Solid state drive (SSD) pricing has taken a pummelling over the last few months with a global flash oversupply. There has rarely been a better time to buy an SSD, especially when considering that many online stores have discount codes and cashbacks which can further reduce the price.
Despite not being in the market for SSDs (as I already have a plentiful supply of drives which serve my needs despite being a little small for my liking), the recent $30 off Newegg Australia promotion really caught my eye. Combining the offer with their listing for an ADATA Ultimate SU650 480GB SSD resulted in a price of just AU$56.49 including GST and shipping, for a net cost of just 12c per GB. That’s quite a bit of a discount compared to the regular prices of the competing drives of the present market and is also cheaper than many of the USB flash drives. That made purchasing the drive almost a no-brainer, but what would it be made of and would it be any good?
The drive comes inside a rather flashy packaging designed for retail, including an almost iridescent, holographic background and clear plastic bubble showing off the drive itself. Because of the low price, the moniker of “Ultimate” feels rather curious – what is it the ultimate of? Performance? Value for money? Low prices? I guess that’s something that’s left to the buyers’ imagination. Of course, at this price point, there are basically zero other inclusions – no adapter cradles, no spacing shims, no screws. The only performance metric on the package seems to be a claim of up to 520MB/s read and 450MB/s write, with the rest of the advertised feature set being practically standard fare for a modern SSD.
Unlike some of the competitors’ more frustration-loaded packages which require scissors to open, this package has a convenient perforation to allow for peeling out the cardboard backing to expose the drive within. Unfortunately, the perforations were not very deep, which meant that peeling it was a bit of a careful exercise to make sure that we weren’t otherwise decimating the package. Then again, I suppose even if you do destroy the package, as long as the drive comes out in one piece, that’s fine?
The warranty information for the drive is printed on the tearaway flap, which claims a three-year warranty for SSDs.
The drive itself feels relatively lightweight and insubstantial – as many value SSDs do. The bottom plate feels to be made of metal, with a top cover made of plastic – another cost-saving measure, no doubt. The labelling on the top of the drive has a nice finish, with the label on the underside providing the expected basic information. The drive claims to be Made in Taiwan.
On the side of the drive, a warranty label is placed across the seam of the casing. Nothing too exciting.
With a drive of this price, I just couldn’t wait to get the covers off – so I went ahead.
The drive itself uses a screw-less design, which also saves money. Removing the bottom label already gives us a hint as to the size of the actual PCB. The two halves of the case separate by defeating the snaps along all sides.
The PCB itself is quite compact, itself labelled with the ADATA brand and serial number. Let’s examine the board more closely.
The top of the PCB reveals a Maxio MAS09024-B2C flash controller dated Week 9 of 2018. This particular flash controller is a special one, as it is one of the few capable of operating QLC memory aside from the Phison controller and shares a commonality with the lower ADATA SU630 SSD which does use QLC. However, unlike the SU630, I believe the SU650 uses TLC memory instead which translates into higher endurance figures of (280TBW instead of 100TBW). The flash controller is paired with two flash packages, remarked with ADATA’s own logo and part number. However, judging from the codes printed near the location mark, these appear to be Intel/Micron flash memory parts. Two independent switching converters phases appear to be populated on this side, one for each flash chip, and more power conversion appears to be handled by a chip marked 3123B 815J1. There is also a four pin header unpopulated (maybe a serial connection for programming the controller) along with a two pin header facing the front of the PCB.
Looking at the bottom, the PCB isn’t very populated, but it seems likely the extra flash memory footprint would be used for the 960Gb model. A polyfuse can be seen to protect the board from overcurrent. Aside from that, the footprints suggest the possibility of supplying power through two switching converters (presumably one for each flash memory package), but neither are populated on this particular board. A number of test points are visible. The board’s silkscreen suggests this is model-specific to the SU650, as the silkscreen reads SU650MB1, this particular board dated Week 34 of 2018.
Being a cost-conscious design, it seems that there are no specific precautions for surprise power removal nor any DRAM cache employed, instead opting for pSLC caching instead.
Testing of the drive was undertaken using the chipset ports of my main workstation (AMD Ryzen 7 1700 @ 3.8Ghz on an Asus X370-PRIME PRO motherboard running Windows 10). I put it through the regular barrage of tests to see how it performed under heavy workloads.
Looking at the SMART data from CrystalDiskInfo reveals that the Maxio controller attributes are not particularly well known at this time, so many of the data values are not labelled correctly. Despite this, it seems that this drive does run hot as soon as any significant workload is placed upon it – due to its low thermal mass due to small PCB/heatsinking arrangements, the temperature rises to about 63-65°C and stays there for the duration of the operation. I’m not sure whether this is illustrative of thermal throttling or not. Nonetheless, the testing proceeded normally and no failure of any attributes was recorded after completing the whole series of tests.
I thought I’d take a few screenshots from CDI during testing to see what happens to the attributes over the course of testing – this way, we can possibly derive the meaning of the attributes without the help of the manufacturer.
In the first set of images on the left, I showed the “new” state SMART data. In the above set, on the left, is the result after one full surface read, and to the right is the result after one full surface write (after the read had been performed).
Finally, the above screenshot shows the numbers for the drive after all tests had completed.
From this, we see that attributes AD, E9, EA, F1 and F2 have changed. It seems likely that the labelling of F1 and F2 are correct, as F2 increased after the first read from 358,559,962 to 1,295,930,563, which is a difference of 937,370,601 suggesting that the unit of the raw values for F1 and F2 is a count of 512-byte sectors. This does suggest that almost 180GB was written to the drive even before I received it, possibly as part of the manufacturing process.
As for attribute AD, it did not change at all on read, but only on write. The initial value is 8,590,196,739 which increases to 12,885,426,182, a difference of 4,295,229,443. This value is approximately 4.58 times higher than the count difference of sectors written to the drive – I wonder if it is in the same units which would imply rather drastic write amplification, or might it be in other units (e.g. 128-byte chunks)?
Attribute E9 and EA both increase slightly during the read – from 626,409,856 to 626,414,464 (+4608) and 230,452,864 to 230,450,816 (+2048) respectively. Could this be due to spurious writes occurring despite the unit not being partitioned, or automatic rewriting from marginal ECC or read-disturb compensation? From here, both increase drastically during the write test, from 626,414,464 to 2,883,172,736 (+2,256,758,272) and 230,450,816 to 1,549,011,200 (+1,318,560,384) respectively. I’m not sure what these attributes mean, but it could be that E9 reflects the writes to pSLC cache and EA represents the writes to the flash memory, although the units are unclear.
Of all the SMART variables, only attributes A9 (Health?), AF (ECC Fail Count?), B4 (MAX PECycle?), C2 (Temperature) and E7 (Lifetime Left?) provide any pre-failure warning.
On a fully written drive, the SU650 was able to achieve an average of 517.9MB/s read speed consistently throughout the whole drive.
Writing to the drive proved interesting – on a fresh drive, high write speeds of about 500MB/s was achieved for the first 96GB of the drive, some of this likely to be the pSLC cache along with the fact that none of the drive’s sectors would have been mapped. A slight slow-down is experienced for another 24GB or so, before the write speed plummets, settling to about 80MB/s until 360GB where it falls a little more to about 60MB/s. The overall average was 185.3MB/s.
Once filled and without secure-erasing, I attempted another write to find that the speeds were averaging just 81.2MB/s as now the drive was operating under a “limited” spare blocks condition and the controller did not seem to handle this as well as some other drives do.
This implies that the true flash memory write speed is rather limited, but also, the controller’s ability to manage high speeds when free blocks are lacking to be rather problematic. This is despite the fact the drive is sold as 480GB user accessible, whereas the onboard flash is likely to be in excess – as some of the flash could have been lost to bad blocks (lower grade memory), be used for pSLC caching and ECC enhancements. Whatever the case may be, it seems operating without TRIM support is probably not advisable unless you manually overprovision (e.g. by creating a partition smaller than the entire drive space and ensuring the remaining free space is never written).
The random read and write tests show the limited IOPS capability of the drive, however, at least access times seem to be low and consistent.
Extra tests were also run, but they are more applicable to hard drives than SSDs.
The file benchmark used a 500MB file and showed consistently smooth high-speed performance, but oddly, a very low write IOPS figure (possibly an issue with HDTune, but also possibly due to the drive’s inconsistent write behaviour). The block size test seems to show accesses of about 64kB are able to make use of the majority of the drives’ throughput.
Freshly formatting the drive and running a test actually puts out rather surprisingly good fitures, with read speeds that very much show the limits of the SATA III interface, and write speeds not far behind. Even the 4kiB accesses look rather healthy – which was a surprise to me. So while the SSD may be low-cost and cost-optimised, the performance for a 1GB test file under CDM is surprisingly good. Sustained performance, however, isn’t likely to reach such lofty heights (as suggested by the HDTune results).
Testing with AS SSD shows just how sensitive the results are to the dirty/clean state of the drive, with the first test done almost immediately after finishing the HDTune Pro tests, not providing sufficient idle time for the drive to manage TRIM-ing the free space. After giving the drive a few minutes to recover, it was able to post the noticeably improved results on the right.
While the unit does push out a score of 831 with sequential scores besting the package claims, testing was performed with the default setting of 1Gb, thus this is reflecting the performance of the pSLC cache more than the bulk of the SSD especially as we are starting with a blank slate.
Moving onto a more traditional benchmark, I decided to run ATTO both with default 256MB test file and 32GB. On the whole, the results didn’t change much, with the drive having reached its full potential by 256kB accesses and being fairly close even down to 32kB. Write IOPS outpaced read IOPS for transactions below 32kB, with the returned speeds showing slightly higher figures than on other benchmarks. This might not be surprising, as the test file still isn’t particularly big compared to the size of the drive.
Running Anvil on the drive results in a score of 4440, which is actually not bad, even besting a Samsung 840 Pro tested on an older platform. However, I suspect this performance is not representative of the sustained ability of the drive, so I decided to try again with some more onerous settings.
Pushing the file size up to 32GB and testing again, a much lower score of 3291 was reported which is more similar to what I might expect from a value SSD, with IOPS figures more in line with the specification data provided in the datasheet. I suspect this indicates that the high peak IOPS is not sustainable and is an artefact of the pSLC caching mechanism.
With the drive formatted as NTFS, it was not possible to test the entire capacity although it did get quite close. On the whole, no data corruption was detected which is the expected result. The achieved write speed was 90.7MB/s and read of 363MB/s. It’s fairly normal to have lower than expected read speeds due to the overhead of verification, but the write speed really suggests that the drive is not suited for write-intensive applications.
ADATA SSD Toolbox
ADATA’s software for diagnosing the health of the SSD, optimising system settings, performing maintenance operations and firmware upgrades is their SSD Toolbox which can be downloaded from their website.
Downloading the latest version from their website and installing it results in a rather cryptic message on first start-up which claims a new version is available and asks whether we should install it or not. The problem is that if you click yes – absolutely nothing happens and the toolbox does not start up. The only way to bypass this and get access to the toolbox’s functionality is just to decline the update altogether.
From the main screen, we can select the drive in question to examine it. Somehow, I suspect the software is not well aligned for use with the Maxio controller of this drive as it somehow claims that 55,222.10Tb has already been written to the drive.
While there is a SMART attributes viewer that does label a few more attributes than CDI does, I don’t trust the labelling as it doesn’t seem to make sense that the Maximum P/E Cycles on a relatively new drive should be 0x123C (or 4668 in decimal).
Arguably the most important feature of the toolbox would be the ability to update the firmware to address any potential bugs or issues. I tried this, and as it turns out, there are no firmware updates at this time.
As a result, I find the ADATA SSD Toolbox to be very limited in its capabilities and not of particular usefulness for this particular drive at this time. Maybe it’s worth installing only when you know there is a new firmware available and that it is necessary for it to be installed – I suspect some of these lower-end drives probably won’t be getting updates anyway.
The ADATA Ultimate SU650 480GB Solid State Drive definitely belongs to the “value” SSD category, fighting it out amongst a number of other candidates but otherwise not being particularly remarkable. Even its own advertising materials claim it to be a great “first SSD”, something I can’t really argue with.
This unit seems to be on a very cost-optimised design, making it as simple as possible while still offering acceptable performance. To its credit, it appears to use TLC memory when some of their competition has already moved to QLC, which should improve its data retention and endurance.
In testing, it showed performance that would mostly best a hard drive, but was not in any way outstanding for an SSD with fairly limited IOPS performance when it comes to write-intensive workloads. The sequential read speeds generally reached close to the SATAIII limits, but the write speeds seem quite dependent on not filling the whole drive, with some clear pSLC caching benefit for up to 96Gb written but turning to something more like 60-80MB/s when the drive is filled for the first time. This suggests that while the drive advertises a 480GB capacity, the amount of overprovisioning is probably not sufficient to run well without TRIM capability as some of this overprovisioning may have gone into providing the pSLC cache and improved ECC.
For everyday users who do not perform large amounts of writing to the drive, the performance of the unit could be very satisfactory. Reliability of the drive and the resilience to unexpected power removal was not assessed.