An SSD is one of the best upgrades you can make today – a machine booting and working with files from a traditional hard drive with spinning platters often sits waiting as the drive seeks away. By using flash memory, where there is no mechanical parts, reliability, noise, power consumption and speed can all be improved.
Early SSDs suffered performance inconsistencies due to differences in flash wear levelling algorithms, limited embedded controller performance, less flash memory channels and slower flash memory interfaces. Early adopters also experienced failures, which has soured the SSD experience for some, due to firmware bugs which can render the data on a drive inaccessible.
Unfortunately, SSDs have been quite expensive and the capacities are still relatively limited compared to hard drives, as flash memory demand has remained high and manufacturing is energy intensive. Over the past few years, the drives have hovered between AU$90-120 for 120/128Gb models and AU$180-240 for 240/256Gb models.
Important specifications are as follows:
- Read: up to 540Mb/s
- Write: up to 300Mb/s
- Consumption: 3.5W (max), 0.4W (idle)
- Weight: 71g (max)
- Form-factor: 7mm
- Global Wear Levelling
- 55-bit BCH ECC
- SMART Monitoring
- 3-year Warranty
I was intrigued – What chipset does it use? What NAND chips are used? How does it perform? Is it worth the buy? So I bought one!
I ordered the product on Monday, and it arrived Friday morning. It came in a large (for an SSD) colour cardboard box.
The box is a desktop upgrade kit, rather than just the bare drive. The front side lists some of its most important features, interestingly it says RAID capability which implies it has some wear-levelling and garbage collection that is satisfactory even in non-TRIM applications.
The rear has extended specifications. Kingmax have been a Taiwanese brand that have been around for a while – I know them best from their flash memory cards and DRAM products which have always been value priced.
Included is a small user manual leaflet, which features some warranty information. It also has small screws to secure the 2.5″ drive, and large screws to secure the 3.5″ bracket to the computer. Of course, the bracket is also provided, as well as a black SATA cable and molex to SATA power adapter cable. This is a very complete kit, especially for the price, although in my case, the SATA cable seems to be broken – the drive does not connect when the included cable is used, but an alternative spare cable works fine.
The drive itself is housed in a black metal shell, which feels a little hollow. The shell itself is in two parts, secured by regular Philips head screws. There is a warranty label over the seam and screw itself.
It’s a new product, it’s got a warranty label so what the hell? Well I couldn’t resist the urge to find out what it was made of, so I took it apart and nulled the warranty on the first day.
Internally, it’s got a label that says 5 years warranty, curiously. The warranty for this product is only three years though. Maybe the SME series has common drives with different labels outside …
The memory itself comes from Micron Technology and appears to be 20nm MLC 128Gbit NAND from the markings. Near each flash package, there are four ceramic capacitors to buffer the power, which is great.
The rear also features a Nanya NT5CB128M16FP-DI 256MiB DDR3-1600 DRAM cache, made Week 52 of 2013.
Nearer the SATA converter appears to be a switching converter to derive the lower voltages for the flash, DRAM and controller.
The underside follows the same pattern with a total of 16 NAND packages on the board. There are diagnostic jumper pads as well for programming the controller, and a space for an EEPROM that isn’t populated.
The controller itself is a Silicon Motion SM2246EN. It appears to be dated week 33 of 2013. I haven’t had any experience with Silicon Motion SSD controllers, however, I have used them in flash memory cards, USB memory keys (earlier Corsair Flash Voyagers) and in the CF-controller that powered the original Asus EeePC701 and that seems to have performed fine. As it has no real heritage, I cannot comment on its long term reliability or whether there are any bugs.
The PCB itself appears to be made by Tak Yan Electronic Shenzhen (TYE) and is dated 27th December 2013 with a code of SSD-S3B3-S14A. It’s a very recent product!
The input side seems to feature a fuse for catastrophic failure protection, but the spaces for the population of a zener diode for transient overvoltage protection are not fitted. Also to note is that there are no thermally conductive pads to sink the heat from the board, and it is likely a choice to keep the price down.
Some of the pads near the controller appear to be used to fit surface mount resistors to configure the controller.
There’s no more mystery now!
I think the thing buyers want to know most is how does it perform? Well, I hope to answer this by testing the SSD using some of the most popular synthetic benchmarks. It was installed in a system using an AMD Phenom II x6 1090T BE @ 3.90Ghz, in a Gigabyte 890FXA-UD7 running Windows 7 Professional. The AMD SATA drivers were used, and the drives were attached to the on-board chipset-hosted SATA III ports.
The comparison benchmarks are the Kinmax SME35 Xvalue 240Gb with firmware M1024E, Samsung 840 Pro 256Gb which was secure-erased and updated to the latest firmware (DXM06B0Q) and the OCZ Vertex 3 120Gb (SF-2281 based, Sync NAND) which was also running the latest firmware (2.25). Unfortunately, as the OCZ Vertex 3 is running the system, and is not fresh state, the performance benchmark results for the Vertex 3 are not directly comparable. Likewise, the Samsung 840 Pro is a completely different calibre of drive, representing the performance end of the market so isn’t directly comparable either.
If I had a dedicated SSD test rig, and dedicated test SSDs, I might be able to do a better job. The use of the AMD controller may also reduce performance results compared to testing with an Intel SATA3 port, so just be aware of that. This is the performance that I get – your mileage may vary!
Fresh out of the box, the SSD shows the SMART data on the left. After all of my testing has been run, the SMART data appears as on the right.
The main values that have changed appear to be C3 (0 -> 15), A4 (1->3892), A5 (1->6), F1 (0->24104), F2 (0->47640). From the values, I would have to assume that C3 may be a wear level delta or wear levelling value. A4 might be related to data written. A5 might be related to the number of whole-media cycles used. F1 and F2 seem to be related to data written and read respectively, although increasing at about 30 counts per Gigabyte.
Without the help from the manufacturer, we won’t know any of this for sure, which is a little annoying. The only variables that are failure predictive are C4 Reallocation Event Count, C7 Unknown, A5 Unknown and A6 Unknown. The rest are informational, as their thresholds are zero.
It also seems that the smart data logging isn’t quite accurate either – the temperature worst value isn’t logged, but depending on how it is mounted, under heavy loads without ventilation I have seen the temperature get to 59 degrees C.
The performance of the Kingmax SME35 in sequential read and write are close to expectations, having reached 514.7Mb/s read and 265.7Mb/s write. The strange shape on the write test seems to suggest that there are different dies with slightly different speeds on the drive itself. It’s not symmetrical because the space between 240 and 256Gb is reserved for overprovisioning and is not user accessible.
The read and write access benchmarks are as follows.
The drive really hits its stride when it comes to write operations at 64kB, and read operations at 1Mb. IOPS values from different benchmarks can vary significantly.
For comparison, here are the screenshots from the same benchmarks being run on a Samsung 840 Pro, read averaging 527.3Mb/s and write averaging 440.0Mb/s.
The benchmarks weren’t run on my OCZ Vertex 3 as it was a boot drive that was very “dirty”, and wouldn’t really give an accurate benchmark when it comes to the full surface.
It seems to show performance in-line with its specifications, and it’s not too shabby when it comes to 4k performance.
By comparison, here is the OCZ Vertex 3 (note, as it is my boot drive, and it is dirty, not directly comparable). Because the SandForce controller implements data compression, the results are different for incompressible (default, on the left) versus compressible (on the right).
The Samsung 840 Pro, our performance reference, doesn’t seem to be too much faster in any regard, and is even slower sometimes. It looks like this Kingmax value series drive is putting up quite respectable performance figures.
AS SSD Benchmark
The results are slightly different to CrystalDiskMark, but very respectable.
In comparison, here is the Samsung 840 Pro – considering the price difference, the Kingmax is not too bad.
The Vertex 3 is badly battered though … and that’s because its performance is best only with compressible data which some of the boot drive data is. It’s also in a dirty state – but there’s another thing which can explain the poor synthetic performance …
AS SSD Copy Benchmark
Another decent result from the Kingmax.
The Samsung 840 Pro for reference, is slightly faster.
The Vertex 3 however, started behaving inconsistently.
Notice the wide variance between these two runs? It turns out that I’ve triggered the write throttling on this drive due to the extensive number of benchmarks run, and it’s artificially limited the write performance to avoid premature flash wearout during the warranty period. Sorry folks, but the Vertex 3’s synthetic numbers are just artificially low and not directly comparable as a result. Unfortunately, many SandForce drives do have this, so it’s not easy to get a real bearing on how the SandForce drives compare – but the user experience is still good.
AS SSD Compression Benchmark
This benchmark confirms that the Silicon Motion controller doesn’t utilize compression techniques, which means predictable throughput regardless of data but also potentially higher levels of flash wear compared to a drive with compression.
For comparison, the Samsung 840 Pro also doesn’t compress its data.
The older OCZ Vertex 3 based on the SF2281 does, and hence the slope in the lines.
ATTO Disk Benchmark
ATTO seems to turn out higher numbers, but I’m not sure why. It’s still pretty much “as the box says”.
For reference, the Samsung 840 Pro.
And the OCZ Vertex 3 – it seems ATTO uses compressible data so the OCZ looks better.
Anvil’s Storage Utilities
Another decent result it seems. I didn’t want to wear out the SSD, so I didn’t run any endurance benchmarks.
For reference, the Samsung 840 Pro, which (as expected) benches better.
And the battered OCZ Vertex 3 …
H2testw was used to ensure the drive does not corrupt any data. I also manually filled the whole drive with random data and ran multiple MD5 passes to ensure consistency. In no case did the drive show any inconsistencies.
The read performance on H2testw was a bit low – but this is a limitation of H2testw. The same test run on a faster Samsung 840 Pro showed it hitting the limitation on both reading and writing.
The power consumption of the drives were made by cutting apart a Molex to SATA adapter, and plugging it into my secondary machine with the Kiethley Model 2110 5.5 digit multimeter performing the measurements under the control of a computer.
A plot of the current consumption was made during a run of CrystalDiskMark with the drive attached to a SATA2 port, as this computer had no SATA3 ports. For comparison, the OCZ Vertex 3 was replaced by my Kingston SSDnow V300, another SandForce based drive, as I couldn’t pull the OCZ out for test purposes. I also compare a 2.5″ rotating platter hard drive, a Western Digital WD16000BEVS 160Gb drive I had in spare.
Despite the 2A rating on the label, the Kingmax does turn in a very good power consumption figure. This test is primarily of interest to people seeking to install the drive into a laptop, as it will give you an idea of whether you will gain any battery life benefits.
The summarized average power consumption in Idle, Read and Write are as follows:
Drive Idle Read Write Kingmax SME35 Xvalue 240Gb 60.5mA 216mA 513mA Kingston V300 120Gb 119mA 372mA 590mA Samsung 840 Pro 256Gb 59.7mA 300mA 386mA Western Digital WD1600BEVS 220mA 680mA 700mA
It’s not a clear victory between the Kingmax and Samsung 840 Pro, but it seems the Kingston SSDNow V300 trails behind. Of course, the hard disk idles at a much higher power, consume much more power when reading, and a little more when writing. It makes good sense to change to an SSD in a laptop!
The package inclusions are great and the price is very alluring. Some corners have been cut, but in general, it does everything it says on the tin. Internally, it is powered by a Silicon Motion SM2246 controller with Micron 20nm MLC NAND. It passed every test I threw at it, so at least when it is fresh, it’s a solid product. Unfortunately, there’s no way to tell for sure if there are any bugs which might impact on the stability or performance over time, or whether there might be any reliability issues.
A direct comparison between the Kingmax and the Samsung 840 Pro can be made, performance wise, but it has to be kept in mind that the 840 Pro is a performance drive commanding close to AU$1/Gb, whereas the Kingmax is commanding only AU$0.58/Gb. The OCZ Vertex 3 as it was “dirty” as my boot drive, wasn’t able to show its best side and can’t be directly compared.
There are higher performance options, but you are paying AU$1/Gb or more. For a value drive, it does appear to be a decent performer, likely to be sufficient for many users, and at a very attractive price. The SMART data isn’t very descriptive, but it seems to show that the drive didn’t suffer much throughout the testing regime and should have some level of endurance (unlike some former value options – e.g. OCZ Petrol).