By now, you have probably heard all about Solid State Drives (SSDs). They are one of the most beneficial upgrades for your computer, and those who upgrade to an SSD will almost never be able to go back. So, let’s say you are convinced; you know the size you need and you’ve decided you’ll get one. Wallet in hand, you are ready to buy one, but you realize there are a lot of them. Which is the best one? This is the question I am here to answer today. Which is the best SSD in a given size class?
First, let me lay out what I am focusing on here: I’m limiting myself to the mainstream consumer SSD market. This means I won’t be taking SLC based drives into account. Their high price pushes them into a separate category. I’m also limiting myself to SATA interface drives, so no PCIe drives or PATA drives. Still, I hope this article will meet the needs of most who read it.
I’ve split the SSDs up into size brackets of 50 GB, up to 299 GB. After that, the available number of drives drops, so I bundled together the sizes into a 300-499 GB category and a 500 GB+ one. I will list the best drive in each category – performance-wise – but with some reliability and ease-of-use considerations taken into account as well. They will not necessarily be the best value drives. I will list a number of runner ups in each category as well a winner; I will, of course, also provide the reasoning behind my choices.
I have collected all the drives available into a spreadsheet. It is not quite complete. There are some really old drives missing and there are new drives showing up every week, but it should cover majority of all SSDs available in this category at the time of publication. You can find it here.
There is one other issue I should discuss before I get to the meat of the matter: the size of the SSD. I’ve put the SSDs in their size category according to the advertised size. But this may not be the same as the actual formatted size available in Windows. Part of this problem you are perhaps already familiar with from standard hard drives: the confusion with decimal (SI) and binary (JEDEC) definitions. A gigabyte, using the SI prefix, is 1000000000 bytes, but using the JEDEC prefix is 1073741824 bytes. One gigabyte can refer to two different values depending on the definition one is using; these are often both denoted ‘GB’. The hard drive manufacturers use 1 GB = 109 bytes but Windows uses 1 GB = 230 bytes. So when you buy a hard drive advertised at 500 GB and put a partition on it, in Windows it ends up being only 466 GB.
With SSDs it gets even more complicated. Not only do you have the problem of SI versus JEDEC, but there is also the issue of over-provisioning. SSD controllers keep some space to themselves so that they have some spare area as a scratchpad to help work around the read-modify-write issue and for wear leveling. But this means that the flash visible and usable to the OS is less than the flash present on the drive. So, a 128 GB drive might reserve as much as 8 GB as a scratchpad. Different SSD makers follow different practices for what they advertise as the size of the SSD. For example: the G.Skill Falcon 128 GB and the OCZ Vertex 120 GB are the same drive underneath but have different advertised sizes.
And the winners are …
up to 49 GB
Winner: Corsair Nova V32
This category is actually one of the more complicated ones. SSDs get their high read/write performance compared to flash cards and thumb drives by having multiple channels on the controller and reading and writing to multiple NAND chips at the same time. In order to get the size of the drive down the SSD makers leave out some NAND chips, but this leaves them with fewer NAND chips to write to. This means all drives in this category suffer performance-wise in different ways.
The winner is the Corsair Nova V32 drive. It uses the new revision of the Indilinx Barefoot controller: the Barefoot ECO. The normal Barefoot controller had some issues working with 30 nm flash, evident in the limited performance of the OCZ Solid 2 drive which has the Barefoot controller and Intel 34 nm flash. The Barefoot ECO fixes this issue and even manages to lift the performance above the normal Barefoot controller with its preferred Samsung flash.
This was by no means a clear victory, for there is also the Intel G2 controller based Intel X25-V 40 GB drive. It has both better random read and write speeds than the Corsair Nova drive but is hampered by an anemic 40 MB/s sequential write speed. This means that in many reading-based usage scenarios it will be faster but the moment you actually write a lot of data to the drive it falls way behind. That is why I didn’t make it the winner. If your usage doesn’t include any big writes to the drive, then the Intel would be the better choice.
The third contender to the throne is OCZ Vertex Turbo 30 GB drive. It uses binned versions of the good old Indilinx Barefoot controller, but is overclocked by 8%. This lifts its performance up above the rest of the Indlinx based drives by a few percent and helps it almost catch the Corsair Nova, and maybe just beat it in some sequential write benchmarks.
Bringing up the rear are the rest of the Indilinx Barefoot based drives: the OCZ Vertex, the Corsair Extreme, the Super Talent Ultradrive ME and the A-DATA S592. All of them are basically the same drive with different stickers on them. Performance-wise they fall only a few percentage points behind the OCZ Vertex Turbo and the Corsair Nova drives.
50 – 99 GB
Winner: OCZ Vertex 2 50 GB
This is a packed category. Almost every controller on the market has a presence here, but when it comes to performance one controller dominates: SandForce. The winner of the SandForce drives is the OCZ Vertex 2 50 GB. More than one review site has declared the Vertex 2 the fastest SSD they’ve ever tested. It uses the consumer version of the controller, the SF-1200. The main differences between the SF-1500 and the SF-1200 are that SF-1500 based drives come with a super capacitor to avoid data loss even in the event of a power outage, and the controller is tweaked to take advantage of it. The SF-1200 also has an artificial limit on its random write performance. OCZ has managed to do a deal with SandForce to get a special firmware without this performance handicap. This lifts it above the rest of the SF-1200 based drives.
The first of the runners up is the 60 GB Vertex 2. Hardware-wise this is the same drive as the 50 GB version. The difference is in firmware: while the 50 GB model reserves 28% of the flash for over provisioning, the 60 GB version only reserves 13%. This comes with a few percentage point performance disadvantage in exchange for a better $ per GB ratio.
While only the OCZ Vertex 2 gets the high performance firmware for SF-1200 officially, Corsair has managed to achieve something similar. The performance handicap was only implemented in the final release firmware: the release candidate version of the firmware didn’t have it. Corsair took the release candidate firmware, disabled a power saving feature that didn’t work and then shipped the drive with that firmware. Thus, the Force gets the same performance as the Vertex 2. The question that remains is what will Corsair do with future firmware updates?
Following on are the OCZ Agility 2 50 GB and the G.Skill Phoenix 50 GB. These are the normal SF-1200, without any firmware tweaks but with the big 28% over provisioning. A little below them in performance are the OCZ Agility 2 60 GB and the Mushkin Callisto 60 GB. They are still normal SF-1200 drives, but with smaller 13% over provisioning and the associated performance drop.
Lastly, are the only non-SandForce drives in this size category: the good old Intel G2 controller based drives. The reason they make it is that they do have higher random read than any of the SandForce drives and lower latency. In writes, though, they fall way behind. The G2 based drives are the ones with 34 nm flash.
100 – 149 GB
Winner: OCZ Vertex 2 100 GB
This is the biggest size category with the most drives to choose from. Still the SandForce drives dominate. The winner this time, as well, is the OCZ Vertex 2 drive, which is pretty much the same as the previous one, only more flash.
As the first runners up is the Vertex 2 100 GB, based on a early SF-1500 design. OCZ produced the Limited Edition Vertex based on a batch of these early controllers. Its performance is the same as that of the Vertex 2, but being a limited edition, I figured it is better to recommend something which has better availability prospects. Then it is the Force F100, the same as in the previous category but with double the amount of flash. The reason I put the Vertex 2 above the Force is because of the uncertainty of future firmware updates to the Force. Performance is similar, with the same 28% over provisioning of flash and the same uncapped random write speed.
The next one is the only non-SandForce drive in this size category, the Crucial RealSSD C300 128 GB. This has the new Micron C300 controller. I had a hard time placing it. In some ways it is faster than the SandForce drives; it has the highest random read performance of any drive on the market and the lowest average response time, as well as the highest sequential read thanks to a SATA 6 Gbps connection. Along with the multiple strong sides the RealSSD C300 has several flaws. The random read is really high, but only if you have your partitions aligned. In a non-aligned situation it suffers from a massive performance drop. While it has really good average response time, its maximum response time is quite high: over 1 second. With such a low average time the maximum response time is very rare, but it indicates that there is some sort of situation where it has problems. The problem is alleviated by the drives support for NCQ. In old-school IDE mode all I/O is synchronous: this means that the I/Os are served one at a time and the following I/O is left hanging until the previous is finished. This was what made the high 1-2 second maximum response time of the old JMicron controller such a big stutter problem. It didn’t support NCQ so all I/Os were synchronous. So when one of those slow 1-2 second I/Os came along the whole queue stalled, since it couldn’t start working on the next one when waiting for the stalled one to finish. This means the whole I/O subsystem stalled and thus every process waiting for a I/O stalled. NCQ means that this is nowhere near as big of a problem. Even if one I/O stalls for 1 second it can continue service other I/Os while waiting for it to finish. The only thing that stalls is the process waiting for that specific I/O. While the SATA 6 Gbps gives the RealSSD C300 the highest sequential read speed of any drive it suffers from a drawback. The performance of an SSD depends on the performance of the SATA controller and the SATA drivers. Currently the best SATA controller and drives are the Intel ones, but they don’t have SATA 6 Gbps. You either end up having to sacrifice random read performance in favor of sequential performance or vice versa. Finally, it has had some firmware issues with ‘bricking’ drives and then the firmware updater ‘bricking’ drives and so on. So, while the C300 is in several ways a faster drive than the SandForce-based ones I felt that its flaws bumped it off the top.
Next up are the Vertex 2 120 GB and the Force F120. These are the same as the earlier ones except with 13% over provisioning. It is also for the same reason I placed the Vertex 2 above the Force F120.
Following these are the OCZ Agility 2 100 GB, the G.Skill Phoenix 100 GB, the Patriot Inferno 100 GB and the A-DATA S599 100 GB drives. All are basically the same drive with different branding. All are SF-1200 with 28% over provisioning. Lastly are the OCZ Agility 2 120 GB and the Mushkin Callisto 120 GB drives. Same drive as the previous ones except this time with 13% over provisioning.
Winner: Intel X25-M G2 160 GB
Runner up: Kingston SSDNow M series 160 GB G2
This is a relatively empty category. SSDs increase in size by going to bigger NAND chips, and they usually increase in power of two. This means most drives jump from 128 GB to 256 GB. The exceptions here are the Intel controller based drives. With ten channels it connects to ten NAND chips. So the 80 GB version connects to ten 8 GB chips and the 160 GB version we have here connects to ten 16 GB chips.
Here picking the winner is easy: Intel X25-M G2 160 GB. It is a bit faster than the 80 GB version with higher sequential write speeds. The runner up is basically the same drive but re-branded as Kingston.
200 – 249 GB
Winner: OCZ Vertex 2 200 GB
There is one other controller which doesn’t do the exact jump from 128 GB to 256 GB, but instead from 100 GB to 200 GB: the SandForce controller. But this also means that there are only SandForce based drives in this size category.
Except for the absence of the Crucial Drive we see the same pattern as in the 100 – 149 GB category. The OCZ Vertex 2 200 GB wins, followed by the OCZ Vertex LE 200 GB, and then the Corsair Force F200 with their 28% over provisioning and high random write speed. Then we have the same Vertex 2 and Force F200 again but with the lower 13% over provisioning. Then the general SF-1200 drives with the 28% over provisioning in many different brands. Lastly we have the SF-1200 drives that have 13% over provisioning.
250 – 299 GB
Winner: Crucial RealSSD C300 256 GB
Runner up: OCZ Colossus 250 GB
In this category there is no SandForce to get in the way of the Crucial RealSSD C300, so it gets to take the top place. Not only that, but it gets to have quite a lead over the runner up. Beyond that there is not much to say. It is the same drive as the 128 GB one, but bigger.
The runner up is a strange beast: the OCZ Colossus 250 GB. First of all, this is the only 3.5 inch drive mentioned here. All the other ones are 2.5 inches. Beyond that, the internals are quite different. It is built around the Indilinx Barefoot controller, not one but four of them. These four Barefoot controllers are hooked up to two two-channel Silicon Image 5923 RAID controllers. These two controllers are then hooked up to a third 5923 controller. That is a lot of controllers for a single drive: this does give it a nice boost in performance, especially in sequential read and write speeds. There are some problems of this design, however. All that RAID setup is still hidden behind a single SATA 3Gbps port, limiting its maximum performance to no better than what a single SandForce controller can do. In random read and writes it does not scale nearly as well because it lacks support for NCQ. The other flaw is that it doesn’t support TRIM either. The TRIM issue is alleviated by OCZ’s garbage collection.
300 – 499 GB
Winner: OCZ Vertex 2 400 GB
In this category there are only SandForce based drives, where one vendor has gone and doubled the size again. The winner is the OCZ Vertex 2 400 GB. For those that aren’t keeping track this is the one with 28% over provisioning. It is followed by the one with 13% over provisioning, the 400 GB version. After that comes the normal performing SF-1200 drives in the form of the Agility 2 drives in a 28% over provisioning and a 13% over provisioning version. At this size, the SandForce drives are a little slower than their smaller brethren.
Winner: OCZ Colossus 500 GB
Runner up: OCZ Colossus 1 TB
In this category there are few drives, mostly because the price of flash is pushing them outside the realm of affordability. The winner is the OCZ Colossus 500 GB, a bigger version of the one we met earlier. The runner up is the slightly slower 1 TB version of the Colossus.
For somebody looking for a big SSD that isn’t a large 3.5 inch drive, you might want to look at the Kingston SSDNow V+ Series 512GB, based on the Toshiba controller. It is quite a bit slower than the Colossus drives and thus I didn’t feel it deserved to be a runner up. The Toshiba controller is based on the new JMicron controller, JMF612, but designed to work with Toshiba NAND. Performance wise it is mediocre, but still a lot faster than a hard drive.
Let me now round up the winners in each category:
|up to 49 GB||Corsair Nova V32||Indilinx Barefoot ECO||32 GB|
|50 to 99 GB||OCZ Vertex 2||SandForce SF-1200||50 GB|
|100 to 149 GB||OCZ Vertex 2||SandForce SF-1200||100 GB|
|150 to 199 GB||Intel X25-M G2||Intel G2||160 GB|
|200 to 249 GB||OCZ Vertex 2||SandForce SF-1200||200 GB|
|250 to 299 GB||Crucial RealSSD C300||Micron C300||256 GB|
|300 to 499 GB||OCZ Vertex 2||SandForce SF-1200||400 GB|
|500 GB and up||OCZ Colossus||4x Indilinx Barefoot, 3x Silicon Image 5923||500 GB|
There is one thing clear here: the SandForce controller dominates. It is the winner in five out of eight categories. The other ones are Indilinx Barefoot ECO, Intel G2, Micron C300 and a four-way Indilinx Barefoot.
In sequential read all the drives perform the same. They bump up at the SATA 3 Gbps limit of 300 MB/s. The reason that the limit is 300 MB/s rather than 375 MB/s as you would expect from 3 Gbps is that the SATA transfer protocol is only 80% efficient and 80% of 375 is 300. Now, none of the drives actually do full 300 MB/s sequential read. This is because there are inefficiencies in the SATA controller and SATA drivers, and there is also overhead to the specific access pattern being done. That is why the actual sequential reading limit tends to vary between benchmarks: they use different access patterns to do the test.
The random read situation is different. In this one, the SandForce based drives fall behind the Micron C300 controller and the Intel G2 controller. If you have a pure read usage situation, the Intel- and Micron-based drives are better choices than SandForce. For such usage the Crucial RealSSD C300 reigns supreme. This is also a metric where NCQ matters, which is why the Barefoot based OCZ Colossus – in not supporting NCQ – does so poorly. Since SSDs get their high performance by reading from multiple NAND chips at the same time, the random read performance depends on the random reads it needs to do being spread out over multiple NAND chips so the controller can do the simultaneous read. NCQ helps with that by letting the controller rearrange read I/Os to keep reading from as many NAND chips at the same time as possible.
Desktop usage is not purely read-based, however. This is why the SandForce controller is dominating in the different categories: its write performance. A normal SSD usually ends up physically writing more data to the drive than what you write to it in the OS. This is because of the read-modify-write issue and wear leveling forcing the drive to move data around causing extra writes. On a bad SSD this write amplification can cause 10x – 20x extra writes. Intel brags about having write amplification of only 1.1x. SandForce ups that by having a write amplification of only 0.5x. It is not because it doesn’t need to do the wear leveling and other moving around of data, but because it compresses the data on the fly. The example they give is that doing a Windows installation results in 25GB of writes to the drive but their compression drops this down to 11GB. This way the SandForce based drives have a sequential write speed that bumps into the SATA 3 Gbps limit. That, of course, still depends on the nature of the data. If the data you are writing already is compressed, like movies, music or images, then it can’t compress it further and the write speed drops to below that of an Indilinx Barefoot based OCZ Agility.
It is not surprising that the SandForce drives also dominate in random writes. Being able to do 160 MB/s lets it take the top spot. But it does not dominate here, for the Micron C300 based drive comes in a close second with 140 MB/s. The Intel G2 drive follows with 45 MB/s and then the Indilinx Barefoot come in last with the OCZ Colossus and Corsair Nova. The lack of NCQ support hurts the Colossus and keeps it way back despite having SandForce levels of sequential write speed. The other non-Indilinx drives use NCQ increase their random write performance by gathering up a bunch of incoming small random writes into a big bunch and then treating that bunch as a single big write and writing it in parallel onto multiple NAND chips. This is why their random write performance increase when you increase the queue depth in benchmarks. It gives the controller more simultaneous small random writes to gather together into bunches, turning the workload into a more sequential write one.
One last issue I should touch upon is reliability. That the MLC NAND chips have a limit to how many times they can be written to is not a big problem. Good wear leveling and low write amplification make sure of this. This is part of why the Intel drives are so reliable: their 1.1x write amplification, and why I expect the SandForce drives to shine in this area with their even lower wear leveling of 0.5x. Of course, the size of the drive also matter here. With the same daily workload you will wear out a small 30 GB drive much faster than a big 256 GB drive. In reality, it is not the know problem or limited writes you should worry about, but the unknown ones. SSDs are new technology and the makers keeps stumbling over issues they hadn’t thought about. I already mentioned that the Crucial RealSSD C300 has had problems. For the SandForce drives I haven’t heard anything, but they are also very new and it may be that there are problems that haven’t been discovered yet. When it comes to reliability the best ones are the Intel drives which have been around a while already and have had time to fix their problems and the Indilinx Barefoot ECO which is just a revision of the tried and true Indilinx Barefoot controller.