SSD scaling outside the sweet spot

Ever since I started reviewing PC components here at TR, my relationship with hardware has become increasingly promiscuous. The Ikea shelving unit that serves as my test rack plays host to an endless stream of new systems—at least one every week, and when I’m really on my game, several in one day. I’m not incapable of commitment, though. The Twins, our tireless tandem of storage test systems, are firmly entrenched on my rack. Since being updated with Sandy Bridge hardware and an expanded test suite late this summer, this tag team has been running almost non-stop benchmarking the latest solid-state drives.

We broke in the duo with a massive round-up of 120-128GB SSDs. With prices in the $200 range, these drives occupy the sweet spot in the market. They’re not the fastest examples of the breed, though. Manufacturer spec sheets routinely promise better performance for higher-capacity models. Modern SSDs situate their flash arrays behind multi-channel memory controllers, so adding NAND dies can allow drives to take advantage of more controller-level parallelism.

How much of a performance boost can you really get from a higher-capacity SSD? I’m glad you asked. For the past few weeks, The Twins have been chewing through a collection of 240-320GB versions of the same drives from our 120-128GB round-up. The numbers have been crunched, graphs have been drawn, and value has been quantified. So, without further ado, let’s introduce the subjects of today’s look at SSD performance scaling.

One of everything

The SSD market is made up of a dizzying number of different drives, but it’s important to realize the number of unique configurations is considerably smaller. Three factors influence performance: the SSD controller, the flash memory, and the firmware that manages the interactions between those things and the host interface. Only a handful of controllers and flash memory types are actively being used by top-tier drive makers. Firmware is typically supplied by the folks behind the controller, and there isn’t always room to extract additional performance via additional tweaking.

In our sweet-spot SSD round-up, we tested similar SandForce drives from three manufacturers. Despite each one having a slightly different firmware revision, the drives offered essentially equivalent performance across the entirety of our benchmark suite. We did see substantial performance gaps between SandForce configurations using synchronous and asynchronous memory, so that’s where we’ve focused our attention today. Corsair’s Force Series 3 will represent budget drives based on the SandForce SF-2281 controller and equipped with asynchronous NAND, while Corsair’s Force Series GT speaks for more exotic models populated with synchronous flash.

  Size Controller NAND Cache Warranty Price
Corsair Force Series 3 120GB SandForce SF-2281 25-nm Micron NA 3 years $180
Corsair Force Series 3 240GB SandForce SF-2281 25-nm Micron NA 3 years $330
Corsair Force Series GT 120GB SandForce SF-2281 25-nm Intel NA 3 years $204
Corsair Force Series GT 240GB SandForce SF-2281 25-nm Intel NA 3 years $460
Crucial m4 128GB Marvell 88SS9174 25-nm Micron 128MB 3 years $198
Crucial m4 256GB Marvell 88SS9174 25-nm Micron 128MB 3 years $381
Intel 320 Series 120GB Intel PC29AS21BA0 25-nm Intel 64MB 5 years $205
Intel 320 Series 300GB Intel PC29AS21BA0 25-nm Intel 64MB 5 years $545
Intel 510 Series 120GB Marvell 88SS9174 34-nm Intel 128MB 3 years $280
Intel 510 Series 250GB Marvell 88SS9174 34-nm Intel 128MB 3 years $569

Among contemporary SSDs based on Marvell’s 88SS9174 controller, Intel’s 510 Series and Crucial’s m4 are the most recent examples. The m4 is decked out with 25-nm synchronous flash chips similar to the ones found inside the Force GT, while the 510 Series uses older 34-nm NAND. Intel and Crucial have their own firmware, as well, so we’ve included examples of both drives.

The latest controllers from SandForce and Marvell are easily the most popular with SSD makers right now. Intel’s 320 Series provides a unique alternative that pairs an old-school 3Gbps controller of Intel’s own design with 25-nm NAND that rolls out of the company’s fabs. Since this example of vertical integration is part of Intel’s push to get SSDs into mainstream machines, we couldn’t leave it out of the fun.

From a parallelism perspective, the 320 Series’ flash controller has somewhat of an edge over its 6Gbps competition. The Intel chip has 10 memory channels, while the Marvell and SandForce designs must make do with only eight. The individual memory channels on the Intel controller are slower than the ones on its rivals, though.

Drive makers differ on whether they offer 120 or 128GB capacities in the sweet spot. There’s even more variety at higher capacity points. The SandForce crowd doubles up on its 120GB variants with 240GB versions. Crucial multiplies by two, as well, scaling its 128GB m4 up to a 256GB version. Intel’s 510 Series starts at 120GB but makes its next stop at 250GB, providing a little more capacity than the SandForce drives. While the 320 Series also offers a 120GB capacity, it leaves 250GB in the dust on the way to 300GB.

Storage capacity is perhaps better discussed in the context of cost—specifically, the cost per gigabyte. That 300GB 320 Series drive runs $25 less than a 510 Series with 50 fewer gigabytes under the hood. To put things into perspective, let’s look at each SSD’s cost per gigabyte, which we’ve obtained using current Newegg prices and the amount of storage capacity accessible to users in Windows. A Western Digital Caviar Black 1TB mechanical hard drive has been included for reference, and the bars have been color-coded by drive family to make the results easier to follow.

Depending on the drive family, higher-capacity models can be slightly better values than their less capacious counterparts. The Force Series 3, m4, and 510 Series all cost less per gigabyte if you climb the capacity ladder. However, the reverse is true for the Force Series GT and the Intel 320 Series, which are better deals at 120GB than at 240GB and 300GB, respectively.

These cost-per-gigabyte calculations are sensitive to prices that change seemingly every few days at Newegg, so the results certainly aren’t set in stone. However, the overall landscape has looked pretty consistent of late. The Crucial m4 and the asynchronous SandForce drives are among the cheapest 6Gbps drives around, while the 510 Series stubbornly resists discounting. Everything else tends to fall between those extremes.

Our testing methods

I’m just going to come right out and say it: SSD testing is hard. In the mechanical era, storage was nice and predictable. Today’s solid-state drives are rather more complex, especially since their performance depends not just on the test you’re running, but also on the test you ran before that. And don’t forget about the performance implications of the block-rewrite penalty inherent to flash memory—or the TRIM and garbage-collection routines designed to combat it.

To ensure consistent and repeatable results, the SSDs were secure-erased between almost every component of our test suite. This returns the drives to their factory fresh state, erasing any remnants of previous workloads.

For some benchmarks, we’ve deliberately tested drives in a used state to illustrate their long-term performance potential. Other tests create their own used states, usually by writing across the full extent of the drive before launching into a workload. In all cases, the SSDs were tested in the same states as their peers, ensuring an even playing field for all. I’ve even gone so far as to avoid running certain benchmarks overnight to ensure that some SSDs don’t spend more time than others idling between tests.

Steps have also been taken to ensure that Sandy Bridge’s power-saving features don’t taint any of our results. All of the CPU’s low-power states have been disabled, effectively pegging the 2500K at 3.3GHz. Transitioning in and out of different power states can affect the performance of storage benchmarks, especially when dealing with short burst transfers.

We’re using fresh Rapid Storage Technology drivers from Intel and the latest firmware revisions on virtually all of the SSDs. The 120GB Force SSDs were tested with Corsair’s 1.3 firmware, while the 240GB drives ran a slightly newer 1.3.2 revision. Corsair tells us the changes incorporated in 1.3.2 don’t impact performance, so the discrepancy shouldn’t tarnish our results. For the Intel 510 Series, our 250GB drive used PWG2 firmware, while the 120GB model ran PPG4. Intel doesn’t offer firmware updates for the 510 Series, so you’re stuck with what you’ve got. For what it’s worth, the company says changes in the SSD’s firmware haven’t been substantial enough to warrant a public release.

We run all our tests at least three times and report the median of the results. We’ve found IOMeter performance can fall off after the first couple of runs, so we use five runs in total and throw out the first two. We used the following system configuration for testing:

Processor Intel Core i7-2500K 3.3GHz
Motherboard Asus P8P67 Deluxe
Bios revision 1850
Platform hub Intel P67 Express
Platform drivers INF update 9.2.0.1030

RST 10.6.0.1022

Memory size 8GB (2 DIMMs)
Memory type Corsair Vengeance DDR3 SDRAM at 1333MHz
Memory timings 9-9-9-24-1T
Audio Realtek ALC892 with 2.62 drivers
Graphics Asus EAH6670/DIS/1GD5 1GB with Catalyst 11.7 drivers
Hard drives Corsair Force Series 3 120GB with 1.3 firmware

Corsair Force 3 Series 240GB with 1.3.2 firmware

Corsair Force Series GT 120GB with 1.3 firmware

Corsair Force Series GT 240GB with 1.3.2 firmware

Crucial m4 128GB with 0009 firmware

Corsair m4 256GB with 0009 firmware

Intel 320 Series 120GB with 4PC10362 firmware

Intel 320 Series 300GB with 4PC10362 firmware

Intel 510 Series 120GB with PPG4 firmware

Intel 510 Series 250GB with PWG2 firmware

WD Caviar Black 1TB with 05.01D05 firmware

Power supply Corsair Professional Series Gold AX650W
OS Windows 7 Ultimate x64

Thanks to Asus for providing the systems’ motherboards and graphics cards, Intel for the CPUs, Corsair for the memory and PSUs, Thermaltake for the CPU coolers, and Western Digital for the Caviar Black 1TB system drives. Also, thanks to all the SSD makers for providing their drives.

We used the following versions of our test applications:

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

Most of 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.

HD Tune — Transfer rates

HD Tune lets us look at transfer rates in a couple of different ways. We use the benchmark’s “full test” setting, which tracks performance across the entire drive and lets us create the fancy line graphs you see below. This test was run at its default 64KB block size.

Not much action here. The higher-capacity drives are a smidgen faster than their 120-128GB counterparts, but the differences only amount to a few MB/s. The one exception is the Intel 510 Series, whose 250GB variant manages an average read speed 15MB/s higher than the 120GB model.

Switching to writes lets the high-capacity drives strut their stuff, and most of them see substantial performance increases. Once again, the 510 Series enjoys the biggest boost from a step up in capacity; its average read speed climbs by nearly 90MB/s when moving to the 250GB drive. The 320 Series and Crucial m4 also enjoy healthy leaps of 69 and 50MB/s, respectively.

Although the Force Series GT still finds itself atop the standings, the 240GB synchronous SandForce drive only enjoys a 29MB/s increase in average write speed over the 120GB unit. The gap in performance is even smaller with the asynchronous Force Series 3, whose high-capacity variant adds just 7MB/s. Interestingly, the differences in minimum write speeds between the two SandForce capacities are more substantial: 34MB/s for the Force Series 3 and 50MB/s for the GT. As you can see in the line graph, the write speeds of the SandForce SSDs oscillate between high and low extremes, while the rest of the field maintains more consistent write rates.

HD Tune’s burst speed tests are meant to isolate a drive’s cache memory, so they shouldn’t be affected by total capacity.

Crucial apparently didn’t get the memo, because the m4’s 128GB model holds a 40MB/s advantage over the 256GB one with writes. Intel’s 510 Series 250GB also has an edge with burst transfers, but it only amounts to 8MB/s for both reads and writes. Otherwise, SSD capacity has little impact on burst performance.

HD Tune — Random access times

In addition to letting us test transfer rates, HD Tune can measure random access times. We’ve tested with four transfer sizes and presented all the results in a couple of line graphs. We’ve also busted out the 4KB and 1MB transfers sizes into bar graphs that should be easier to read.

At least with reads, the 120-128GB SSDs have no problem keeping up with their higher-capacity counterparts. Will that hold true when we switch to writes?

That depends on the controller and the transfer size. The results are unchanged with random 4KB writes and with the SandForce drives overall. However, the 1MB transfer size shows a clear advantage for the high-capacity versions of the Crucial m4 and both Intel SSDs.

TR FileBench — Real-world copy speeds

Our resident developer, Bruno “morphine” Ferreira, has been hard at work on a new file copy benchmark for our storage reviews. FileBench is the result of his efforts. This shining example of scripting awesomeness runs through a series of file copy operations using Windows 7’s xcopy command. Using xcopy produces nearly identical copy speeds to dragging and dropping files using the Windows GUI, so our results should be representative of typical real-world performance.

To reduce the number of external variables, FileBench runs entirely on the drive that’s being tested. Files are copied from source folders to temporary targets that aren’t deleted until all testing is complete. Copy speeds were tested first with the SSDs fresh from a secure erase and a second time in a “tortured” used state after 30 minutes of IOMeter thrashing through a workstation access pattern loaded with 32 concurrent I/O requests.

To gauge performance with different kinds of files, we tested with five sets. The movie set includes six video files of the sort one might download off BitTorrent. Total payload: 4.1GB. 101 uncompressed images from my Canon Rebel T2i make up the RAW file set, totaling 2.32GB. Our MP3 file set uses a chunk of my music archive, which is made up of high-bitrate MP3s and associated album art. This one has 549 files that add up to 3.47GB. The Mozilla file set includes the huge selection of files necessary to compile Firefox. All told, there are 22,696 files spread across only 923MB. Finally, we have the TR file set, which contains several years worth of the images, HTML files, and spreadsheets behind my reviews. This set has the largest number of files at 26,767, but it’s heftier than the Mozilla set with 1.7GB worth of data.

For nearly every drive family, the higher-capacity SSD is faster across all five file sets. Only with Crucial m4 does the lower-capacity model have quicker copy speeds, and then only with the small files in the TR and mozilla file sets. For the most part, the deltas between our 120-128GB and 240-300GB drives are most pronounced with the larger files that make up the movie, RAW, and MP3 file sets. Those gaps tend to shrink when with the TR and mozilla sets, which produce slower copy speeds overall.

The SandForce posse is perhaps the best example of the overall trend. For both Force SSDs, the performance differences between capacities are more pronounced with larger file sizes than they are with smaller ones. The synchronous GT sees a much bigger jump in performance moving from 120GB to 240GB than its asynchronous sibling. Despite being hampered by a relaxed TRIM mechanism that results in slower used-state copy speeds, the SandForce drives are among the best performers overall.

As with the Force SSDs, the higher-capacity Intel 510 Series enjoys much bigger increase in copy speeds with larger files than it does with smaller ones. The 510 Series is pretty quick overall, and its performance scales up much better than the Crucial m4 even though the two drives share the same controller. The m4 is the lone exception to the overall trend—the differences in copy speeds between its 128 and 256GB capacities are actually smaller with the larger file sets. It’s interesting to note that the m4 is much more competitive in those larger file sets, too.

The Intel 320 Series is the slowest SSD of the bunch, but the 300GB version is a fair bit faster than the 120GB, especially with bigger files. With a couple of file sets, the 300GB drive is even quick enough to sneak ahead of the 240GB Force Series 3.

TR DriveBench 1.0 — Disk-intensive multitasking

TR DriveBench allows us to record the individual IO requests associated with a Windows session and then play those results back on different drives. We’ve used this app to create a set of multitasking workloads that combine common desktop tasks with disk-intensive background operations like compiling code, copying files, downloading via BitTorrent, transcoding video, and scanning for viruses. You can read more about these workloads and desktop tasks on this page of our SSD value round-up.

The traces that make up this first batch of DriveBench workloads are an imperfect measure of real-world performance because they only account for small snippets of disk activity. A much larger trace on the following page addresses that deficiency, but we’re going to keep around the old ones to give you a comparative point of reference to our reviews of older drives.

Below, you’ll find an overall average followed by scores for each of our individual workloads. The overall score is an average of the mean performance score with each multitasking workload.

DriveBench 1.0 is run right after five rounds of our usual IOMeter access patterns, so all the drives start in a thoroughly used state. This test will only run on an unpartitioned drive, so we delete the IOMeter test file (which spans the entire capacity of each drive) and the accompanying partition before launching DriveBench.

The 240GB Force SSDs push quite a few more IOs than their smaller siblings. The gaps aren’t as large with the Intel drives, but both get a decent bump when scaling up to higher capacities. Alas, the same isn’t true for the Crucial m4, whose overall DriveBench score is only slightly higher for the 256GB model.

Breaking down DriveBench’s individual tests reveals consistently large performance deltas between the different sized SandForce drives. The gaps between the Intel SSDs are more dependent on the task. The 510 and 320 Series both scale up with additional capacity in the copy workload, but the higher-capacity models have smaller advantages in the other workloads.

Despite pushing more IOs than much of the competition, the m4 isn’t much faster at 256GB than it is at 128GB. That’s not necessarily a bad thing. Taken another way, the 128GB model isn’t much slower than the 256GB one.

TR DriveBench 2.0 — More disk-intensive multitasking

As much as we like DriveBench 1.0’s individual workloads, the traces cover only slices of disk activity. Because we fire the recorded I/Os at the disks as fast as possible, the drives also have no downtime during which to engage background garbage collection or other optimization algorithms. DriveBench 2.0 addresses both of those issues with a much larger trace and a slightly different testing methodology.

For our new trace, I recorded about two weeks of disk activity on a test system pressed into service as my primary desktop. The system was left on at all times, and it was used mostly for web surfing, email, photo editing, gaming, and working with the HTML, Excel, and image files that become TR content. To make things more interesting, I fired up disk-intensive multitasking workloads alongside those more mundane desktop tasks. The multitasking workloads were similar to what’s included in DriveBench 1.0: compiling code, copying files, downloading torrents, transcoding video, and scanning for viruses.

Although my bursts of disk-intensive multitasking were contrived in nature, the goal was to come up with a demanding test that would probe drives for weakness in a sea of everyday I/O. The system was limited to a single partition, which housed not only the OS and applications but also all of the associated data and downloaded, ahem, Linux ISOs. We plan to add at least one more DriveBench workload that more strictly models the life of an OS and applications drive, but that’ll have to wait for a future article.

As it stands, our multitasking-infused trace is loaded with more than 25 million read operations totaling over 1.1TB of data. The workload has plenty of writes, too: 14 million that add up to nearly 525GB. That’s a busy couple of weeks.

DriveBench 1.0 all but eliminates disk idling, but this second revision gives the SSDs plenty of idle time for background processing. The test begins with drives fresh from a secure erase, but it writes across their full capacity to ensure that all flash pages are filled before performance is measured. Things are a little different on that front, too. Instead of looking at a raw IOps rate, we’re going to explore service times—the amount of time that it takes drives to complete an I/O request. We’ll start with an overall mean service time before slicing and dicing the results.

The 510 Series offers little improvement in performance above its sweet-spot capacity of 120GB. That shuffles the standings when we look at the higher-capacity drives, of which the 510 Series is the slowest overall. Otherwise, all of the higher-capacity SSDs offer decent reductions in service times when compared to their smaller siblings. The Crucial m4 cuts its mean service time by over one third, which is the biggest drop of the bunch.

Let’s split that overall score with a breakdown of service times for reads and writes.

As we saw in HD Tune, the divide between high- and low-capacity models is pretty narrow with reads. The situation changes with writes, which execute substantially faster on the 240-300GB drives—with the exception of the 510 Series, of course. Intel’s take on the Marvell controller offers nearly identical performance at our two capacity points. I’d be more encouraged by the 120GB unit keeping up with its big brother if either offered a lower mean service time than our lone mechanical hard drive.

There are millions of I/O requests in this trace, so we can’t easily graph service times to look at the variance. However, our analysis tools do report the standard deviation, which can give us a sense of how much service times vary from the mean.

The 240-300GB SSDs have lower standard deviations with both reads and writes, which means they offer more consistent performance than the lower-capacity models. Once again, the differences are more pronounced with writes than with reads. The Intel 510 Series is the one exception here, too.

If I haven’t already scared you off with too many graphs and statistics, this next pair will do it. We’re going to close out our DriveBench analysis with a look at the distribution of service times. I’ve split the tally between I/O requests that complete in 0-1 milliseconds, 1-100 ms, and those that take longer than 100 ms to complete.

When we look at the distribution of service times across our different capacity points, it’s clear writes benefit from the larger capacities more than reads.

IOMeter

Our IOMeter workloads are made up of randomized access patterns, making them perfect candidates to exploit the wicked-fast access times of solid-state storage. This app bombards drives with an escalating number of concurrent IO requests and should do a good job of simulating the demanding environments common in enterprise applications. We’ve previously tested with the “pseudo random” access pattern, but that re-uses a buffer that is only filled with random data once, which doesn’t strike us as very random at all. For this round of testing, we’ve cut out the recycling with IOMeter’s fully random setting.

This decision mostly impacts the write-compression technology inside SandForce controllers, which will struggle to work its magic on truly random sequences of 1s and 0s. While we’ve observed little drop in performance moving from pseudo to fully random workloads on SandForce drives with server-style overprovisioning percentages, the same can’t be said for current consumer-grade drives that set aside much less of their flash capacity as “spare area” for the controller.

Over the last few years, we’ve watched new storage controller drivers effectively cap IOMeter performance scaling beyond 32 outstanding I/O requests. The Serial ATA spec’s Native Command Queue is 32 slots deep, and more than one drive maker has told us that this queue is rarely full. As a result, we’re only testing up to 32 concurrent I/O requests.

Our IOMeter results are mixed to say the least. The SandForce-equipped Force SSDs offer higher transaction rates at higher capacity points across all four workloads. They perform much better versus their rivals when executing the file-server, workstation, and database access patterns, which are made up of a mix of reads and writes. In those three workloads, the 240GB Force Series GT is the fastest overall. Its asynchronous counterpart is among a clump of drives vying for second place.

That cluster includes the Crucial m4, which trumps all of the other SSDs with the web-server access pattern, a workload made up entirely of read operations. Surprisingly, the 128GB m4 is quicker than the 256GB drive in that test. With the other access patterns, the higher-capacity model crunches moderately more IOs.

The 120GB Intel 510 Series is also in the running for second place overall. However, its 250GB counterpart produces lower transaction rates in each access pattern. With all but the web-server workload, the 250GB 510 Series flirts with the back of the pack. The Intel 320 Series exhibits similar behavior. While its 120 and 300GB flavors are closely matched with the web-server access pattern, the 300GB drive is substantially slower with the others. I suppose the only saving grace for the 320 Series is the fact that it didn’t have as far to fall as the 510 Series.

Boot duration

We’re limited in how we can measure storage performance with actual applications, but we have come up with a handful of load-time tests that do just that. This is the only batch of performance tests that presses the competitors into service as system drives housing the operating system. It’s only fitting, then, that we start by timing how long it takes to load the OS. Here, we’re relying on Windows 7’s own performance-monitoring capabilities to clock the boot duration, which is the time between BIOS initialization and when the system has loaded all processes and idled for 10 seconds. We’re reporting the boot duration minus those superfluous seconds.

The 120GB SSDs all load Windows 7 faster than their high-capacity kin, although the differences only amount to fractions of a second. There’s essentially no difference in boot times between the 128 and 256GB flavors of the Crucial m4.

Level load times

Upgrading to a fancy solid-state drive will likely have little impact on in-game frame rates, but you will be able to load levels faster.

Most of the results are too close to call—we have to time these tests with a stopwatch, so I wouldn’t worry about minute differences in the graphs above. That said, the Intel SSDs consistently loaded our game levels quicker at 120GB than they did at 250 and 300GB.

Compiling

We’ve long thought that code compiling might be able to tease out meaningful performance differences between different storage solutions, so we’ve taken one more shot at the problem with a little help from FileBench creator Bruno “morphine” Ferreira. This test starts with version 2010.05 of the Qt application framework source, which is compiled with multiple threads using the MinGW port of GCC 4.4.0. Mad props to morphine for packaging this test so nicely.

This multi-threaded compiling job can’t tease out any meaningful performance differences between our SSD capacity points. That’s to be expected given that the test doesn’t even benefit from transitioning to solid-state storage from a mechanical hard drive.

Power consumption

We tested power consumption under load with IOMeter’s workstation access pattern chewing through 32 concurrent I/O requests. Idle power consumption was probed one minute after processing Windows 7’s idle tasks on an empty desktop.

Most of the SSDs have comparable power consumption between the capacities we’ve gathered. There are a few exceptions, however. The 240GB Force Series 3 consumes a little less power than the 120GB drive at idle, even though the situation is reversed under our IOMeter load. The higher-capacity Force GT and Intel 510 Series SSDs also draw more wattage than the 120GB versions, but only under load.

The value perspective

Still with me? Congratulations, you’ve reached our famous value analysis, which adds capacity and pricing to the performance data we’ve explored over the preceding pages. We used Newegg prices to even the playing field for all the drives, and we didn’t take mail-in rebates into account when performing our calculations.

Our value calculations require a single performance score, which makes things a little complicated. We’ve come up with an overall index that normalizes SSD performance against a common baseline provided by the Caviar Black. This index uses a subset of our performance data, including HD Tune’s random 4K response times and average transfer rates, our used-state FileBench results, scores from all five DriveBench 1.0 workloads, mean DriveBench 2.0 service times plus the percentage above 100 ms, IOMeter transfer rates for each access pattern with eight outstanding I/O requests, the Windows 7 boot duration, and our load times in Portal 2 and Duke Nukem Forever.

Time constraints prevented us from using a slower baseline drive than our Caviar Black, which actually scored better than a few of the SSDs in a couple of DriveBench metrics. To prevent those scores from jacking with the overall results, we’ve fudged the numbers slightly to match our mechanical baseline. Calculating overall performance scores is an imperfect science, and I may have to dust off our old 4,200-RPM notebook drive to set a new baseline for future reviews.

We’ve been using a harmonic mean to generate our overall score for storage performance because it does a good job of handling normalized results that can vary by several orders of magnitude from one test to the next. After much reading on the subject and calculating numerous performance scores in previous storage reviews, we’re convinced this is the best approach for our particular mix of tests.

In terms of raw performance, the SandForce-based Force SSDs scale up to higher capacity points better than the competition. The asynchronous and synchronous configs both enjoy big increases in overall performance when moving from 120 to 240GB. The other big winner is the Intel 320 Series, whose 300GB variant vaults the drive from last place to a smidgen ahead of the fastest Crucial m4. Meanwhile, the 510 Series offers the least incentive to add capacity.

So, what happens if we mash this overall performance score with cost and capacity? Magic! Or, rather, performance per dollar per gigabyte, which is divides each SSD’s overall score by its cost per gigabyte. We’ll express this value metric as a single score in a line graph before exploring the relationship between performance and cost-per-gigabyte in a scatter plot.

Taking the cost per gigabyte into account also factors in the range of capacities we’re dealing with at each step on the ladder. The 240GB Force Series 3 occupies an enviable spot on our scatter plot, offering the lowest price per gigabyte and better performance than all but its GT stable mates. One could argue that the step up in performance to the GT is worth the extra cash.

If you’re not keen on any of the SandForce SSDs, the Crucial m4’s low cost per gigabyte at 256GB is rather intriguing. The high-capacity Intel drives are much pricier.

Although this analysis is helpful when evaluating SSDs on their own, what happens when we consider the cost of drives in the context of a complete system? To find out, we’ve divided our overall performance score by the sum of our test system’s components (which total around $800 at Newegg before adding the SSDs).

Gobs of solid-state storage capacity ain’t cheap, and considering the cost of a complete system shuffles our scatter plot. The 120-128GB drives are mostly lined up in the middle of the plot, while the others drift off into pricier territory to the right. Looking at the spread, the Force Series GT 120GB stands out as a pretty good deal. The Force GT 240GB is still the fastest drive overall, and its price premium looks justified as a result.

Conclusions

Although we’ve only scratched the surface of the SSD scaling picture, we’ve learned a few important things about how contemporary controllers and drive configurations behave when transitioning from the 120-128GB sweet spot to higher capacity points. For the most part, performance goes up. The improvements are much more substantial with writes than with reads, but each SSD family has its own set of quirks.

For the Intel 320 Series, the big quirk is comparatively sluggish overall performance. The 3Gbps Serial ATA interface doesn’t do the it any favors, but the drive does gain more performance than a couple of its rivals when moving up the capacity ladder. To be fair, the 320 Series jumped from 120 to 300GB, while the rest of the pack stayed anchored in the 240-256GB range.

The 320 Series 300GB actually scores higher in our overall performance index than the Crucial m4 256GB, whose performance gains at its higher capacity are almost as impressive. The 256GB drive’s improved write performance addresses a key weakness of the 128GB model, but only in a few instances is the m4 quick enough to keep up with the leaders. Comparatively slower write performance does little to blunt the m4’s appeal as an OS and applications drive. In fact, the 256GB model’s sub-$400 asking price offers nearly the lowest cost per gigabyte of the bunch.

Despite using the same controller as the m4, the Intel 510 Series doesn’t score nearly as well overall. Performance generally steps up nicely between the 120 and 250GB versions of the drive, but the 250GB model also trails the 120GB on a few occasions. Either way, the 510 Series is almost completely ruined by street prices that seem unreasonably high given the cost of the competition.

The fact is that SandForce-based SSDs are a lot cheaper than the 510 Series—and a lot faster. Their performance also scales up much better when moving from 120 to 240GB. Configurations with synchronous memory like the Corsair Force Series GT are the best options if you want pack-leading performance, but the asynchronous stuff exemplified by the Force Series 3 is pretty compelling, too. We can’t mention SandForce SSDs without a word of caution about the BSOD bug, of course, but the latest firmware releases purportedly resolve at least some of the issues users have encountered.

I should probably wrap this up with a confession of sorts. Our testing of 240-300GB SSDs was motivated as much by an interest in performance scaling as it was by the impending arrival of a handful of new SSDs. Drives typically debut at higher capacity points, and given the time it takes one of The Twins to run through our suite, we decided to get a head start on gathering comparative data. With that done, our sights can now turn to performance scaling in lower-capacity SSDs at budget price points. Stay tuned.

Comments closed
    • kamikaziechameleon
    • 8 years ago

    I loved your RAID test a while back and would really enjoy you revisiting SSD in different raid arrays when they get to $1 per gigabyte.

    I can’t wait for the next gen of SSD’s I can almost afford a 250 Gb SSD!!!

    • indeego
    • 8 years ago

    Last page mistake:
    [quote<]The 3Gbps Serial ATA interface doesn't do the it any favors,[/quote<]

    • anotherengineer
    • 8 years ago

    Did I miss the section where you list the firmware revisions the drive were tested with?

    And again no tests with an AMD 950/850 SB chipset for comparison.

    Other than decent review, I am still saving for the 256GB M4…..be awhile yet

      • Dissonance
      • 8 years ago

      As with previous storage reviews, all the firmware revisions are detailed on the test methods page.

      We prefer to use the fastest platform possible storage tests, and Intel has the fastest SATA controller by a fair margin. Coming up with comparative data on an AMD platform would have literally taken weeks. We’ll continue to revisit AMD SATA performance as new platforms come out, but given the number of days it takes to run just a single SSD through our test suite, it’s likely that storage testing will remain contined to a single platform.

    • green
    • 8 years ago

    in regards to the m4 256/512gb variants
    i’m starting to think either the limit of the marvell controller has been hit and needs a mjor update rather than minor bump/shrink, or that ‘too much’ processing is going on on the controller
    if it’s too much processing it would explain why in some cases the 128gb performs better than the higher capacity versions
    where the amount of data passing through the 128gb is just hitting the limits, whereas the 256/512gb overwhelms it
    (sort of like having a 800bhp engine under the hood, then using bald tires to grip to the road)

    either way i’m very happy i went with the 128gb m4
    despite it not having top performance across the board, the fact is crucial has had the least amount of problems for this generation
    and with it crucial again released a firmware update that significantly boosted performance
    it may not got the best ratings from places like anandtech (they still haven’t touched the 128gb)
    but overall the initial c300 series were pretty darn solid. the m4 is just as good.
    can’t wait to see how crucial go with their next gen ssd’s

    • FranzVonPapen
    • 8 years ago

    Funny how you never include the [b<]Overall performance per $/GB[/b<] for the Caviar Black 1TB used as reference elsewhere in the comparison.

      • Buzzard44
      • 8 years ago

      I initially thought the same thing, then realized the Caviar Black would have such good performance on a $/GB basis that it would drastically skew the graph, making the difference between SSD price/performance ratios look insignificant.

        • FranzVonPapen
        • 8 years ago

        That’s my point, really. Rotating media remains an outstanding value.

          • Waco
          • 8 years ago

          Just like a $.10 pencil has a better performance per $ than a nice mechanical pencil…but which would you rather use every day?

            • paulWTAMU
            • 8 years ago

            I prefer regular ones personally but that’s jsut me…

            • Bauxite
            • 8 years ago

            Go with a car/bus analogy, it will work out better

            • indeego
            • 8 years ago

            I think I’m doing it wrong, but mechanical pencils break constantly when I have used them.

          • sjl
          • 8 years ago

          Outstanding value compared to what, exactly?

          Yes, I know you meant on a dollars per GB basis. On that particular metric, you’re absolutely right: cheap, commodity hard drives can’t be beaten. But there are other metrics as well that are important.

          For example: how much electricity do you need to keep the data constantly available? If you can shuffle a few hundred TB off to tape, because you only need a small subset instantly available at any given point in time, the tapes can sit on a shelf, and won’t need any electricity. (And yes, I’ve worked for companies that do exactly that.)

          Or perhaps IOs per second is what’s important to you. Traditionally, you’d RAID up a bunch of fast-spinning disks, and use maybe 10% of their capacity so you could get the IOPS up. SSDs are king in this space – even though their capacity is low in comparison to hard drives, they cannot be beaten for random IO, and if that’s what’s important, they’ll whomp all over spinning rusty platters..

          Maybe you’d like to be able to read data sequentially at high speeds (250 MB/s or higher)? RAID up some hard drives, or put it on tape. (LTO4: 120 MB/s native; 240 MB/s compressed. LTO5: 140 MB/s native, 280 MB/s compressed. T10000C: 240 MB/s native, 480 MB/s compressed. Per drive.)

          It all comes down to what you need out of your storage, and speccing it out accordingly.

          (Why yes, I do work with storage for a living. However did you guess?)

    • jthh
    • 8 years ago

    No OWC SSDs?

      • Dissonance
      • 8 years ago

      We haven’t noticed any real difference in performance between different SandForce implementations with this generation or the previous one. Since the focus of this article is on scaling, we decided to focus only on configurations that might have different scaling stories to tell. Each SSD takes several days to test, so this article would have come out much later had we tried to include multiple SandForce partners. That’s why you don’t see results for the Agility 3, Vertex 3, or HyperX SSDs that were in our 120-128GB roundup.

    • duke_sandman
    • 8 years ago

    Awesome work guys — the technical clarity and depth surpass even niche sites that focus on storage. Very very nice.

    I was surprised that there was not a large difference between the compile times for the MinGW/QT application for the SSD vs. mechanical HD. My main development application went from a 5min30sec multi-threaded compile with a Caviar Black to 2min flat with a C300 SSD. This is a math/science application that leans heavily on the Boost uBlas linear algebra library, btw, and #includes a couple thousand header files….

    • SuperMonkey
    • 8 years ago

    I’m looking for a smaller SSD for Intel’s SRT.
    Looking forward to the upcoming reviews.

      • stdRaichu
      • 8 years ago

      Didn’t Intel already release a 20GB SLC SSD expressly for their SSD caching thingy? Shame it’s a software-only solution…

        • SuperMonkey
        • 8 years ago

        Yes but 20 GB is a bit small.

    • crsh1976
    • 8 years ago

    I’m still on the fence with regards to SSDs; prices are more tempting nowadays, but the on-going Sandforce controller BSoD issue keeps me at bay. There are more stable models available, thankfully.

    • stdRaichu
    • 8 years ago

    FWIW I’ve got a 512GB Crucial M4 and in sequential tests I can get about 515MB/s reads and 270MB/s writes using the 0009 firmware.

    Just bought it’s 64GB kid brother to use in an HTPC. Best drives of the current generation in my inevitably biased opinion 😀

    Edit: got my figures mixed up, was 440MB/s seq reads on 0002 and 515MB/s on 0009.

    • Bensam123
    • 8 years ago

    I’m surprised all the OCZ drives got cut out of the mix from the first review. OCZ also has a couple new SSDs out that would be cool to see the results from in addition to a couple oddballs currently on the market like Adata, Gskil, Patriot, and Samsung. Samsung being the most obvious as they make standard HDs as well.

    It’s really quite surprising that none of the major HD manufacturers have hopped on board the SSD train yet besides with that bastardazition Seagate made awhile back.

    • Firestarter
    • 8 years ago

    So… what happens if you throw RAID 0 in the mix? Just about every motherboard out there has it, and supposedly Intels variant isn’t half bad. I wouldn’t recommend it for a buddy just yet, but I would be seriously considering striping 2 128GB SSDs versus a single 256 SSD for my personal beastly computer.

    I realize this is whole ‘nother can of worms that you guys might be avoiding because of its complexity. But I can’t help to wonder and I’m not sure I’d trust anyone else to tell me which is faster 😀

      • Bensam123
      • 8 years ago

      Second. Some of the lower price points are becoming viable in terms of match up in performance vs those at higher price points. They don’t have a middle sweet spot with linear performance like standard HDs or even graphics cards. I guess they haven’t figured out how to completely whore out the price range yet as the market is relatively new and changing a lot.

      • mattthemuppet
      • 8 years ago

      I think the main issue with an SSD RAID is Trim. Old info, but I don’t think Trim/ Garbage Collection works with RAID, so the performance will degrade with use due to the block rewrite penalty. Things might have changed recently though 🙂

        • Yeats
        • 8 years ago

        I believe garbage collection works on RAID0 – my brother’s Crucial SSDs in RAID are still performing “like new” after several months. Could depend on controller, though.

        Update: exchanged emails with my brother, he ran Crystal DiskMark 3.0.1 and said that the measurements were as good or better than when we set up the system initially.

          • Waco
          • 8 years ago

          I run my old-ass Indilinx drives in RAID 0 as well. The numbers haven’t budged a bit since I installed my OS on them over 9 months ago with pretty heavy usage in the meantime. Garbage collection, at least on the older Indilinx drives (with the newest firmware) works well enough to keep any noticeable change under control.

          The drives specifically are a G.Skill Falcon 128 GB and a Corsair X128. Both are flashed with the newest Falcon firmware.

    • Sunburn74
    • 8 years ago

    The fact that regardless of speed of the drive (where some drives are literally 2x as fast as the others) it takes 8 seconds to boot a level implies there is a uncovered and undetermined bottleneck when it comes to level loads once you get past a certain read speed with your drive.

      • Firestarter
      • 8 years ago

      Would decompressing a gigabyte worth of JPGs do it for you?

      • Bauxite
      • 8 years ago

      perhaps slow I/O masks the cpu running code, they are using the same system as far as the rest of the parts.

      • Kaleid
      • 8 years ago

      CPU speed likely. Compiling benchmark supports this.
      It would be interesting to see the 2500K 3.3GHz at vastly higher clocks to see if it results into a larger performance difference.

    • Buzzard44
    • 8 years ago

    What? No LTO-5 in the review?

    Seriously though, the m4 is my sweetheart. Can’t imagine a better drive at the moment.

    • Palek
    • 8 years ago

    From page 5:
    [quote<]For each and every drive family, the higher-capacity SSD is faster across all five file sets.[/quote<] This seems to contradict the bottom two graphs - TR and Mozilla - where the 128GB m4 is faster than the 256GB m4. Accidental data swap?

      • Dissonance
      • 8 years ago

      Just blurred vision from staring at too many graphs. Fixed.

        • Palek
        • 8 years ago

        I must say, tabulating all that data into readable graphs is a part of your job I do not envy at all. 🙂

        Great job with the article, by the way! The m4 keeps looking like a great alternative.

          • stdRaichu
          • 8 years ago

          I’ve just written the performance part of a business case for a new 3par SAN at work. This involved condensing about 20MB of iometer CSV’s into three not-for-techies graphs and is a process I would not wish on my worst enemy.

    • phez
    • 8 years ago

    Force GT 120 or m4 128?

    Tough call on performance. Though, the numerous issues I read regarding the Corsairs isn’t helping their cause.

    But really, unless we’re looking at $150 drives, I’m not sure what else you could review to remove the m4 from it’s throne 😉

    Oh , and, thanks of course !!

    • Buzzard44
    • 8 years ago

    Awesome review guys, keep it up!

    Also:
    Looks like newegg dropped their prices quite a bit on several of the drives since you filled in that table in the article. Might want to update that if you have the time.
    Over $50 on the Corsair Force 3! Not sure how that affects your conclusions.

    Edit: Also there’s no link on the comments page back to the article.

      • Dissonance
      • 8 years ago

      Doh, I swear some of these prices are changing daily. The Force 3’s lower price had actually been updated in the graphs already, but I missed it and a few others in the chart. Everything should be up to date now. For now.

        • tay
        • 8 years ago

        This is a fantastic review, thanks.

        • SomeOtherGeek
        • 8 years ago

        First, great review!

        Second, I know this is probably asking too much, but is it possible to make the price/GB graph real-time, so that if the price changes we could plug in the price and see the new standings? You have all the data there, just a input for the price and see how the result changes? I love what you did and I could play with those numbers all day! Anyway, keep up the great work.

        • Bauxite
        • 8 years ago

        The intels are still staying above MSRP though, kindof annoying

    • NeelyCam
    • 8 years ago

    What’s the difference between Intel NAND and Micron NAND..? I mean, aren’t they produced by the same one company?

      • Bauxite
      • 8 years ago

      Yes, IMF Tech. (lit. Intel Micron Flash)

      Are you forgetting controller, firmware, etc…

      *gasp* intel isn’t perfect and amazing and magical? /blasphemy

Pin It on Pinterest

Share This