Crucial M4 128 GB and 256 GB

Crucial M4

Micron announced the RealSSD C400 at CES on the 4th January. A development of the RealSSD C300, this SSD is known as the M4, Crucial being Micron’s general consumer arm. Launched one year earlier, the C300 stood out from other SSDS due to its adoption of the SATA 6 Gbits standard which opens the door to theoretical speeds of 600 MB/s, as against 300 MB/s for SATA 3 Gbits. Combined with aggressive pricing and high end random access performance, this was one of the flagship SSDs of 2010.

The M4 is also based on a Marvell 88SS9174 controller but in its BLD2 or BKK2 revision, while the BJP2 is used on the C300. According to Crucial, the firmware is what makes all the difference between the C300 and the M4 and not the controller revision. We actually managed to get hold of a 256 GB version in the BLD2 and BKK2 revision and performance levels were identical. It came with a 256 MB DRAM cache, compared to 128 MB on the C300, and the memory used is now 25nm IMFT, while it was 34nm IMFT before.

Here are the figures announced by Crucial in terms of performance:


Read speeds are up 16.9% at equal capacity. Write speeds are up by between 26.7 and 20.9%. Although random writes are also up (+11.1 to 33%), the same can’t be said for random reads: 60,000 IOPs on the C300 but just 40,000 IOPS on the M4.



The Crucial M4 128 GB tested combines a Marvell 88SS9174 controller, a Micron DRAM chip and 16 Micron 29F64G08CFACB flash chips. These chips are engraved at 25nm and combine two 32 Gb dies. The pages are 4 KB in size and a block is 1 MB.



The Crucial M4 256 GB uses different flash chips to the 128 GB version and has 29F128G08CFAAB chips instead. Their capacity is doubled as they use two 64 Gb dies. This time the pages are 8 Kb and the blocks 2 MB.

On the Crucial M4s, 6.9% of the flash memory is used for wear leveling and internal optimisations, which corresponds to the difference between the size of the flash chips (given on a basis of 1 KB = 1024 bytes) and the capacities of the SSDs (given on a basis of 1 KB = 1000 bytes).

OCZ Vertex 3 120 GB and 240 GB

OCZ Vertex 3

The OCZ Vertex 3 is the first SSD to use the new generation SandForce SF-2000 controller. OCZ and SandForce have been very tight since the release of the Vertex 2s, which were also the first SSDs equipped with the previous generation controller. This SSD was, like the C300, one of the standout SSDs in 2010.

According to SandForce this new controller gives both reads and writes of up to 500 MB/s, namely double the speeds of the previous generation, while IOPS have also been doubled from 30,000 to 60,000. In practice, with the Vertex 3, OCZ is combining the general consumer and high end versions, the SF-2281 with 25nm IMFT memory. As with the Vertex 2s, the SandForce doesn’t use any external memory, which makes the SSD cheaper to produce.

For now, only 120, 240 and 480 GB versions are available, which leaves the field to the Crucial M4 in terms of the 60/64 GB versions. Note the disparities in performance between these three models, with the 120, 240 and 480 GB versions scored respectively at:

- Reads: 550, 550 and 530 MB/s
- Writes: 500, 520 and 450 MB/s
- Random reads: 20K, 40K and 50K IOPS
- Random writes: 60K, 60K and 40K IOPS

All these figures were obtained with highly compressible data, the most favourable type. To recap, in contast to other controllers, the SandForces compress data before writing it to flash memory. While this procedure is a boon with compressible files (text files for example), recompression isn’t really effective with files that have already been compressed, ie. those that generally take up most space (audio, photos, videos or simply RAR/ZIP type files).

In spite of the use of highly compressible data you can see that the 120 GB is significantly down on the 240 and 480 GB versions in terms of random reads.




In its 120 GB version, the Vertex 3 uses a SandForce SF-2281TB1-SDC controller (the end of the codename is masked), combined with sixteen 25nm Intel 29F64G08ACME2 chips. Unfortunately we don’t know how these chips are organised internally.



The 240 GB version tested, a preseries model, uses an SF-2281VA1-SDC-ES controller. ES must be what is masked on the 120 GB model. The 120 GB model was made for market but OCZ appears to be using SandForce ES chips, or prototypes, for SSDs that are nevertheless on sale in stores! This isn’t the first time we’ve seen this and it isn’t necessarily a problem as there hasn’t necessarily been a revision between this and the final version. It does explain however why the Vertex 3s were released ahead of the other models. The flash chips are 25nm Micron 29F128G08CFAAB chips that combine two 64 Gb dies and are made up of 8 KB pages and 2 MB blocks.

On the Vertex 3s, some flash space is used for wear leveling, internal optimisations and SandForce RAISE technology, which is designed to prevent the loss of data should part of the flash memory fail.

Intel SSD 320 120 GB and 300 GB

Intel SSD 320

The Intel SSD 320 is the new marketing name for what is in fact a third generation X25-M. Launched in September 2008 with 50nm flash, the first X25-Ms were replaced by a second version, codename Postville, in summer 2009. These new SSDs used 34nm memory, which opened the way for a reduction in costs and therefore sales price, and supported TRIM. The Postvilles dominated the SSD market in 2009 alongside the OCZ Vertex SSDs that were based on the Indilinx Barefoot controller.

What was called “Postville Refresh” on the Intel roadmaps is in fact an update of the current X25-M second generation. Still using the SATA 3 Gb/s standard on an Intel PC29AS21BA0 controller, this time they have 25nm flash memory. This is less costly per GB than 34nm memory but also has a shorter lifespan. The MTBF remains at 1.2 million hours, without Intel giving any figures on the volume of data that can be written to the SSD, except a minimum, whatever the model, of 20 GB per day over 5 years (typical usage is about 5 to 10 GB).

In terms of performance, here are the figures announced for the range, with capacities from 40 to 600 GB (!):


Here are the current X25-V/X25-M specs for comparison:


The performance levels announced are up, especially when it comes to sequential writes, with gains of between 28% and 65% on the previous generation. The same goes for random writes, with between 48% and 144% improvement! In sequential reads, the gain is between 8% and 17%, against between 8% and 20% improvement for random reads. Latency is however up slightly, except on the 40 GB version in writes, while energy consumption at idle is also up very slightly.



The SSD 320 120 GB has an Intel PC29AS21BA0 controller and 64 MB of Hynix DRAM. The flash is organised in quite an original fashion with four Intel 25nm 29F64G08ACMEI 8 GB chips and six 29F16B08CCMEI 16 GB chips, making a total of 128 GB of flash memory. Intel had to use this combination because its controller can apparently only function with 5 or 10 flash chips, which means that with same size chips you can only have 80 to 120 GB of space.



The 300 GB version includes twenty 25nm 16 GB Intel 29F16B08CCMEI flash chips and includes two 64 Gb dies.

On the Intel SSD 320s, around 12.7% of the flash is used for wear leveling and internal optimisations.

Intel SSD 510 120 GB and 250 GB

Intel SSD 510

By keeping more or less the same controller since 2008, Intel has obviously allowed itself to be surpassed in terms of performance. While waiting for the arrival of a new in-house controller, the company recently launched a new SATA 6 Gbps SSD range, the Intel SSD 510 Series.

Using 34nm flash memory, they’re based, like the C300 and the Crucial M4, on a Marvell 88SS9174. Not the same revision however. The BKK2 is used on the Intel SSD 510, whereas the BJP2 is used for the C300 and the BL02 for the M4. Both Crucial and Intel say however that the revision isn’t as important as the firmware, which, both say again, has been developed in partnership with Marvell.

Peformance is announced as follows:


The sequential speeds are very good, 10 to 20% up on the M4s but this is to the detriment of random accesses which are between two and six times lower. Compared to the Intel SSD 320 Series, speeds are also significantly up, while random accesses are down.



The 120 GB version of the Intel SSD 510 therefore uses a Marvell 88SS9174-BKK2, 128 MB of Hynix DRAM and sixteen 34nm 29F64G08CAMDD flash chips.



The 250 GB version however uses 29F16B08JAMDD chips.

Around 12.7% of the flash is used for wear leveling and internal optimisations on the 120 GB version, compared to 9.1% on the 250 GB version.

Test protocol

Test protocol

As we said in our article on SSDs, TRIM and IOMeter, testing SSDs correctly in an environment compatible with TRIM is not necessarily an easy exercise. Various parameters need to be taken into account and the fact that SandForce controllers include data compression algorithms only complicates things further.Some very popular tools such as CrystalDiskMark give turnkey solutions, but these results are unfortunately not entirely conclusive. For example, the test file is only 4 GB in size, which is fine for a quick test but gives pretty variable results. Also the random tests only address the SSD across a restricted area.

Other tools such as h2Bench, HD Tune or HD Tach were designed for testing hard drives and sometimes operate on drives that contain no data. They can then get mixed up during read measurements, as random reads are turned into sequential reads by some controllers when an SSD is empty!


The marvellous box of tools that is IOmeter is what’s known as a synthetic benchmark. We used it to measure performance in several cases:

- Sequential reads by blocks of 2 MB
- Sequential reads by blocks of 4 KB
- Random reads by blocks of 4 KB
- Sequential writes by blocks of 2 MB
- Sequential writes by blocks of 4 KB
- Random writes by blocks of 4 KB

This gives us a vision of the speed of any given storage device in terms of throughput, but also in terms of I/0s. The tests on 4 KB blocks were carried out with 1, 2, 4, 8, 16 and 32 simultaneous commands so as to highlight the controller’s ability to parallelise accesses.

Next we ran the practical tests, first of all with reads and writes of various groups of files. These groups were composed as follows:

- Extra large: 731.17 MB on average
- Large files: 5.2 MB on average
- Medium sized files: 800.88 KB on average
- Small files: 48.78 KB on average

We used a RamDisk as the source or the target for reads or writes on the SSD. In view of the high performance of recent SSDs and so as to obtain results that are subject to as little variation as possible, we used Robocopy with an in-house piece of software which enables us to carry the tests out repeatedly.

To finish with we carried out some purely practical tests, namely timed operations after installation of Windows 7 (64 bit) on each of the SSDs:

- Boot Windows 7
- Installation of Photoshop CS 5
- Start up 3D Studio Max
- Start up 3D Studio Max + Photoshop + Word + Excel
- Launch a game in Civilization V
- Launch a game in Crysis 2

We tried to add another test halfway through, a test that involved processing groups of images in Photoshop. The problem is that even with a simple processing task such as JPEG recompression, the performance of the various devices don’t have much of an impact.

We left out benchmarks such as PC Mark Vantage, which are halfway between synthetic and practical, or any of the other pieces of software that repeat recorded accesses, known as traces. We don’t consider such results to be pertinent as they consist in carrying out accesses as fast as possible on the storage devices. They tell you that device X is the fastest but if at the end of the day practical performance levels are limited by the machine’s processing capabilities, what’s the point?

If we had recorded the accesses carried out by Photoshop in processing groups of images and used a trace benchmark, we would typically have observed differences of a magnitude of around 200% between SSDs, while in practical usage there’s no difference. Also, these traces simply record the type of access but don’t take into account the content, which can show SandForce controllers in a more positive light depending on whether compressible data or data that has already been compressed is used.

The test configuration consisted of a Core i7-2600K mounted on an Intel DP67BG with 16 GB of DDR3-1333 and a Radeon HD 5870. The synthetic and file copying tests were carried out in Windows 7 (64 bit) with the storage system tested as the secondary drive and the boot drive a Crucial M225 64 GB, all with Intel RST 10 drivers in AHCI mode. The other tests were carried out with the test storage device as the primary drive.

For comparison we also included the star SSDs from the previous generation, namely the:

- Corsair Force 120 GB (SandForce SF-1200 + 34nm IMFT flash)
- Crucial C300 128 GB (Marvell 88SS9174 + 34nm IMFT flash)
- Crucial C300 256 GB (Marvell 88SS9174 + 34nm IMFT flash)
- G.Skill Falcon II 128 GB (Indilinx Barefoot + 34nm IMFT flash)

Sequential speeds

Sequential speeds

We started the tests with sequential writes, measured using IOMeter. We tested accesses by 2 MB blocks, in reads and writes. As we have in the other pages of the report, we’ve chosen to display several performance graphs that can be consulted by moving the mouse over the links situated beneath the graph. The graphs displayed are as follows:

- 3G/6G – Incompressible data:
Results on the Intel SATA 6G motherboard port, therefore at SATA6G on those SSDs that support this standard, or at SATA3G for the others. Random data used in IOMeter, which makes it resistant to compression by the SandForce controllers.

- 3G/6G – Compressible data:
For the SandForce SSD we used non random data here (a series of 0s or 1s), which was therefore highly compressible on SandForce SSDs. This didsn’t change the results on the other SSDs and we’ve therefore given the same results as previously.

- 3G – Incompressible data:
Results using the Intel SATA 3G motherboard port for the SATA 6G SSDs, so as to limit to this transfer speed, with random data.

- 3G – Compressible data:
Results using the Intel SATA3G motherboard port for the SATA 6G SandForce SSDs, with non random data.


[ 3G/6G – Incompressible data]  [ 3G/6G – Compressible data ]
[ 3G – Incompressible data ]  [ 3G – Compressible data ]
The two Vertex 3s are alone in giving read speeds in excess of 500 MB/s, whatever the type of data used. Note that in contrast to the previous generation, the type of data has little impact on speeds. Next come the Intel SSD 510s, the Crucial M4s and the C300s, the only old generation SATA 6G range. While the C300s managed 350 MB/s, the fastest SATA 3G SSD was the X25-M with 262 MB/s. The SSD 320s don’t do a great deal better with 271 MB/s.

In writes the SSD 320s do however give a gain as at equal capacity (120 GB) with speeds of 131 MB/s against 111 MB/s on the Intel X25-M120. This is however a long way behind the SSD 510 250 GB which gives speeds of 316 MB/s, followed by the Vertex 3 240 and the M4 256 GB. In its 120-128 GB version, the SSD 510 remains in the lead but the M4 overtakes the Vertex 3.

The Vertex 3 120 GB os a good way down on the 240 GB version when it comes to sequential writes, in contrast to what is stated in the official specs. True, with highly compressible data, the Vertex 3 240 and 120 GB give explosive results, recording speeds of 462 and 451 MB/s respectively and leaving all the competiton trailing in their wake. Unfortunately this sort of posturing doesn’t do a great deal for these models in practice as the large files we generally transfer on our computers are to a large extent made up of data that can’t be compressed much by the SandForce controller as they have already been compressed (audio, photo, video files etc.).

Naturally, the new generation SSDs are greatly limited by the SATA 3G port. They then give comparable levels of performance to the old guard in terms of reads but gains are still visible in writes.

Random reads

Random reads

We then moved on to random accesses, looking at reads first of all. Of course in standard usage, reads are far more common than writes and when it comes to a system disk, random accesses are of primary importance, especially in intensive multitasking. A storage device’s capacity to process a large number of random accesses, something that is directly linked to access times, is what will save it from slowing down the rest of the PC when it’s asked to access data situated in several places at the same time.

A standard hard drive with an access time of 10ms won’t manage more than 100 operations per second (IOPS), while SSDs have access times of between 0.13 and 0.23ms, or 7800 to 4300 IOPS. A huge difference! As in the previous test we measured random accesses on an Intel SATA 6G port and, in addition, on an Intel SATA 3G port for the 6G SSDs and with compressible or incompressible data with the SandForces.

The tests were carried out with 1, 2, 4, 8, 16 and even 32 simultaneous accesses (QD1 to 32 in the graphs). This gives us an understanding of an SSD’s capacity to process these accesses in parallel, with the ideal scenario being that performance is doubled between 1 and 2, 2 and 4 and so on. While it’s definitely worth checking this out, we shouldn’t lose sight of the fact that in standard usage, the level of simultaneous accesses is somewhere between 1 and 4.

MB/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]
IO/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]

MB/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]
IO/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]
While the new generation SSDs give substantial sequential speed gains, the same can’t be said for random speeds. Worse, none of the new SSDs can match the performance levels managed by the C300!

This is notably the case for the Crucial M4, which is however the fastest of the new SSDs tested. The 128 GB version actually obtains markedly better results than the 256 GB version. The explanation for this is in fact fairly simple and is down to the memory chips used. On the 128 GB version, the chips used are organised with pages – the smallest readable unit – of 4 KB, whereas the 256 GB version is constituted of 8 KB pages. To read the 4 KB blocks required in this test, the 256 GB version therefore has to access a full 8 KB page.

The Intel SSD 320s are in second position with not much difference in performance between the two capacities. They’re more or less comparable to the X25-M from QD1 to QD8 and do better beyond that.

The Intel SSD 510s and Vertex 3s are the slowest of the new generation SSDs here. The Intel SSD 510s are faster from QD1 to QD4, the most common loads in standard usage, while the Vertex 3s are in front beyond this.

The SandForce SSDs do vary depending on the type of data, but not by much. The impact of the 3G limitation on 6G SSDs is generally low, but noteworthy, especially with the C300 256, C300 128 and M4 128, which are then limited to 190 MB/s at QD32 while they manage 220 – 236 MB/s at SATA 6G.

Random writes

Random writes

After random reads, we went on to test random writes. As we’ve already said, reads are generally much more frequent than writes on your standard desktop and having high random write performance only really comes into its own for serveur type usage. Measuring random writes allows us to check an area of performance that sometimes betrayed early SSDs (the sadly notorious OCZ Cores based on the JMicron JMF602).

As in the previous test we carried out the tests on the Intel SATA 6G port, or at 6G for SSDs that support this mode and 3G on the others, but also on a 3G port so as to limit the 6G SSDs in this mode. We carried out the tests on the SandForce SSDs with compressible and incompressible data. Lastly, we carried out tests with 1, 2, 4, 8, 16 or 32 simultaneous accesses.

MB/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]
IO/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]

MB/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]
IO/s: [ 3G/6G - Incomp. ]  [ 3G/6G - Comp.]  [ 3G - Incomp. ]  [ 3G - Comp. ]

The most attentive readers will have noticed that random writes are a good deal better than reads. Why is this? When you write data randomly to an SSD, the SSD actually writes the data sequentially, while reorganising its internal allocation table to make the addresses known by the OS on the storage system (the LBAs) correspond to the right memory cells.

This time the new generation offers gains on the old generation. With incompressible data, the Crucial M4s take over at the top, but the Vertex 3s are in the lead at QD1 as well as with compressible data, with an advantage for the 120 GB version over the 240 GB. This strange behaviour is mirrored with the SSD 510s, the 120 GB version outperforming the 250 GB model. Note that level of perfomance on the Intel SSDs is only very slightly up on the previous generaton and that they’re a good way behind the new leaders. The SATA 3G limitation only really has an impact on the M4 256 GB and the Vertex 3s when they’re processing compressible data.

Practical tests: Files

Practical tests: Files

Moving onto the practical tests, we started by looking at write and read speeds for various groups of files. These groups were composed in the following way:

- Extra large: 731.17 MB on average
- Large files: 5.2 MB on average
- Medium sized files: 800.88 KB on average
- Small files: 48.78 KB on average

We used a RamDisk as the source or the target for reads or writes on the SSD. In view of the rapidity of recent SSDs and so as to obtain results that are subject to as little variation as possible, we used Robocopy with an in-house piece of software which enables us to carry the tests out repeatedly.

[ 3G/6G - Reads ]  [ 3G - Reads ]
[ 3G/6G - Writes ]  [ 3G - Writes ]

[ 3G/6G - Reads ]  [ 3G - Reads ]
[ 3G/6G - Writes ]  [ 3G - Writes ]
In reads, as you’d expect, the fastest speeds are achieved with the largest files. We scored the Vertex 3240 and the Intel 510 250 at 399 MB/s, with the Crucial M4s hot on their heels. With 5.2 MB files, the gaps are smaller and while the Vertex 3s still have the lead, the Intel 510s and Crucial M4s aren’t far behind.

The results for small files are very interesting as only the OCZ Vertex 3s and Intel SSD 510s do better than the C300s, which were previously the fastest. The performance levels on the Crucial M4, however, drop a good deal, though they’re still reasonable, with the 128 GB version faster than the 256 GB. The Intel SSD 320 Series models are however a long way back!

When all the SSDs are limited to SATA 3 Gbps, the gaps between them are levelled off and you can see that changing the interface even has a negative impact in transfers of small files.

With large files, the Intel SSD 510 Series 256 takes the lead, followed by the Vertex 3 240 and the Crucial M4 256. The difference between the two is fairly small and the M4 has the advantage when you compare 120/128 GB versions. While the specifications announced for the M4s were confirmed by these tests, with the C300s coming in slower, not all the SSDs across the board live up to their specs!

The Vertex 3s for example are clearly down on the speeds of 500 MB/s or more announced in the specs, the 120 GB version as much as three times slower! This illustrates the issue of theoretical figures obtained with highly compressible data as compared to our test files, which are made up of photo, audio and video files, namely the most common big files found on PCs.

With small files which are more easily compressible, the Vertex 3s dominate in the same way as the Corsair F120 did in the past. Without data compression, the Intel SSD 510s are however very competitive! The M4 is then some way back, even a little slower than the C300.

Practical tests: Applications

Practical tests: Applications

To finish with we carried out the purely practical tests, namely timed operations after installation of Windows 7 (64 bit) on each of the SSDs:

- Boot Windows 7
- Start up 3D Studio Max
- Start up 3D Studio Max + Photoshop + Word + Excel
- Launch a game in Civilization V
- Launch a game in Crysis 2
- Installation of Photoshop CS 5

We did our best to limit the variations as much as we could, however, as the timing was carried out by hand and the results themselves can vary slightly, there is a definitely a small margin of error here. As you can see, the results for the various SSDs are very close one to the next and there was no point in trying to test for the differences between SATA 3G and SATA 6G: only the 6G results are therefore given.

We tried to add another test halfway through, a test to process groups of images in Photoshop. The problem was that even with a simple processing task such as JPEG recompression, the performance of the various devices didn’t have much of an impact. Processor speed is what limits performance here, not the storage device. We also thought we might combine some measurements with writing files at 5 MB/s, simulating a download, but the impact was minimal and not outside of the margin of error for timing.

For comparison purposes we added a hard drive, the very rapid Western Digital Caviar Black 2 TB, and the score obtained on an 8 GB RamDisk! With sequential speeds measured at over 5 GB/s and 100,000 IOPS in random 4KB accesses (QD1), the RamDisk is 10-15 x faster than the best SSD!


Windows 7 start-up was timed from the appearance of the load screen to full load of desktop (hourglass disappears). The results are very close across the board as the worst SSD, the Falcon II, boots Windows in 15 seconds, against 13.8 seconds for the best, the Crucial C300 256 GB. In fact only 3 SSDs take over 14.5 seconds: the Falcon II and the SSD 320 120 and 300 GB.

The times for 3D Studio Max start-up are also close, varying between 19.5 seconds on the Vertex 3 240 aand 22.7 seconds on the Falcon II, which, along with the SSD 320s, are the only solutions higher than 22 seconds.

With respect to the combined launch of 3D Studio Max, Photoshop, Word and Excel, the gap between the various SSDs is also quite small, with the fastest taking 20 seconds and the slowest 23.8 seconds – SSD performance is much much better than the hard drive, measured at 71.5 seconds here!

This last test is in fact the most representative of the gains SSDs can give in your day-to-day work, even if launching four applications at once isn’t something you’d often do. Sure, as we saw for Windows or 3ds on their own, an SSD is fater than an HDD, but this is accentuated as the number of simultaneous accesses goes up: launching three additional applications has hardly any impact on the launch of 3ds on an SSD (between 0.4 and 1.1 seconds slower) while this process adds 37.5 seconds with the HDD. With an SSD, the responsiveness of your machine will remain at a very good level even when you’re doing intensive multitasking, which certainly won’t be the case with an HDD.

It is however difficult to state a preference for one SSD over another as the times obtained are very similar from one model to the next. While the old generaton G.Skill Falcon II is a little behind, this is also the case with the brand new Intel SSD 320s, which are slower in practical tests than the X25-M. This is no doubt linked to their poor scores for reads of small files. The OCZ Vertex 3, Crucial M4 and Intel SSD 510 are all very close to one another, but don’t really do any better than the previous generation C300s.

The RamDisk does give peformance gains but they remain quite low given the gigantic difference in performance in comparison to SSDs on paper. Accessing data in double quick time is all very well but when it doesn’t simply consist of reads/writes, this data still has to be processed by the system.

Practical tests: Applications (cont.)

Practical tests: Applications (cont.)

Now we come to the loading of games. We used two games here, Crysis 2 and Civilization V.


As with the previous tests, there’s only a very slight difference in performance levels between the SSDs, with the gain over a standard hard drive less significant here. Loading data quickly is all well and good but you still have to process it and this is what takes most time when you load up a match already underway in a game. Once again the Intel SSD 320 is very slightly behind, while the Vertex 3, M4 and SSD 510 and C300 are all very close one to another. The RamDisk makes some difference in Crysis 2 but this is negligeable in Civilisation V. Of course SSDs do reduce any in-game load times, definitely a positive. In some games such as WoW, they also make quite a difference in terms of load times, but we weren’t able to include this title in our test because of the random nature of loads linked to the fact that the number of characters in a certain place can vary.

This brings us to the final test, a timed installation of Photoshop CS5 from the archive downloaded from the Adobe site. This installation is carried out in two stages: decompression from the archive and then the actual installation.


The advantage of the SSDs over the HDD is quite small and the HDD is actually faster than the slowest SSDs. Overall, there’s not a great difference in performance and the gains given by RamDisk remain limited.

Performance over time & TRIM

Performance over time & TRIM

As we’ve already mentioned on several occasions on this website, an SSD’s performance levels can deteriorate over time. There are several reasons for this, the first being structural: a hard drive can read, write (unwritten space) or rewrite (space that has previously been written to) data by 4 KB packages. With flash memory you can only read, write or rewrite by 4 KB packages, 4 KB and … 512 KB (or even 8 KB, 8 KB and 2048 KB for a 25nm 8 GB flash chip).

When a space that is already occupied by a file has to be rewritten, there can be an impact on performance! Worse still, if the file has been deleted by the OS and the OS doesn’t support TRIM, the SSD then doesn’t know that the file has been deleted and acts as if it had to rewrite the data rather than write it. This problem is however resolved with TRIM as the OS then tells the SSD that the space is to be considered as unwritten.

This structural issue is accentuated by the existence of optimisations inside SSDs aimed at improving random write performance and write amplification. Simply put, when a request to write data randomly to an SSD is received, the SSD writes it sequentially to the flash memory, making sure that the addresses in its internal allocation table correspond to the addresses known by the OS (the LBAs) and the corresponding flash pages. For such a mechanism to be efficient, the blocks of flash memory do have to be available, which can be problematic without TRIM support.

What does all this translate to in practice? Testing the impact of wear on an SSD’s performance is no easy thing. We nevertheless carried out a few tests on the 120/128 GB versions of the Intel SSD 510, Intel SSD 320, OCZ Vertex 3 and Crucial M4.

The idea behind our test is simple. We put the SSDs in an extreme situation. We created a partition, leaving only 10 GB of space on the SSD that we then filled with video files. By leaving just a small amount of available flash on the SSD, this accelerated any performance deterioration.

We then carried out several tests for 3 minutes with IOMeter on the remaining 10 GB, using data that can’t be compressed a great deal:

- Random writes (= "New" graph)
- Sequential writes (= "Used" graph)
- 10 GB formatting using the TRIM command
- Sequential writes (= "TRIM" graph)
- Random writes (= "Used" graph)
- 10 GB formatting using the TRIM command
- Random writes (= "TRIM" graph)

Finally, we reset the SSD and, after putting it back in the same state, we carried out sequential writes to the SSD (= "New" graph)

Here are our results:


[ New ]  [ Used ]  [ Trim ]
Writing to a space previously used for sequential writes doesn’t have much of an impact on random writes, as you can see if you compare the differences between the ‘new’ and ‘used’ graphs. The impact of TRIM support can’t therefore be measured in this case. Note however that the Intel SSD 320 is least impacted, with peformance only dropping off at the end of the test.

In fact the SSD 320’s performance does drop after 210 seconds, when the SSD has already written 7.59 GB to the 10 GB partition (but write amplification has to be added to this). The dip in performance actually comes latest on the Vertex 3, after 180 seconds, by which time it has already been asked to write 13.5 GB to the 10 GB partition, but without helping it with TRIM! On the SSD 510 and Crucial M4, the dip in performance comes more quickly, after 60 and 90s, or respective writes of 3.9 and 6.4 GB. This is still a good level of performance.


[ New ]  [ Used ]  [ Trim ]
Things change with sequential writes. On a new SSD, the level of performance remains stable throughout the test, even though 30 to 45 GB of data is being written continuously. The only exception to the rule is the Vertex 3, whose performance takes a hit after 60s (or 9.4 GB written), dropping from 161 to 142 and then 132 MB/s.

This can be felt when writing sequentially to a space that was previously written with random writes, except on the SSD 320, which is hardly affected. Performance levels on the SSD 510, M4 and Vertex 3 then drop to 120, 83.1 and 48.3 MB/s respectively in the first 30 seconds! As you continue to write to the used SSDs however, performance gradually return to the sort of levels you’d expect and by the end of the test they’re all quite close to their initial performance.

Using TRIM allows the Crucial M4, Intel SSD 320 and 510 but not the Vertex 3 to kick in with their initial levels of performance from the start. The Vertex 3 stays at around 130 MB/s (against 160 MB/s when ‘new’) and even suffers a few other dips during the test. Remember however that we have put the SSDs in what are extreme conditions for the SandForce controller here as we used data that either wasn’t easily compressed or couldn’t be compressed at all, whether this was videos stored on the first partition or writes carried out with IOMeter.

In the end, the new SSDs do pretty well and even though we do advise you to use an OS that is TRIM compatible, it isn’t totally out of the question to use them when TRIM isn’t supported.

Energy consumption

Energy consumption

Finally we measured energy consumption with a clip ammeter. Energy consumption was taken at idle and in one of the most intensive SSD loads, sequential writes.


At idle, the Falcon II, an older generation SSD, is the most economical, at just 0.4 Watts, whereas the Vertex 3 240 draws the most (1.65 Watts)! Remember, the Vertex 3 240 is a preseries model and its firmware doesn’t include the energy economy modes during inactivity, unlike the 120. The scores for the two Vertex 3s in their final versions should be pretty similar.

During sequential writes, the Corsair F120 and the Intel SSD 320 120 draw the least power. They are however also among the lowest performance solutions in this domain, in contrast to the SSD 510 250 and Vertex 3 240 which draw over 4W. The C300 256 GB also consumes over 4W, but the M4 256 manages to be a good deal more economical in spite of its high performance with sequential writes. Whatever the SSD used however, they’re more economical than standard 3"1/2 drives by far. Standard HDDs struggle to manage anything under 3W at idle and can go as high as 8W in load (the 1 TB versions for example).

Conclusion

Conclusion

With throughputs of up to 500 MB/s, the new generation SSDs are a mouthwatering prospect on paper. These sequential speed scores need to be put in context however and the dip in random access reads takes them below the Crucial C300s in terms of performance in this domain.


Overall, none of these new SSDs really stands out from the lot. The Vertex 3s give the highest performance in sequential reads but the SSD 510s dominate in writes once you move onto data that can’t be compressed by the SandForce controller, which is more realistic. Note for example the difference in the Vertex 3 120’s handling of the different types of data, going from an announced speed of 500 MB/s to 451 MB/s measured with highly compressible data and just 163 MB/s with incompressible data in our test files.

The Crucial M4 is a long way ahead of the pack with random reads, the 128 GB version having the advantage over the 256 GB due to the memory chip architecture (4 KB vs 8 KB pages). In random writes the Vertex 3s and Crucial M4s share the lead.

Note that we haven’t mentioned the Intel SSD 320s. This is because they are never the highest performance solution. Limited to SATA 3G, in practical tests its level of performance is lower than the X25-M, whether when reading small files or in the various applied timed tests. A part to avoid!

On the other hand, in our practical tests, the Crucial M4s, OCZ Vertex 3s and Intel SSD 510s are all pretty much on a par. This new generation doesn’t give much of a gain over the Crucial C300s when it comes to use as a system disk and the absolute gain remains very low even in comparison to the good old G.Skill Falcon II (based on the Indilinx Barefoot) and X25-M! As shown by our readings taken on a RamDisk, which is however 15x as fast, there isn’t much room for improvement: accessing the data is one thing but it’s quite another to process it when this doesn’t simply consist of reads / writes.

What new generation SSD can we recommend out of the Crucial M4s, OCZ Vertex 3s and Intel SSD 510s? You’ll basically have to see how these three excellent solutions are priced and make your choice accordingly. For the moment Crucial seems to have the advantage as the 128 and 256 GB versions cost €220 and €430 respectively, while the Intels and OCZs cost between 10-15% more for the 120/250 and 120/240 GB versions! Another positive when it comes to the Crucial M4s is that a 64 GB version is also listed, while such a capacity hasn’t yet been announced for the SSD 510 and Vertex 3 ranges.