G.Skill Titan 128GB Solid State Drive Review

Overall Score

Solid State Disks (SSDs) are a relatively new product segment for the enthusiast computer user, promising fantastic data transfer speeds, lower power usage, and greater durability than traditional Hard Disk Drives (HDDs).  The technology driving SSDs, however, is now almost 30 years old and is based on non-volatile NAND flash memory chips.  NAND memory differs from traditional Random Access Memory (RAM), which is used to load up programs while your computer is in use, in that the NAND memory cells can reliably maintain their contents without the continual application of electric current.  This simply means that data stored on NAND chips has an extremely long life even if the drive is not turned on, for say years.  Conversely, pull the plug on RAM and the data stored in it quickly degrades and the memory must be reloaded on the next boot cycle.

The NAND flash chips used in SSDs come in two varieties: single-level cell (SLC) and multi-level cell (MLC).  SLC devices can only store one bit of information per cell, but have the distinct advantage of faster access times for writing to the cells.  MLC devices, however, can store more than one bit of information per cell by having different levels of electrical charge applied to the cell, but generally take 3-4 times longer to write a bit to the cell.  Assuming the manufacturing process size is the same, MLC devices always have a higher capacity than SLC devices with the obvious major drawback that you will get slower write speeds than with an SLC device.  Additionally, due to the nature of MLC cells, they also have a substantially shorter lifespan than comparable SLC parts.  Manufacturers have priced their components according to these differences in speed and reliability and therefore the SSDs geared at the consumer market segment are primarily MLC based.

SSDs originally have been the storage medium of choice for high-grade commercial and military devices.  If you ever wondered why that military laptop could be dropped 15 feet and still be more than a pile of scrap after the event, SSDs are one of the reasons.  Unfortunately, until the past 6 to 9 months, that durability also carried with it a military-grade price tag.  While the chips for SSDs are more difficult to produce than RAM, and are binned much higher than the NAND flash used in USB thumb drives, there has also been a significant period of cost-recovery going on for the past 2 years that has kept prices artificially high.

So why the push to introduce SSDs to the consumer segment now?  The reasons are many, but include three primary factors.  First and foremost is that SSD manufacturing technology has finally progressed to the point where the devices are faster than high-end consumer hard disk drives (for the most part) and have become more reliable during the development process.  By its very nature, NAND flash can only sustain a limited number of read/write cycles before it becomes unreliable.  This cycle number is one of the key components in determining the Mean Time Between Failure (MTBF) of the drive.  Improved NAND chips, as well as the implementation of wear-level algorithms to ensure that no group of cells is written to more than the average, has brought SSD reliability up to at least the level of consumer-grade Hard Disk Drives (HDDs).

Second, the number of devices using NAND flash has increased exponentially over the past 10 years and now includes cell phones, flash drives, and portable media players to name a few.  The need for ever increasing and faster flash to run these devices has spurred investment by major integrated circuit companies like Samsung, Intel, Micron, and Hynix to expand capacity to meet demand.  In doing so, these companies also look to sell any excess capacity that they might have, especially their highest-binned chips which they can sell out at a premium price.

[img]intel-ssd-potential.jpg[/img]

The third and final driving force behind SSD adoption is the enthusiast segment itself.  While we have seen great strides over the past 20 years in CPU, GPU, and RAM speeds, the performance curve that HDDs sit on is far lower.  The limitations to HDD performance are inherent in their design.  You can only spin a drive so fast before disk wobble makes reads unreliable on the outer segments.  Additionally, repositioning the read/write heads via the controller arm takes time and costs precious milliseconds before data can be accessed.  Some of these limitations have been alleviated by banding drives together in RAID 0, but the reality is that HDDs are still the greatest bottleneck in a system and have increased in speed and size at a far slower pace than most other components.

SSDs over the past two years have made remarkable strides in both capacity and performance.  For the most part, high-performance drives have been relegated to SLC-based designs, and thereby smaller capacities.  MLC-based designs have been able to achieve far higher capacity, but their write speeds have suffered and been typically just a fraction of a good 7200RPM HDD.  The high cost of the drives coupled with these drawbacks has significantly hindered SSD adoption rates.  However, as the NAND chips that populate these drives have improved, so has performance.
 

The JMicron Controller Issue

[img]jmicron.jpg[/img]

The first foray into "budget" SSDs was made by a group of companies (G.Skill, OCZ, Patriot, and others) around July 2008 utilizing JMicron’s JMF601A/602A 8 channel flash controller coupled with Samsung NAND MLC flash memory.  Devices based on the JMicron controller sported sizes of 32, 64, and 128GB with up to 120-140MB/s read and 80-90MB/s write speeds.  These drives were targeted specifically at users that wanted to try out SSDs but were unwilling to fork out $1000+ for SLC-based drives.  Initial prices were $170-$480, based on capacity, a discount of approximately 50% off the going rate per GB for SSD drives.

Unfortunately, what users of these SSDs quickly found out is that the drives were flawed when writing small files, say 8K and smaller.  The drives would "stutter" on average for about 250ms before completing the write operation.  This fatal flaw gave users a far from smooth computing experience.  JMicron was quick to address the issue and release a revised controller, the JMF601B/602B, to improve random write latency.  The new controller added 16KB of on-chip cache that did help with the stutter problems, but performance was still far from ideal as many users still complained of slow writes when doing heavy I/O tasking.  Since the release of the revision B controller JMicron has acknowledged the issues with the JMF601/602 series and encouraged their use only in low I/O devices, such as Netbooks.  JMicron is currently designing a new controller for more robust I/O performance, but it is not expected to be released before Q3 2009.
 

Intel – the X25-M

[img]X25-M_combo.jpg[/img]

 

Approximately 3 months after the JMicron-based SSDs were released, Intel stormed onto the SSD scene with its MLCbased drives.  Instead of copying the competition, Intel rolled its own design for their MLC-based SSD line.  The Intel X25-M uses Intel’s own SATAII controller chip employing native command queuing (NCQ) to access 10 channels of 50nm MLC NAND-flash simultaneously.  The entire setup is buffered with 16MB of high-speed cache.  From a design perspective, the X25-M effectively is running a 10-unit RAID 0 array under the control of a very efficient controller.

Intel’s specifications for the X25-M claimed a less than 0.1ms access time and sustained read and write speeds of 250MB/s and 70MB/s, respectively.  Once out on the market, this drive was one of the few MLC-based SSDs that lived up to its hype.  Unlike the JMicron-based drives, there were no issues writing small files and even with mediocre write performance users have been exceptionally pleased with the performance of the X25-M.  The X25-M has not been without its own set of issues, however.  Early revisions of the X25-M exhibited speed degredation issues after extendend periods of heavy writes.  Intel appears to have since corrected this issue with revised firmware.  Our sample has been running smoothly under heavy write conditions for 3 months now and still benches the same was when it was purchased.

Like most things, however, there was a catch.  Intel knew that the performance of the X25-M was head and shoulders above the competition, and they priced the drive that way as well.  Initial prices for the 80GB X25-M were almost $600 and remained there, or higher, for the first 3 months after introduction. Recently Intel cut pricing on the X25-M 80GB to just under the $400 mark.  This has helped Intel to remain competitive as other companies second generation SSDs, like the G.Skill Titan, are coming to market and hope to cut the performance gap and take market share from Intel.
 

The computer market, especially the enthusiast segment, moves at lightning speed.  Because of the flawed design with the JMF601/602 controllers, SSD companies needed to re-think their strategy if they wished to remain competitive.  Unfortunately, because no one other than JMicron and Intel currently make consumer-grade SATAII SSD controllers, the options these companies faced were limited to either sticking it out with the flawed first generation products until the new JMicron controller arrives in Q3 2009, or coming up with a way to make what they have work better.  Obvious choice, right?

 

[img]titan128.JPG[/img]

The solution these companies came up with was remarkably simple, and elegant.  Two controllers are better than one.  To overcome the buffer fill limitations of the JMF602B controller and prevent stuttering, G.Skill implemented a design that puts dual JMF602B controllers under direct control of a single JMicron JMB630 RAID controller in a RAID 0 configuration.  The dual-controller design should theoretically give the Titan twice the throughput of drives using a single JMF602B controller and thereby eliminate, or at least greatly reduce, the small file write issues.

We think including a DRAM buffer similar to that used in the Intel X25-M would have aided performance as well, however it is unclear if this enhancement was left out of the final design because of cost considerations or because of difficulties integrating it with the 3 controllers now on the Titan SSD.

 

G.Skill was kind enough to send us a 128GB Titan to reivew and here are some snapshots of the little gem.

[timg]G.Skill_1.jpg[/timg]     [timg]G.Skill_2.jpg[/timg]

[timg]G.Skill_3.jpg[/timg]     [timg]G.Skill_4.jpg[/timg]

SSD benchmarking has not turned out to be as straightforward as we would hope and our readers would expect.  Most of the applications out on the market for measuring storage performance have been tailored to how HDDs have worked for decades.  Many, but not all, of these applications rely on "spot checking" the read/write speeds of HDDs over various areas of the disk and then building a composite from these fragments of data.  While this method is entirely valid to determine HDD performance, with SSDs these methods fail to produce consistent results because the wear-leveling algorithms in the SSD firmware that prevent the benchmarking applications from reading and writing to the same cells in a consistent manner.  Therefore, any testing methodology for SSDs needs to include synthetic tests that evaluate the entire drive at multiple file sizes, as well as real world performance metrics such as boot time and application load times.

Keeping these principles in mind, we decided to use the following benchmarks in our analysis:

  • ATTO 2.34 – A good metric for determining Input/Output Operations per Second (IOPS) as it measures transfer speed from the drive using various file sizes.
  • HD Tune 3.50 – Tests sustained file transfer read and write speed over large portions of a drive.
  • Everest Disk Benchmark 4.60 – Measures read and write speed of pre-determined block sizes from 4k to 1MB in size over the entire capacity of a drive.
  • PCMark Vantage HDD Suite – Includes timed tests for Windows Defender, game level loading, picture loading, Windows Vista start-up, video editing, Media Center file performance, adding music to Windows Media Player, and general application loading.  A composite score is generated from the performance of these individual sub tests.  Xtreme CPU only uses the composite score in our reporting, as the Professional version of Vantage that allows viewing of the sub-scores is prohibitively expensive.
  • Windows 7 Beta 1 Start-up Time – Measured from selection of SSD/HDD from boot menu to full load of applications after automated login.

 

 

Testbed

  • Intel Core i7 920 running at 3.70GHz
  • Copper Thermalright Ultra-120 Extreme
  • ASUS P6T Deluxe running BIOS 1102
  • Patriot Viper PVT36G1600ELK 6GB (3 x 2GB) PC3-12800 DDR3 Memory at 9-9-9-24 timings
  • Dual eVGA 9800GX2 Video cards in Quad SLI
  • All drives connected to SATA Port 0 of ICH10R controller for testing
  • Windows 7 Beta 1

* Windows 7 Beta 1 was chosen as the benchmarking OS because it is the first Microsoft OS that is purported to be SSD optimized.  While we have no way to verify this claim as of yet, we felt we should give it a whirl and see.

 

 

G.Skill Titan FM-25S2S-128GBT1 2.5" 128GB SATA II SSD

[img]G.Skill-Specs.jpg[/img]

 

Intel X25-M 80GB

[img]Intel-Specs.jpg[/img]

 

Samsung Spinpoint F1 HD103UJ 1TB

[img]Samsung-Specs.jpg[/img]

Samsung Spinpoint F1 HD103UJ 1TB

[img]ATTO%20-%20Samsung.jpg[/img]

We chose the Samsung F1 1TB HDD as our baseline marker of what a user can expect out of good hard drive for $100 or less right now.  In ATTO this drive hits its stride at 32KB chunks and larger and can maintain read and write speeds just in excess of 100MB/s in these file sizes.  Small file I/O speed, by comparison, is significantly slower.

 

Intel X25-M 80GB

[img]ATTO%20-%20Intel.jpg[/img]

The Intel X25-M shows exceptional read speeds, up to as much as 270MB/s.  Write speeds, however, are much slower and max out at 80-95 MB/s.  Interestingly, both read and write speeds obtained on this drive were both in excess of the rated technical specifications listed by Intel.  The small file I/O performance (4KB and smaller) on the X25-M surpassed the other two drives in ATTO by at least a factor of two.  Users constantly accessing small files will notice that the X25-M has a much quicker load time than the other drives.

Observant readers will note the significant drop in read and write performance at 512KB block size.  We observered this drop in speed on every run of ATTO, however the block size at which it occured varried in every single run.  We strongly suspect that what we are seeing here is a flushing of the cache on the X25-M that slows down performance temporarily.  The drop always occured at block sizes above 128KB, and never fully crippled or stalled the drive, but did show up on our tests.

 

G.Skill Titan FM-25S2S-128GBT1 2.5" 128GB SATA II SSD

[img]ATTO%20-%20G.Skill.jpg[/img]

The Titan did not show any of the cache flushing phenomena we observed with the Intel SSD.  This makes sense, as the Titan only has dual 16KB caches on the JMF602B.  Read speeds were very good on the Titan, topping out at about 200MB/s, the rated speed for this drive.  Write speeds were by far the best of the bunch once we got above the 8KB block size.  Users shuffling larger files to and from the Titan will appreciate this performance.

 

Samsung Spinpoint F1 HD103UJ 1TB

[timg]HD_Tune_-_Samsung_-_Read.jpg[/timg] [timg]HD_Tune_-_Samsung_-_Write.jpg[/timg]

The Samsung F1 1TB HDD shows a nice 100+MB transfer rate on the outer sectors that gradually declines to 52MB/s on the inner sectors of the drive.  This performance pattern is typical of HDDs.  Access time on the F1 is a mediocre 13+ms.

 

Intel X25-M 80GB

[timg]HD_Tune_-_Intel_-_Read.jpg[/timg] [timg]HD_Tune_-_Intel_-_Write.jpg[/timg]

Read and write speeds on the X25-M were close to the Intel specified values.  No surprises here.  Access time was a brisk 0.1ms, right at the rated specs.

 

G.Skill Titan FM-25S2S-128GBT1 2.5" 128GB SATA II SSD

[timg]HD_Tune_-_G.Skill_-_Read.jpg[/timg] [timg]HD_Tune_-_G.Skill_-_Write.jpg[/timg]

Read speeds on the Titan were consistently fast, starting out at 150MB/s and topping out at about 200MB/s.  Linear writes on the G.Skill, however were extremely erratic.  We repeated this test no less than 8 times, on two different computers, to verify the results.  During all the write tests using HD Tune Pro 3.5 the drive would oscillate between 0.1MB/s and 150MB/s writes.  We suspect that in this test, writing to the drive continuously saturates both of the JMicron JMF602B controllers and they must pause before they can continue.

 

The Everest HDD Benchmark measures read and write speed over the entire capacity of a drive.  We set the block speed manually to every setting from 4KB to 1MB for both the G.Skill and the Intel and ran the tests three times.  Because Everest reads and writes to the entire capacity of the drive, it takes quite a while for each run at each block size (from 20 to 90 minutes).  All in all, the data represented here took about 2 days to collect and we feel represents what a user can expect out of each drive in terms of performance at various file sizes.  Because of the large size of the Samsung drive (1TB), we did not include it in these tests, as each run would have been approximately 4-12 hours long.

 

[img]Everest_Linear_Read.jpg[/img]

Both drives offer exceptional read performance in the large block tests, but the Intel X25-M shows the strengh of it’s more advanced controller with smaller block transfers.  At 4KB the X25-M starts off at almost 4 times the transfer rate of the G.Skill Titan and maintains a significant advantage throughout the testing.  These results are very similar to what we observed in the IOPS measurements using ATTO.

 

 

[img]Everest_Linear_Write.jpg[/img]

Write tests for the G.Skill Titan were surprising, to say the least.  We repeated this set of measurements no less than 3 times to verify our results.  In Everest the G.Skill Titan struggles with write speed until we get to the very large block size of 1MB.  Similar to the write issues we saw with HD Tune, we suspect that the JMF602B controllers are just not up to the task of writing a large number of files to the disk.  While there was no stuttering observed during these tests, the write speeds were a disappointment.

 

For those wishing to see the raw data for each performance point of the graphs above, we have included screencaps of all the Everest runs in the Appendix of this article.

 

The HDD testing suite of PCMark Vantage runs a series of reads and writes on a drive to simulate boot time, application loads, and file writes to give a performance rating that can be used to evaluate a drive’s real world performance.

These tests include:

  • Windows Defender
  • Game level loading
  • Picture loading
  • Windows Vista start-up
  • Video editing
  • Media Center file performance
  • Adding music to Windows Media Player
  • General application loading.

Xtreme CPU only uses the composite score in our reporting, as the Professional version of Vantage that allows viewing of the sub-scores is prohibitively expensive.  Each score represents the average of 3 individual runs.

 

[img]Vantage.jpg[/img]

The Titan has its best showing by far in this benchmarking suite, just edging out the Intel X25-M in overall performance.

 

As mentioned in our testbed description, we chose Windows 7 Beta 1 as the OS for SSD testing.  The reasoning behind this is simple: Microsoft has openly stated that Windows 7 will be their first OS that is optimized for SSDs.  This is significant because the kernels for Windows XP and Windows Vista contain decades of optimizations for HDD access patterns and no optimizations for SSD file access patterns.

 

[img]Win7_boot.jpg[/img]

Now this is where you get your bang for your buck in SSDs!  File access patterns for OS booting are primarily sets of random reads.  With HDDs, such as the Samsung F1, access time plays a key role here becuase you have to reposition the read/write head over the next piece of data before the process can continue one.  With SSDs you just make a call to the appropriate memory cell, immediately load the data, and move on to the next set of instructions.  Boot up is also a read heavy process, and it is no surprise to us that the X25-M wins out in this test.

 

After using the G.Skill Titan for a week, and the Intel X25-M before that, we would like to give our readers our impressions on both drives in day to day usage.  While we did not notice any stuttering with the G.Skill Titan, under heavy I/O situations such as multiple application launches, there was a perceptible hesitation with the drive.  For example, on multiple occasions we noticed that opening Firefox would result in a framed window with no contents for approximately half a second, then the program would resume and our homepage load.  Disk activity on the drive was pegged out during these hesitations and we suspect that the combination of continual reads and small file writes to the Firefox cache saturated both controllers on the Titan.  The Samsung F1 HDD did not exhibit this issue either, but admittedly program load times were far from snappy compared to either of the SSDs.

With the X25-M application loads were nearly instantaneous, even when accessing multiple programs simultaneously.  Day to day usage with the Intel X25-M was extremely smooth, with the one exception of anytime we made substantially large writes to the drive.  In contrast to the G.Skill Titan there were no hesitation issues, but the overall write speed of the X25-M is somewhat disappointing after you get accustomed to the exceptional read speeds.  Early revisions of the X25-M reportedly showed large drops in write performance after extensive heavy usage.  Our sample has shown no issues in this regard, even after 3 months of extremely heavy use.
 

Conclusions

These are very exciting times for the enthusiast computer user.  Steady improvements in performance coupled with dropping prices should make SSDs one of the most sought-after upgrades for 2009.  The G.Skill Titan is a good step in the right direction, bringing drives with better performance and lower prices.  The innovative design of using dual controllers in an internal RAID 0 configuration is a step in the right direction, although we feel like the overall design is a little incomplete.  Adding a significant cache to the system would have most likely alleviated the small write issues we see with the drive and we hope to see this implemented in the next revision, whenever it is set to be released.  Overall the G.Skill makes a great showing and improves leaps and bounds from the first generation JMicron drives it is set to replace.  With that said, we still feel that the Intel X25-M gives better performance, albeit at a higher price per GB.

Pros:

  • Up to 200MB/s and faster sustained read speeds.
  • Up to 160MB/s sustained write speeds on large files.
  • Dual internal JMicron Controllers in RAID 0 eliminates the stutter problems of previous generation.
  • Lower power consumption than traditional HDDs.
  • Capacity up to 256GB in a 2.5" form factor.
  • 2 Year warranty.
  • <2ms access time.
  • Durable casing.

Cons:

  • Mediocre write performance with small file sizes.
  • Cost per GB (still an issue for all SSDs, and not exclusive to this drive).

Recommendations

Despite our complaints about the small write performance of the G.Skill Titan we at Xtreme CPU cannot over-emphasize how pleasurable it is to use an SSD for the boot drive on your computer.  Boot times are almost cut in half, programs load exceptionally quick, and the drives are completely silent.  Either the G.Skill Titan or the X25-M would make an exceptional addition to any enthusiast’s computer.  Furthermore, a RAID 0 setup of either of these drives would simply be heavenly in terms of performance, which will be the subject of a subsequent article where we compare performance of 2 G.Skill Titans and 2 Intel X25-M in RAID 0.

 

Discuss in the forums

We know how many users want to "see the data for themselves" and we appreciate that kind of critical eye.  Just for you we have included screen caps of each Everest run from our SSD testing.  Just click on the image for a larger version.

 

Intel X25-M 80GB

Read   4K                                       8K                                        16K                                        32K

[timg]Everest_Intel_Read_4KB.jpg[/timg] [timg]Everest_Intel_Read_8KB.jpg[/timg] [timg]Everest_Intel_Read_16KB.jpg[/timg] [timg]Everest_Intel_Read_32KB.jpg[/timg]

Read   64K                                  128K                                     256K                                      512K

[timg]Everest_Intel_Read_64KB.jpg[/timg] [timg]Everest_Intel_Read_128KB.jpg[/timg] [timg]Everest_Intel_Read_256KB.jpg[/timg] [timg]Everest_Intel_Read_512KB.jpg[/timg]

Read   1MB

[timg]Everest_Intel_Read_1MB.jpg[/timg]

 

Write   4K                                       8K                                        16K                                        32K

[timg]Everest_Intel_Write_4KB.jpg[/timg] [timg]Everest_Intel_Write_8KB.jpg[/timg] [timg]Everest_Intel_Write_16KB.jpg[/timg] [timg]Everest_Intel_Write_32KB.jpg[/timg]

Write   64K                                  128K                                     256K                                      512K

[timg]Everest_Intel_Write_64KB.jpg[/timg] [timg]Everest_Intel_Write_128KB.jpg[/timg] [timg]Everest_Intel_Write_256KB.jpg[/timg] [timg]Everest_Intel_Write_512KB.jpg[/timg]

Write   1MB

[timg]Everest_Intel_Write_1MB.jpg[/timg]

 

 

G.Skill Titan FM-25S2S-128GBT1 2.5" 128GB SATA II SSD

Read   4K                                       8K                                        16K                                        32K

[timg]Everest_G.Skill_Read_4K.jpg[/timg] [timg]Everest_G.Skill_Read_8K.jpg[/timg] [timg]Everest_G.Skill_Read_16K.jpg[/timg] [timg]Everest_G.Skill_Read_32K.jpg[/timg]

Read   64K                                  128K                                     256K                                      512K

[timg]Everest_G.Skill_Read_64K.jpg[/timg] [timg]Everest_G.Skill_Read_128K.jpg[/timg] [timg]Everest_G.Skill_Read_256K.jpg[/timg] [timg]Everest_G.Skill_Read_512K.jpg[/timg]

Read   1MB

[timg]Everest_G.Skill_Read_1MB.jpg[/timg]

 

Write   4K                                       8K                                        16K                                        32K

[timg]Everest_G.Skill_Write_4KB.jpg[/timg] [timg]Everest_G.Skill_Write_8KB.jpg[/timg] [timg]Everest_G.Skill_Write_16KB.jpg[/timg] [timg]Everest_G.Skill_Write_32KB.jpg[/timg]

Write   64K                                  128K                                     256K                                      512K

[timg]Everest_G.Skill_Write_64KB.jpg[/timg] [timg]Everest_G.Skill_Write_128KB.jpg[/timg] [timg]Everest_G.Skill_Write_256KB.jpg[/timg] [timg]Everest_G.Skill_Write_512KB.jpg[/timg]

Write   1MB

[timg]Everest_G.Skill_Write_1MB.jpg[/timg]

SHARE THIS POST

  • Facebook
  • Twitter
  • Myspace
  • Google Buzz
  • Reddit
  • Stumnleupon
  • Delicious
  • Digg
  • Technorati

Leave A Response