• Home
  • Blog
  • Testing budget SSDs Adata Ultimate SU630 480 GB and Patriot P220 512 GB

Testing budget SSDs Adata Ultimate SU630 480 GB and Patriot P220 512 GB

17.01.2024 09:08

Today's material concludes our testing of SATA drives with a capacity of half a terabyte. This will probably also be the last article dedicated to SSDs with this interface. This does not mean that they have lost their relevance in general, but over time, older equipment that is incompatible with more modern solutions is gradually being phased out. However, they still need repairs or upgrades, and it is for such cases that users are looking for more affordable SSDs.

Old computers continue to be used because their owners do not see the point or cannot switch to new devices. That is why when choosing SSDs they focus on budget options. Manufacturers, in turn, strive to make these drives as affordable as possible, often using old budget controllers, cheap QLC memory and other cost-effective solutions.

There are no stable configurations in this segment, and it is important for buyers to look at reviews and ratings in stores. It is important to note that in this segment it is difficult for manufacturers to offer significant improvements, and therefore changes are usually small. However, recently, even SSDs with a capacity of 1-2 TB with TLC memory are becoming rare, and QLC memory is already used even in models with a capacity of 250 GB.

Exceptions include the Crucial MX500, Kingston KC600 and Samsung 870 Evo, which still rise above the rest in this segment. However, they are not new. Despite the small number of alternatives, the choice of budget SSDs is quite limited.

Adata Ultimate SU630 480 GB

The line was introduced to the market five years ago, and from the very beginning Adata honestly announced that it would use QLC memory. This caused some interest, especially given the announcement of 240 GB SSDs, as doubts were raised about their reliable operation. Some users expressed concerns. However, most already believed that such configurations would not work reliably. In light of this, many preferred to pay attention to models such as the SU650, which, as a rule, used more reliable TLC memory. Despite the use of the Realtek RL6468 controller with its features, this proposal seemed more promising than models with explicitly “four-bit” memory, such as the SU630 and SU635.

However, this situation was typical only three to five years ago, when the SATA market was just beginning to decline. Since then, the «Ultimate» line has retained only its beautiful name (which rather sounds like a mockery), and there is only one worthy model left in it — the Ultimate SU800, which we studied in detail when testing the Transcend SSD230S. Below appeared a chaos of various budget platforms, changing chaotically and regardless of the initial statements. For example, in the same stores you could find SU750 on QLC and SU630 on TLC. Which one? Depends on luck. Controller — almost any from Maxiotek, Realtek or Silicon Motion. However, always bufferless and two- or four-channel. Thus, “ultimates” have long become a reflection of the entire segment as a whole — there is one decent and more or less predictable solution, as well as a lot of uncertainty, which, to a first approximation, differ only in labels. Prices are just as confusing, so if you want to play these games, you can just get the cheapest drive — it won't hurt so much if you're unlucky, but doubly nice if you're lucky.

What ticket did we get? Dual-channel Maxio MAS1102 controller and 144-layer Intel N38A QLC memory. In general, we were counting on something like this, so it did not come as a surprise to us. However, this may come as a surprise to some buyers. Let's take a look below at what awaits them, but for now let's give an idea of the second hero.

Patriot P220 512 GB

The vendor, which proudly points to its California headquarters, was once one of Phison's most trusted partners. If you need a budget SSD and for some reason prefer a controller from this manufacturer, then Patriot is an excellent choice. If we are talking about a budget SATA with a Phison S11 controller (which still has some advantages over similar controllers), then it is logical to take Patriot Burst. However, nothing was ever said specifically about the type of memory used, or at least its type, with the exception, perhaps, of SSDs with a capacity of 240 GB or more, introduced in 2020 — which then seemed a little provocative, but is now the norm.

But all this already belongs to the past, since about a year later the concept changed: if we talk about Patriot, we now mean HikVision. At first, production capacity was focused mainly on SATA products, then some originality disappeared in the PCIe segment. Patriot's top products currently include the surviving InnoGrit IG5236-based Viper VPR4300 and the newer Maxio MAP1602-based Viper VP4300 Lite with 232-layer YMTC TLC memory. Although this is generally not bad, it is not at all original — both are presented, for example, in the Digma or Netac assortment.

If we return to SATA… SATA is still SATA. The similarity with HikVision is evident in everything here — from configurations to chip markings and lines in the firmware. For example, our P220 identifies itself as “HKSSD”, although it is common for drives on Maxiotek controllers (including parts produced by HikVision even for itself) to be “MKSSD”. The controller is, as you might guess, Maxio MAS1102. As for memory, we have not seen this one before — 128-layer QLC flash YMTC. A special feature is the 1.33Tbit dies, although most manufacturers prefer to use 1Tbit QLC dies. However, this increase in volume can be useful for flexible configuration of plant operation. For example, something that is not capable of functioning normally as a QLC can be easily converted into terabit TLC crystals. But for its intended use, deviation from powers of two is not very convenient. The tested SSD has three memory crystals — enough for 512 GB. Terabyte models are usually equipped with six crystals, which leads to double interleaving on some controller channels and no interleaving at all on others. In both cases, it is necessary to provide load balancing in the firmware. The struggle for low cost often turns out to be more difficult than the development of top-end devices, but in this segment everyone has low speeds, and there is no need to complicate things too much. Now let's see how this scheme works in practice.

Testing

Samples for comparison

When we recently tested the Crucial BX500, we did a comparison including the KingSpec MT-512 on the Maxio MAS1102 bufferless dual-channel controller. Both drives used Micron N48R QLC memory, which is also common on the Crucial BX500. We will look at the same pair today, since the MT-512 is ideal for comparison with the two main devices in our testing (same controller, but different memory). The Crucial BX500 will also be used to benchmark against devices running on other budget platforms.

Filling with data

The maximum cache write speed is about 500 MB/s, but it is achieved in a small area (exclusively in the CrystalDiskMark test). Not all cells participate in caching, but the cache size exceeds 20% of free space, which is considered acceptable. Despite the fact that the memory write speed cannot be called high, it turned out to be unexpectedly acceptable for QLC memory. To record large amounts of information, a speed of 60 MB/s is, of course, insufficient. The whole process took a little over an hour and a half. Compared to, for example, the Crucial MX500 of the same capacity, which copes with this task in 20 minutes, the difference is noticeable. However, further comments seem unnecessary, but...

The inner pessimist believes that things won’t get worse, while the optimist happily claims that it will, and how it will happen. This situation is similar to the previous case, but the SLC cache here is smaller and slower (except for a small static area — as usual), and outside of it the write speed does not exceed 50 MB/s and often even drops to 20 MB/s. In the case of the SU630, recall, the speed was 60 MB/s and sometimes dropped to 40 MB/s. The overall result is two hours and forty minutes to record the same miserable half a terabyte.

At the same time, I can’t say that the memory from YMTC is so bad — since we’ve been climbing with Micron for well over three hours. And despite an even greater reduction in the cache, we don’t even go above 30 MB/s.

With the change of controller, the situation improves slightly, but not enough to make a significant difference — the overall test execution time returns to the Patriot P220. Just.

It's safe to say that Intel engineers did not put their efforts into improving their QLC memory in vain: it can indeed be rewritten relatively quickly. In older consumer SSDs, even from Intel itself (now Solidigm), you can reach several hundred megabytes per second, and server models are capable of operating gigabytes per second. In both cases, this does not take into account SLC caching, since server SSDs do not even use it. However, to achieve such results, a more or less normal controller is required, which cannot be said about budget SATA drives. So maybe they'll run a little faster or a little slower, but they'll always remain pretty slow. These models demonstrate normal write speed only within the SLC cache. If this space is not enough (or there is none at all), then the recording speed is reduced to low values. In such cases, it is better to refrain from recording altogether :)

Maximum speed characteristics

Low-level benchmarks in general, and CrystalDiskMark 8.0.1 in particular, have long been limited to SLC caching, and are essentially only capable of testing this cache. However, the information manufacturers provide about device performance is also limited by cache limits, so it's always a good idea to check how devices perform in real-world conditions. This check is important because all caching work is done with the goal of achieving as many cache hits as possible and maintaining high speeds despite decreasing memory costs.

Even with full use of the SLC cache, low-end budget SSDs are not able to fully exploit the mixed bandwidth of the SATA interface. For the same financial costs, but using other interfaces, you can get much more productive solutions. In these segments, controller technologies are constantly improving and developing, while in budget SATA SSDs there is a lack of innovation, since their implementation is not justified from an economic point of view. When at least two consecutive multidirectional flows are activated, the total speed decreases by another one and a half to two times, as a result of the actions of simple controllers and cheap memory.

The Patriot P220 turns out to be a clear outsider in our testing. This is due to the use of three memory dies on two controller channels, which creates a performance bottleneck. Even with an indicator of 4.5 kIOPS in the case of working with requests in everyday life, which, however, exceeds system requests, given that they usually do not exceed 2-3 kIOPS, the device cannot effectively use request queues. This makes it not the most suitable for installation in a computer designed to perform serious tasks. The same applies to other test participants in full.

Write speed turns out to be less critical in practice, and achieving better results is quite simple, at least when using an SLC cache. However, two SSDs based on Micron N48R memory show worse results in this regard. Despite the different controllers, both devices face the same problems. Adata retains its leadership using Intel memory, which allows us to repeat the previously expressed compliments to the engineers of this company.

It's funny, but at the next stage of this Paralympics (unlike the previous ones, which is very relevant from a practical point of view), MAS1102 comes forward in tandem with N48R. On the other hand, all the participants are traditionally quite dull — but at least there is some variety.

With a new leader, the gap becomes even greater. On the one hand, this emphasizes that the controller, as usual, makes the main contribution to the results. The MAS1102 simply handles these scenarios better than the SM2259XT2. But it is too early to draw far-reaching conclusions from this.

Mixed mode is also important because in real life it is rarely the case that data only needs to be written or only read. This is especially true in a multitasking environment and given the variety of operations of modern operating systems. However, there is nothing new here — everything is predictable. Even in the ideal case, when the entire process is completely within the SLC cache (and this often also increases read speed), minimum-level SSDs cannot impress with their results. So all that can be said positively is that they are doing their job. However, low-level benchmarks are only limited by their ability to outperform hard drives. Nothing more.

Working with large files

Despite the impressive results in low-level benchmarks, it is not always possible to achieve such speeds in practice. This is because real-life use cases are always more complex. For example, CrystalDiskMark operates on small pieces of information within a single file. Firstly, in modern conditions this file is often and always guaranteed to be in the SLC cache during testing. Second, the test does not take file system overhead operations such as MFT modification and logs into account. Real file recording includes not only a sequential operation, but also many other, different and small ones. In this context, the Intel NAS Performance Toolkit is of great practical importance. This tool allows you to test the device not only in terms of cache, but also in conditions that are closer to reality, for example, when there is almost no free space left. These are the tests we usually conduct.

Existing SSDs have long exceeded speeds of 500 MB/s, while these models do not reach this figure. However, significant differences between them cannot be distinguished either. This is especially important from the point of view of the owner of an old laptop, where the only alternative may be a regular hard drive, which, even in a single thread and when reading data sequentially, is three to four times slower.

It is worth paying attention to the fact that when switching to multi-threaded mode, the Crucial BX500's performance even decreases, while the other three models increase it, approaching the interface limitations. The difference is explained primarily by the choice of controller. MAS1102 can be considered a more preferable solution compared to Silicon Motion SM2259XT2, although the choice of controller itself when using QLC memory can hardly be considered truly optimal.

The Crucial BX500 also shows lag during the recording process, even under ideal conditions when the size of the SLC cache is maximum and there is always free space in it. In reality, such situations will not occur so often, especially in the budget segment, where disks are often purchased with a minimum supply. In these conditions, as expected, the Adata SU630 comes out ahead, where the write speed of the memory itself is lower than other models, but not as low.

Since the main problem is related to the characteristics of the memory itself, the changes in results are not that significant. If these characteristics can be masked by SLC caching, then all SSDs become fast, their performance is comparable to reading. If this is not possible, then all devices turn out to be slow. However, the Adata SU630 can at least compete with the same laptop hard drives, which may be an important factor for certain categories of users.

Naturally, here too it is the writing speed that is critical, not the reading speed. And the funny thing again is that both losers have different controllers, but the same memory. Which should potentially be fast (after all, Micron’s similar TLC flash turned out to be excellent), but it didn’t work out.

For hard drives, sequential access is key, while for SSDs it has long been less important which access scenario is used. However, in this case, we see that the Crucial BX500 reduces its speed and is noticeably inferior to other devices, whose results have even increased. This fact is probably worth noting primarily thanks to Silicon Motion, which perhaps could have taken a more responsible approach to the development of the SM2259XT2 controller. While on the other hand, it is worth noting that in this segment, attempts to sort by quality and speed can be thankless.

Comprehensive performance

Currently, PCMark 10 Storage is the best comprehensive benchmark for assessing storage performance. While briefly describing this tool in our review, we noted that not all three tests included in the kit are equally useful. The most valuable is the use of the full Full System Drive test, which includes almost all common scenarios, from loading the operating system to copying data (both internal and “external”). The other two tests are subsets of this full test and are, in our opinion, less «interesting». However, this full test is useful because it provides accurate measurements of not only real-world throughput for practical problems, but also the resulting latency. Although averaging these metrics across scenarios and then reducing them to a single number is a bit synthetic, it is currently one of the best ways to obtain an overall, realistic assessment of drive performance.

It is important not to get hung up on ideal scenarios for a conditionally empty drive. Despite the fact that in this case 50 GB are occupied, which represents the operating system and a small number of applications, with such requests, few choose a budget SSD of half a terabyte, since this is often more expensive than a “quarter” or “octagon”. Thus, it is better to focus on a scenario where there are only 100 free gigabytes available, which is a more optimistic estimate.

In this case, SLC caching may not cope with setting up a smoke screen, and it is clear who is capable of what. Formally, the Adata Ultimate SU630 turns out to be the best, and the Crucial BX500 the worst, but in practice this is more interesting for education. The problem is that all of these drives are slow. Switching from a hard drive will on average be an improvement, but in some scenarios even a home user may notice irritation. The best models provide stable performance without sudden “brakes” from scratch, which cannot be said about lower-quality SSDs.

Total

We have reached a certain point in our market research, especially in the context of SATA drives. In our purchase, we were lucky: out of ten devices, only three used QLC memory, and, fortunately, these were devices that could be counted on. Having excluded from the list the Samsung 870 Evo and Transcend SSD230S, which used non-QLC memory, we were faced with the reality: the risk of encountering QLC drives increases every day, and this trend seems to continue in the SATA segment.

Buyers don't have many options left. One of them is to avoid SATA drives entirely, especially if you have a newer computer that supports NVMe drives. Another option is to choose one of the remaining «decent» SATA drives, but this list has gotten smaller and will probably continue to get smaller. However, such models are usually more expensive and less popular.

The last option is to take a chance on this lottery, where you may have a little more or a little less luck. The difference, however, will be small. Even SSDs on modern SATA controllers with TLC memory do not promise stars from the sky, and QLC, unfortunately, makes the situation even worse. However, within certain limitations they can still provide acceptable performance, although problems may appear over time. The best place to start looking for better solutions is in other segments.