowest, though.
Game level load times are also pretty close across the board. Again, the new X25-M is a little slower than the old model.
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. We’ve converted those completion times to MB/s to make the results easier to interpret.
Vista’s intelligent caching schemes make obtaining consistent and repeatable performance results rather difficult with FC-Test. To get reliable results, we had to drop back to an older 0.3 revision of the application and create or own custom test patterns. During our initial testing, we noticed that larger test patterns tended to generate more consistent file creation, read, and copy times. That makes sense, because with 4GB of system memory, our test rig has plenty of free RAM available to be filled by Vista’s caching and pre-fetching mojo.
For our tests, we created custom MP3, video, and program files test patterns weighing in at roughly 10GB each. The MP3 test pattern was created from a chunk of my own archive of ultra-high-quality MP3s, while the video test pattern was built from a mix of video files ranging from 360MB to 1.4GB in size. The program files test pattern was derived from, you guessed it, the contents of our test system’s Program Files directory.
Even with these changes, we noticed a little more variability in FC-Test performance than we’d like to see. Normally, we run tests three times and average the results, but for FC-Test, we’ve run each test five times before averaging. We also had to perform some additional test runs to replace obviously erroneous results that cropped up occasionally with each of the drives, usually on the first one or two test runs. The X25-M G2 was a particularly egregious offender on this front, with its performance varying wildly through the first three to four test runs before settling down to consistent, repeatable levels. We’ve asked Intel why that is, but have yet to get a response.
Once G2 the settles down, it boasts the fastest file creation rates with two of our three test patterns. What’s even more impressive is the fact that the new drive is more than twice as fast as the old one with the MP3 and video test patterns. It’s not too far off that mark with the program files test pattern, either.
The gen-two X25-M isn’t always faster than its predecessor when we switch to reads. However, it is consistently quicker than the Indilinx- and Samsung-based drives.
Our copy tests stress both read and write performance, and although the X25-M G2 loses the lead to the Summit in two test patterns, it’s well ahead when dealing with a collection of MP3 files. The new X25-M is about 10MB/s faster than the old drive here.
IOMeter presents a good test case for both seek times and command queuing.
Well, there’s your improved random write performance. With IOMeter workloads that include write operations (everything but the web server workload), the new X25-M offers notably higher transaction rates than the old model. This boost in performance is good enough to put the Intel drive back in the lead, ahead of the OCZ Vertex.
The G2 picks up where the original left off in the read-dominated web server workload, too. Though it’s not that much faster than the older model, the new drive is still miles ahead of the competition.
Along with the G2’s superior transaction rates comes higher CPU utilization. Because some of the drives are doing more work, these results are better put into context by looking at the transaction rate per percent CPU utilization.
Higher numbers are better here, and the X25-M G2 looks very competitive. It’s clearly the most efficient drive with the web server workload.
We tested HD Tach with the benchmark’s full variable zone size setting.
The new X25-M is every bit as fast as the old drive in HD Tach’s sustained read speed test, which makes it quicker than the Summit and the Vertex. However, those drives more than double the X25-M’s transfer rate in the sustained write speed test. At least the G2 drive is a little quicker than the first revision. Amusingly, both exceed Intel’s 70MB/s maximum sustained write speed rating.
The move to a slower but larger DRAM cache costs the X25-M a few MB/s in HD Tach’s burst speed test. Still, the G2 is still quicker than the OCZ drives here.
HD Tach’s random access time test doesn’t tease out meaningful differences between these SSDs. Their near-instantaneous access times are simply too quick.
Our CPU utilization results are within HD Tach’s +/- 2% margin of error in this test.
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. We were able to calculate the power draw from each voltage rail and add them together for the total power draw of the drive. Drives were tested while idling and under an IOMeter load consisting of 256 outstanding I/O requests using the workstation access pattern.
The new X25-M’s power consumption is nearly identical to that of the old model. Keep in mind that at 160GB, the G2 we’re using has twice the storage capacity of the first-gen unit.
That said, the Intel drives are still more power-hungry than those based on Indilinx and Samsung controllers. The Summit’s extremely low idle power consumption is particularly impressive, but we’re still only looking at differences of fractions of a watt.
Intel’s tick-tock approach to processor development delivers new architectures on each tock and then shrinks them to finer process technologies with each tick. This philosophy now appears to be influencing the company’s solid-state drives. The second-generation X25-M retains the same architecture as its predecessor, but takes advantage of new manufacturing technology and some targeted tweaks to deliver real improvements. The move to a more advanced process node has allowed Intel to cut the X25-M prices dramatically, with new 80GB models selling for just $225 in bulk quantities and 160GB drives running only $440.
Lower prices put the X25-M G2 drives around $2.80 per gigabyte, which is still much more expensive than mechanical storage. However, it’s certainly competitive with the cost of other SSDs on the market. The 120GB OCZ Summit, for example, sells for around $350 online. And the 120GB Vertex? It’s just under $400, at least for now. Soon after the 34nm X25-M was announced, OCZ revealed a new price list for its Indilinx-based drives. The 120GB Vertex should hit its freshly discounted suggested retail price of $290 (which translates to $2.42 per gigabyte) before long. Neither OCZ nor Corsair has promised price cuts for Samsung-based drives, though.
Of course, there’s more to the 34nm X25-M than lower prices. The reductions in internal latencies and improvements in random write performance pay big dividends. The gen-two X25-M is faster than its forebear nearly across the board, and by dramatic margins in FC-Test and IOMeter. The fact that the drive excels when faced with both real-world sequential transfers and demanding, highly random workloads is a testament to the strength of Intel’s storage controller design. The Samsung controller does well in FC-Test but turns in an abysmal performance in IOMeter, while the Indilinx controller thrives in IOMeter but is woefully slow in FC-Test.
The original X25-M was already an impressive all-around performer, and since this new revision is even faster, I think we have new storage performance king, at least among MLC-based SSDs. With Windows 7 TRIM support promised in a fourth-quarter firmware update and a manual TRIM tool coming for Vista and XP users, the latest X25-M appears to have all the bases covered. Indeed, the drive’s only real weakness is its comparatively slow sustained write speeds in synthetic benchmarks like HD Tach. That’s a result I’m inclined to overlook given the G2’s otherwise strong showing with real-world write operations, though.
At the end of the SSD round-up we published earlier this month, I concluded that it was probably best to wait until Windows 7’s release to splurge on an SSD. We simply won’t know how TRIM will affect performance until then. However, the new X25-M is so much faster overall than its competitors that it seems unlikely the current generation of Indilinx and Samsung drives will be able to make up the difference, even with TRIM support added to the mix. That’s why we’re tentatively granting the X25-M G2 TR Recommended distinction. If I were looking to plunk a solid-state drive into a netbook, notebook, desktop, or even web server today, I’d go with Intel’s latest X25-M in a heartbeat.