Testing the top-of-the-range SATA SSD Samsung 870 Evo 500 GB

Having tested the Kingston KC600 in the spring, we considered closing the section of “decent” SATA SSDs, since almost all known models had already been studied and little remained on the shelves. The main segment of this device is dominated by more affordable SSDs, since most buyers prefer to save money. Such drives are most often used to improve the performance of old laptops and desktop PCs where other types of drives cannot be installed. However, higher value usually does not play a decisive role.

There is a demand for low capacity SSDs such as 120GB, although such «decent» models have not been produced for several years. We think there's no point in considering anything under 250GB due to the almost identical prices. In such cases, you should prefer a half-terabyte SSD to avoid disappointment in the future.

To test such products, it is necessary to take into account the preferences of real customers and the current assortment of stores. Many buyers do not understand the details of the technical characteristics and choose at random. However, there are those who are more careful about their purchase, especially when choosing a 500 GB SSD.

It is interesting to note that among the 500 GB SSDs, the most popular was the relatively expensive Samsung 870 Evo, even surpassing the Kingston A400. The price difference between them was only about 1000 rubles, which could be a key factor for more knowledgeable buyers. However, the situation in stores may have changed, and they may now occupy different positions in the popularity rankings.

We have not tested this model before. Initially, time was short — two years ago this model was only a minor update of the well-known 860 Evo. At the same time, SSDs with the PCIe Gen4 interface appeared on the market en masse, which represented a more interesting object for testing, and that’s exactly what we were doing. At that time there were many normal SATA models. However, a year and a half ago, their number sharply decreased, and in the case of the Samsung 870 Evo, a small scandal arose from the first batches. Because of this, we postponed the issue until we fully clarified the circumstances. Now, it seems, the situation has returned to normal, and there seem to be no other top-end SATA drives anymore.

Kingston recently released its DC600M data center SSD, which is essentially an updated version of the DC500M with the same Phison S12 controller, but with new memory (apparently Micron B47R). However, these SSDs are expensive, and the choice is mainly limited to models based on the Silicon Motion SM2259 controller. This controller is generally considered successful, although it is not a leader compared to the Samsung platform.

Therefore, it is still worth testing the 870 Evo to make sure that the information about it is complete. Regarding the events of last year, they also deserve special attention.

Subconsciously, we have always been waiting for a similar story regarding Samsung products in recent years. Many people praised the company's products, and usually such popularity ends with some kind of incident. However, Samsung products themselves, including the 850, 860, 950 and 960 models, did not cause any complaints. There is only an unfortunate story with the 840 Evo, but that was a long time ago when good results were not expected from TLC memory. Moreover, it quickly became a thing of the past.

However, even if the products weren't outstanding, they were quality. The main thing is that Samsung did not sell budget models on the retail market, which affected its reputation. If a company maintains a brand at the level of top solutions, but does not sell budget solutions, then the level of trust in the brand automatically increases. And if products are sometimes released that are superior to competitors' top models, this only strengthens the brand's position.

You don't need to invent anything special to qualify as a market benchmark. For example, the 850 Pro has withstood all market changes while competitors have reached the level of the 850 Evo or below. Samsung later released the 860 Pro, but at that time the SATA segment was dying out. This trend continues in the NVMe direction: one product is at the level of the top competitors and another one is higher. There are no budget models on the market that could compete directly with Samsung.

Samsung regularly made changes to its products and no one objected to these updates. For example, the 850 Evo started with 32-layer memory, then moved to 48-layer and finally 64-layer. These changes were noticeable, but did not cause dissatisfaction. The model number change in the 870 Evo is due to the transition to a new controller and improved warranty conditions.

After mass testing in early 2021, the 870 Evo was forgotten. Compared to the 860 Evo, their speed characteristics remained virtually unchanged, and the warranty conditions did not change. Buyers switched from the 860 Evo to the 870 Evo, glad to have at least one high-end SATA line on the market. Yes, they are a little expensive, but they have an impeccable reputation.

True thunder was heard closer to the fall of 2021, when reports began to appear on many forums about the occurrence of bad blocks and data loss on SSDs of the 870 Evo line. Initially, these reports were met with disbelief, but their number constantly increased. Samsung, without providing any explanation, began quietly replacing the SSD under warranty. The new firmware was released only at the end of November, with a strong recommendation to update. Various conspiracy theories began after this, but they gradually died down by mid-2022, along with a decrease in the number of new reports of problems.

What really happened? Since no clear information has been received from the manufacturer, we can only speculate. One or more problematic flash memory batches have probably been detected. After all, storing data in flash memory has a lifespan, and manufacturing problems could have significantly shortened it. However, this does not mean that there are 100% defects or global problems with Samsung’s 128-layer TLC memory as a whole. For example, the 250 GB SSD 870 Evo or 980 Pro remained virtually flawless, while the 870 Evo with 256 Gb die, rather than 512 Gb, encountered problems. It was the latter who had slightly more problems than usual. Also keep in mind that it took several months for these problems to be discovered, and by that time the 870 Evo revision had already been changed. Many users got scared and tried to avoid this line altogether.

Why don't we suspect the firmware? First, the symptoms do not indicate problems that would be related only to the capacity. In addition, updating the firmware did not help those who were already experiencing problems — their level continued to increase. For those who have not yet had problems, they may simply have not been noticed, but no one decided to conduct tests. Problems with the production of any chips and some variation in quality even in one batch are well known. Perhaps in some downloads or even in a specific batch the quality turned out to be slightly worse than standard. This specifically affected the 512 Gbit crystals, so the main victims are the 870 Evo, but not the 250 GB. Considering that each SSD contains at least eight memory crystals, even one defective crystal makes the entire device poor quality. However, relative to the calculus, there were few such “low-quality” crystals, so the problems turned out to be localized in time.

Could this situation happen again in the future? Yes, this is true for all manufacturers. The scale may vary, but the essence remains the same. At the beginning of 2022, Western Digital and Kioxia had to dispose of a considerable amount of products, as a defect was discovered right at the factory. However, the time period for detecting such problems also varies. There is no such thing as a reliable storage device, and reliable data storage is a process that requires proper organization and ongoing maintenance, rather than hoping to find a silver bullet that can solve all problems.

When it comes to flash memory, a certain percentage of defects will always exist, as the quality of the dies that fail testing varies. Therefore, they are graded, and it is assumed that the highest grade will not cause problems, even when used in critical areas. However, consumer SSDs have lower memory quality requirements than server ones. Therefore, even expensive consumer models do not use the highest quality category. This scheme works fine, but sometimes causes failures, including massive ones, when the percentage of defects goes beyond the typical 1%. You need to learn to just exist with it without feeling afraid.

However, some may still have concerns about this line of SSDs, despite the lack of new problem reports, and given that the drives currently sold are from distant revisions of the 2021 series (at the time of the problems it was no longer possible to purchase the «problematic» revision, and now at least the third modification is on sale). We will not convince anyone — after all, the health of the nervous system is irreparable. Our goal was, firstly, to finally test this model, since there are too few decent SATA SSDs left on the market, and secondly, since there are no more drives at the level of the 870 Evo at all.

What do we mean? A powerful eight-channel controller with two cores, rather than a four-channel single-core controller like the Silicon Motion SM2259. In addition, the presence of a DRAM buffer of a traditional size, and not 512 MB for all modifications, as in the third version of the Crucial MX500. The presence of fast TLC memory is an element that was once perceived as standard, but is now rarely seen anywhere. Perhaps only the Kingston conditional enterprise models on the Phison S12, but they deactivate the SLC cache, which affects performance. Additionally, models like Kingston's DC450R are priced even higher than the Samsung 870 Evo. And in some cases, the KC600 model costs more.

Thus, there is every reason to buy. But, remembering the events of the end of 2021, it’s a little scary. Although from our point of view there is no reason for concern, we will not delve into this topic. We can just see how this device works and whether we can actually evaluate the theoretical benefits. This is why we must do this.

To make the analysis more diverse, we added to the testing list the WD Red SA500, similar to the 2020 Blue 3D model — it is based on the Marvell 88SS1074 controller and uses BiCS4 memory with 512 Gbit crystals. Even though the SA500 is available on the market at quite high prices, and models of this type have not appeared for a long time, limiting the choice only between the Crucial MX500 and Kingston KC600 seems a narrow option. However, as already mentioned, there are simply no more options for analysis. In some ways, the new WD models are interesting, but their level is obviously lower due to the use of bufferless simplified controllers. As we recently saw, the Transcend SSD230S can no longer compete with the Crucial MX500 due to the too slow memory installed there. It’s difficult to say exactly about the characteristics of the Adata SU800. Once again, this is the same Silicon Motion SM2259, which will not provide significant advantages over the Crucial MX500 and Kingston KC600, even with luck. This segment of the SATA market is becoming increasingly boring and dull — it is dominated by cheap but very slow models. This is why we no longer recommend using SATA in all cases where possible. However, sometimes you still have to turn to SATA. But you don't always want to buy something that is supposedly slow, even if it is supposed to be sufficient. In such cases, the savings may not seem so justified, especially if you are limited to models with “popular” containers, as is the case in this test.

Let us recall that in 2020 the “upgrade” included an increase in the capacity of memory crystals, which led to a decrease in the interleaving ratio in the 500 GB and 1 TB models. Buyers of the latest model did not notice any special changes, but owners of the first model noticed significant improvements: previous Blue modifications, with such capacity, “faced” interface limitations, while the new ones (including the Red one that joined them) no longer experienced this problem. However, this is not surprising — in those years, manufacturers focused on reducing the cost of SATA drives, even at the expense of performance degradation, since most buyers were already more interested in price.

On the other hand, Micron B47R is also 512 Gbit, but the memory is even too fast for any SATA controllers. Therefore, the Kingston KC600 shows the same graph that we had become accustomed to before 2020. Against the general background of the state of affairs in this segment, this is a great achievement.

But the Crucial MX500 cannot repeat it exactly. There are clearly some flaws in the firmware — otherwise there is nothing to explain such a failure immediately after filling the SLC cache. The Kingston KC600 manages somehow without it. Will this make a difference in practice? For any exits outside the cache — mandatory. But you may not have time to notice that the MX500 then restores speed — the data will run out.

As for our protagonist, it is clear that 512 Gbit memory crystals turned out to be an unfavorable factor. Even taking into account the fact that Samsung 128-layer flash memory has long ceased to be the leader in speed. For example, the Samsung 850 Evo faced interface limitations even with a capacity of 500 GB, but thanks to the use of 256 Gbit memory crystals, of which 16 were required. Thus, we had double interleaving on eight controller channels. However, starting with the 860 Evo, striping disappeared for half-terabyte drives, and full speeds were only available from 1TB models. This had already become a common problem by the time the Samsung 870 Evo arrived in early 2021. However, some available models have managed to improve their performance slightly. However, this is only one of many possible scenarios, so it is better to postpone final conclusions for now. In addition, it is important to note that some “second generation” models from WD, which had similar problems, were able to noticeably improve their performance.

Low-level benchmarks in general and CrystalDiskMark 8.0.1 in particular have long been victims of testing limitations caused by SLC caching. They are limited only by the cache itself and are not capable of conducting broader tests. However, the information provided by manufacturers about device performance is usually limited by cache limits, so testing devices on real-world tasks is always useful. This is especially important as caching work is being done with the goal of maximizing cache utilization in real-world scenarios and delivering high speeds while reducing memory costs.

Consistently reading or writing data is not difficult these days, especially when SLC caching helps the latter. Therefore, at least in this program, a “decent” SATA drive must rest against the interface — which is actually what all four do. Now, if you read and write at the same time, then there is already a difference between the controllers. And what is typical is that Samsung is the worst, losing even to some bufferless solutions. And the best one is Marvell from 10 years ago in a not too fast environment. Well — and this happens, perhaps. But we wouldn't be surprised if it characterizes the benchmark more than its test subjects.

Because what’s in the minds of low-level utilities—their creator alone knows. For example, here Silicon Motion SM2259 looks very good, although at the time of its birth it was considered the worst (but at least not one of the worst).

And here everyone is approximately equal — except for the obvious outsider from Marvell. However, remembering the SLC caching settings in the WD SSD, it’s possible that it’s not his fault—it’s just that there wasn’t enough cache. For the rest, it turned out to be enough, so they showed the maximum for SATA in such scenarios.

The impact of such operations on the speed of real software is much greater than it might seem. In real conditions, «long» queues are indeed much less common, but data blocks other than 4K bytes are very common. Although the number of operations per second on «large» blocks may be slightly lower, the size of these blocks is usually larger, resulting in a higher resulting data transfer rate in megabytes per second.

That is why many try to configure the operation of their systems and programs to use precisely such data blocks. This allows you to optimize drive processes for these use cases. In this case, Silicon Motion demonstrated good performance on data blocks larger than 4K, and the Samsung 870 Evo becomes the leader in data transfer speeds on any blocks larger than this size.

For recording, this is not entirely true, but in practice it is of less importance. Moreover, the 870 Evo looks good on the smallest blocks.

Mixed mode is also important — after all, in reality (and not in test utilities) it rarely happens that for a long time data has to be only written or only read. Especially in a multitasking environment — and given the rich inner life of modern operating systems. But the impressions from the results are twofold. However, the slight loss of the Samsung 870 Evo can be attributed to the high “responsiveness” of the 176-layer Micron memory, which a more powerful controller cannot beat, but when the block is enlarged, the fact that it is more powerful begins to have an effect.

Despite impressive results in benchmarks, such high data transfer rates are not always achievable in practice. Working with real applications is much more difficult than running tests in utilities like CrystalDiskMark. This type of testing operates on small pieces of data inside one file.

Almost always, when conducting such tests, the files are in the SLC cache, which provides higher speeds. In addition, the actual write operation of one file also includes file system service operations. For example, this could be modification of the Master File Table (MFT) and file system journals (since most file systems in use, such as NTFS, support journaling). During actual recording, data is not written sequentially to one place, but partially to different places, sometimes in small blocks.

That is why greater practical accuracy in assessing drive performance comes from using tools such as the Intel NAS Performance Toolkit. This tool allows you to test not only the cache, but also the operation of the drive under conditions that are closer to real, for example, when the device is almost completely full. This allows us to get more realistic results of the drive's performance under real-world usage conditions, which is what we always do.

Working in one thread is the most common (146% of cases), but also the most difficult scenario. True, it is complicated for top models — equipped with a completely different interface. But running into (or almost running into) the limitations of SATA is not a tricky matter. Everyone has been able to do this for a long time. The truth is a little different — and the leader here turns out to be WD Red, from which no one expected records. However, the spread of results itself is negligibly small.

And here Red lost to everyone, although it’s completely ridiculous to take this difference seriously. Perhaps if we were testing terabytes, there wouldn’t be any left, and everyone would cheerfully run into the conditional “560”. One of the reasons why they stopped chasing speed indicators in this segment a long time ago — when the ceiling is so low, it makes no difference who hits it how hard.

With the recording, everything happens now that we were already prepared for. Indeed, a combination of SM2259 and B47R could do without SLC caching. The memory itself turns out to be too fast, which this controller cannot manage “normally” (all its advantages are revealed in completely different segments), but what is there is enough. “Five hundred” on other platforms is objectively slower. If we were testing drives with a full terabyte of capacity, the situation would be slightly different. But neither WD nor Samsung can quickly write outside the cache. The first one ends up being the slowest, since the cache is very small. The second one is faster. But not enough to keep up with the latest versions of the Kingston KC600 or Crucial MX500.

The situation does not change fundamentally, since the mechanism of problems arising is similar. If, of course, we consider them problems, let us recall that today we are testing the fastest surviving SATA drives. One of them was not preserved at all — and was added only to enliven the picture. Its “replacement” is much slower, and we’ve already seen what’s happening in the budget segment more than once. Against this background, everyone is beautiful. But, of course, you can look for some minor differences.

Repetition of what has been covered. One pair of devices turns out to be less dependent on SLC caching, the second — more so. In global terms, this no longer means anything, but for increasing general education it is interesting. As has been said more than once, you should not focus too much on the characteristics of the controllers — memory also greatly affects the final result. Most often limiting the controller's abilities. And sometimes turning out to be excessive for him — and stretching the speed to a higher level than could be expected. On the other hand, if Samsung had kept the same approach in the 860/870 Evo lines as in the 850 Evo, that is, using 256 Gbit crystals for 500 GB modifications, there would have been a different result. But the desire to save money is also inherent in Samsung. The only difference is that the company usually does not go overboard with savings. And for Crucial and Kingston, the bulk of shipments are made up of BX500 and A400 with all that that implies.

What the Samsung 870 Evo does well is to smash to smithereens its roughly the same age in the WD version. It’s very possible that the same thing would have happened with the then versions of the MX500 and KC600 — it’s just that no one expected a trick from the new Micron memory. And also the fact that it will be so cheap that it will be used even in budgets, not to mention mid-range ones. And, we repeat, increasing the capacity could change the picture, since here it would be difficult not to run into the interface. But at the target level — so. Which we didn't really expect, generally speaking.

Today, one of the best comprehensive benchmarks for assessing storage performance is PCMark 10 Storage. This set of tests includes several different tests, of which the «Full System Drive» test is the most informative and useful. It covers a wide range of real-world use cases, including booting the operating system and copying data (both internal and «external»).

Other tests included in PCMark 10 Storage are only partial variations of «Full System Drive» and, in our opinion, are less informative. However, this comprehensive benchmark is useful because it provides accurate measurements of both throughput and latency for practical tasks. While averaging the results across different scenarios and bringing them to a total number is a bit of an oversimplification, it provides a more realistic assessment of overall performance. This is currently one of the best ways to get an overview of the drive's performance in various practical use cases.

After all the previous tests, the results seem a little unexpected at first glance, but upon closer examination, they are quite expected. For example, the Samsung 870 Evo SSD showed itself to be a leader in read operations with arbitrary addresses and blocks from 16K. This type of operation constitutes a significant portion of the application and operating system startup load, and is one of the main testing scenarios. Its good performance with mixed read and write operations is also noticeable. Although the 870 Evo lagged slightly behind its main competitors in pure recording, it performed better in mixed operations. It is important to note that this was not a simple simulation of artificial conditions with a uniform queue and blocks, but realistic mixed scenarios with small pauses that are closer to real-life operation.

Low-level benchmarks are focused on studying narrow aspects of drive operation and do not always take into account the complexities of real-life scenarios. PCMark 10, on the other hand, provides a comprehensive assessment that is close to real-world usage conditions. Although testing occurs in fast playback mode, it covers a wide range of scenarios and can take a significant amount of time. Unfortunately, this is the only such benchmark on the market, and its results cannot be cross-checked in other tools.

The results obtained, admittedly, were somewhat unexpected. Although the Samsung 870 Evo SSD set a new record in the PCMark 10 Storage Full System Drive test, its results are not entirely up to expectations. Based on the internal design and expected features, the Samsung 870 Evo was expected to be the undisputed leader in all areas where it is possible. Perhaps when testing a higher capacity SSD it would have shown itself in a different light. For this comparison, however, we chose the 500GB model, and here the latest versions of the Kingston KC600 and Crucial MX500 stand out with their performance.

Despite this, there is little reason for disappointment. Each SSD has its pros and cons, and choosing between two nearly identical models is no real choice. For example, the Crucial MX500 stands out as having the lowest price of the three but only has a one-year warranty, while the Kingston KC600 is more expensive but offers a full five years of warranty. The Samsung 870 Evo in the market also comes with a one-year warranty, but the brand has its own fan base and loyalty despite some previous issues.

It's difficult to draw definitive conclusions based only on past problems — they won't necessarily arise again. However, the absence of problems in the past does not mean their absence in the future, which is always important to remember, regardless of the manufacturer. And regarding the choice of SATA SSD, besides these three models, there is practically nothing on the market. If you need something really “decent”, but not at the price of a spaceship, then this choice is limited. Some brands, such as Kingston, have updated their DC series, but only in expensive versions, without offering analogues in the more affordable segment. So the choice is really limited at the moment.