Flash drives built on four-bit QLC 3D NAND memory are gradually becoming more widespread. And although such memory is still distrustful of many users, sooner or later it will become the predominant option used in mainstream solid state drives. And this will happen, obviously, for the most banal reason: due to the higher storage density of information, QLC 3D NAND further reduces the cost of solid-state drives, making it possible in the near future to make available to the mass buyer such SSD models, the capacity of which is measured not in gigabytes, but in terabytes.
We already had the opportunity to get acquainted with one of the first SSDs based on QLC 3D NAND, the Samsung 860 QVO SATA drive. However, Samsung is not the only manufacturer that has started mass production of 4-bit memory and solutions based on it. The Intel / Micron alliance also has its own QLC 3D NAND, and these manufacturers also use it in mass products for the consumer market. And they started doing it even before Samsung.
But it is necessary to pay attention to the QLC products of Intel and Micron not even because they brought this technology to the mass market first. The fact is that these manufacturers began to solve somewhat different tasks with the help of QLC 3D NAND, which is why their QLC products differ from Samsung’s offerings ideologically. While the Samsung 860 QVO is a simple consumer SATA SSD that the South Korean manufacturer has tried to keep the price to a minimum, Intel and Micron’s strategy has been more inventive. They started introducing QLC 3D NAND from NVMe drives, that is, not from the lowest, but from a higher price category. This approach is good in that it allows you to demonstrate the potential of four-bit memory in a bright and understandable example, confirming that QLC 3D NAND is not necessarily an option for low-cost and slow drives with a reduced resource. Such memory can also be suitable for products of a higher rank, which are precisely the new QLC-new Intel SSD 660p and Crucial P1.
Along the way, Intel SSD 660p and Crucial P1 solve another problem. With their help, manufacturers want to declare their ambitions in the low-cost NVMe SSD segment, which is currently on the rise. Recently, we have seen many attempts to bring the cost of high-speed drives with an NVMe interface closer to those of the usual SATA SSDs. But all of them were associated either with the transition to a bufferless design, or with the use of controllers with reduced functionality, which ultimately had a bad effect on performance. QLC 3D NAND, on the other hand, allows you to make NVMe SSD cheaper in another way, avoiding obvious design flaws on the controller side.
As a result, in the face of Intel SSD 660p and Crucial P1, we have options that are very interesting in terms of price and performance. Among NVMe drives, they have almost the lowest price, which can be compared with the cost of popular SATA models, but at the same time they promise quite good speed characteristics that significantly exceed the specifications of SATA drives. It turns out that Intel SSD 660p and Crucial P1 are really asking for the role of the best alternative for aging SATA SSDs.
However, does everything work out so smoothly in practice — or does QLC 3D NAND still leave an indelible imprint on the performance and other consumer characteristics of NVMe SSDs? This review will be devoted to the answer to this question, the main character of which will be the Intel SSD 660p.
The first thing that sparks interest in the Intel SSD 660p? — This is Intel’s proprietary QLC 3D NAND. The fact is that such a memory is somewhat different from the already familiar Samsung QLC memory in implementation, and this causes some features of the characteristics of the Intel drive, for example, its long, five-year warranty period, despite the popular belief that QLC 3D NAND is extremely unreliable.
However, the flash memory used in the Intel SSD 660p is a real QLC 3D NAND, each cell of which can have sixteen logical states, due to which it is possible to store four bits of information and a 33 percent increase in data recording density compared to TLC -flash. The QLC 3D NAND chips manufactured by Intel have a 64-layer structure, like the proprietary second-generation TLC 3D NAND, but the capacity of such crystals has been increased to 1 Tbit.
At the same time, Intel continues to rely on the vertical floating gate cell design, while Samsung, Toshiba, and Western Digital have settled on charge trapping. Each approach has its pros and cons, but Intel considers the floating gate to be more suitable for QLC memory, as it achieves better mutual cell isolation and prevents uncontrolled charge leakage. In other words, this approach allows Intel engineers to easily get around the main problem of QLC memory — the low reliability of data storage in the off state.
That is why the company did not cut the warranty period of the Intel SSD 660p, as if forgetting that this drive is built on cheap four-bit memory. However, as part of a full-fledged five-year warranty, the manufacturer announces a resource that is not too generous even for the average user, implying the possibility of daily rewriting of only 10% of the total drive capacity. This means that the flash memory array of the Intel SSD 660p can only be overwritten 200 times during its lifetime. And this is really not enough, because even Samsung allows almost twice as many rewrite cycles (360) for its QLC-drives of the 860 QVO series, although it gives only a three-year warranty. Drives based on TLC memory, from the point of view of the official guarantee, are allowed to be overwritten on average about 600 times during their lifetime.
Nevertheless, you need to understand that all the numbers given in the last paragraph are just declarations that may be far from reality. And to make sure that there are no special claims to the reliability of the Intel SSD 660p, when preparing this review, we carefully studied the reviews on popular foreign online sites. And indeed, despite the fact that this SSD has been on sale since August last year, there are no claims related to its premature failure or any problems with data safety in the comments of buyers. On the contrary, in most reviews, buyers praise 660p for its favorable prices and the opportunity to save money.
Users complain about a completely different thing — a not too high level of performance, which often turns out to be lower than expected. But this is not surprising. QLC 3D NAND is slower than TLC memory for a very obvious reason: the controller, when digitally processing data read from four-bit cells, has to recognize twice as many different states, which, of course, is a more complex computational task. In addition, the low degree of parallelism of the flash memory array assembled from QLC 3D NAND devices also affects. The capacity of the crystals in this case is 1 Tbit, therefore, in the design of a 512 GB drive, for example, only four devices are involved, which is clearly not enough for efficient parallelization of accesses.
The base controller for 660p is also not the flagship one. Intel continues to work closely with Silicon Motion, and for the QLC drive, it has chosen the dual-core SM2263 chip, which is a stripped-down quad-channel version of the SM2262 controller used in the popular Intel SSD 760p. For a simplified controller, computing capabilities are claimed to be about one and a half times worse, so it is not surprising that the declared performance of the Intel SSD 660p against the background of Samsung 970 EVO-level solutions looks unimportant.
But do not forget that the Intel SSD 660p is not at all going to compete with productive NVMe models. It’s more of a compromise, so the official specs shouldn’t come as a surprise.
|Form Factor||M.2 2280|
|Interface||PCI Express 3.0 x4 — NVMe 1.3|
|Flash memory: type, manufacturing process, manufacturer||Intel 64-layer 1-Tbit QLC 3D NAND|
|Controller||Silicon Motion SM2263|
|Buffer: type, volume||DDR3L, 256 MB|
|Max. sustained sequential read speed, MB/s||1500||1800||1800|
|Max. sustained sequential write speed, MB/s||1000||1800||1800|
|Max. random read speed (blocks of 4 KB), IOPS||90 000||150 000||220 000|
|Max. random write speed (blocks of 4 KB), IOPS||220 000||220 000||220 000|
|Power consumption: idle / read-write, W||0.04/4.0|
|MTBF (mean time between failures), mln h||1.6|
|Recording resource, TB||one hundred||200||400|
|Dimensions: L × H × D, mm||80.15×22.15×2.38|
|Warranty period, years||five|
You might even think that the official specifications do not take into account the SLC caching technology. But this is not so, the drive in question has accelerated write technology, but the Intel SSD 660p really does not break records and does not fill the entire bandwidth provided by the PCI Express 3.0 x4 interface for any type of operation. However, for an NVMe drive, the unit cost of which is close to $0.17 per gigabyte, this is quite normal. The Intel SSD 660p is on par with most inexpensive Kingston A1000 or Transcend SSD 110s class NVMe SSDs.
As for SLC caching, this is a matter of special pride for 660p developers. For a drive based on QLC memory, the write speed of which is extremely low, the quality of this technology means a lot. Judge for yourself: the half-terabyte version of the Intel SSD 660p writes data directly to the QLC 3D NAND array only at a speed of about 55-60 MB / s, and this is two to three times lower than the linear write performance of modern mechanical HDDs. For this reason, a QLC drive without an SLC cache would be a completely pointless device, and the size of such a cache should be as large as possible so that the user does not have to deal with the real speed of QLC 3D NAND.
For this reason, the Intel SSD 660p switched to a dynamic SLC caching scheme, in which most of the flash memory array operates in fast SLC mode, and cells are transferred to four-bit QLC mode only as necessary, when writing in one-bit mode, there is not enough space in the flash array. Generally speaking, this approach is typical for almost any drive based on Silicon Motion controllers, but Intel has so far used a static SLC cache in all of its products.
As a result, the continuous linear write speed on a pure Intel SSD 660p with a capacity of 512 GB is as follows:
Approximately 50% of the flash memory array can be transferred to SLC mode on the Intel SSD 660p 512 GB. This allows you to write to the drive in accelerated mode up to 70 GB. But even in this case, the linear write performance is limited to 900-950 MB/s, not to mention the fact that after the cache is full, when writing starts to the main memory array in a four-bit mode, the performance drops to very deplorable indicators. However, despite the fact that the Intel SSD 660p is a rather slow SSD by the standards of NVMe models, in most cases it is still obviously faster than SATA drives.
The Intel SSD 660p has three interesting features in the implementation of SLC caching algorithms. Firstly, the SLC cache of this drive has a static part of 6 GB for every 512 GB of SSD capacity. It allows you to maintain a high write speed even if the SSD volume is almost completely occupied by user files and it is not possible to deploy a capacious dynamic cache.
Secondly, the contents of the SLC cache are not transferred to the QLC memory immediately, but only during relatively long idle times of the drive. On the one hand, this somewhat reduces the efficiency of the cache, which may be full at the right time, but on the other hand, it also allows you to speed up read operations if data is accessed immediately after they are written. This tactic is especially effective in benchmarks that measure the speed of reading immediately after the creation of a test file.
In addition to this, Intel came up with something very special — the developers decided to put some control over the SLC cache in the hands of the user. In the SSD Toolbox proprietary utility for 660p, it is possible to send a command to the controller to force the transfer of all information from the SLC cache to the QLC memory. And this, obviously, allows, if necessary, to pre-prepare the drive for recording large amounts of data.
⇡ # Appearance and internal structure
The range of Intel SSDs has a clear structure. The 500 series includes SATA drives, the 700 series includes NVMe SSDs with a good level of performance, and the 600 series is somewhere in the middle. On the one hand, such drives have an NVMe interface, and on the other hand, they are inexpensive and record-breaking products. And the fact that the Intel SSD 660p is not a premium product at all can be seen right away. While the Intel SSD 760p comes with a black PCB and black label, the green PCB 660p SSD with a plain white sticker looks more like a solution for OEM builders than enthusiasts.
However, for an inexpensive NVMe SSD, the appearance does not matter much, it is much more interesting to look at what components the 660p is assembled from. For testing, we took the 512 GB version. Without an information label, it is shown in the photo below.
The Intel SSD 660p is made in the traditional M.2 2280 form factor, and in its “thin”, one-sided version. Here, the advantage of capacious QLC 3D NAND crystals is clearly manifested. When stacking 16 of these flash memory chips, you can get chips with a total volume of up to 2 TB, so in theory a single-sided 660p with a volume of 8 TB can become a reality. But so far, Intel has not released such a capacious version of its SSD, and therefore manages with chips with only two or four 64-layer QLC 3D NAND crystals inside. For example, the Intel SSD 660p 512 GB uses dual-chip chips, so you can see two flash memory chips on the photographed board. Next to them, free contact pads are reserved for another couple of microcircuits — they are required only in the 2 TB SSD version.
The SM2263 controller, which controls the quad-channel flash memory array, is easily recognizable by its nickel-plated cover that improves heat dissipation. Pay attention to the size of this chip: compared to the eight-channel SM2262, it has become noticeably smaller, which clearly hints at its reduced performance.
This is also indicated by the DDR3L SDRAM chip installed on the Intel SSD 660p. Typically, dynamic memory chips store a copy of the address translation table, which requires an amount of 1 MB of DRAM per 1 GB of flash memory. However, 660p of any capacity has a DRAM buffer size of 256 MB. Consequently, the translation table is not fully buffered in fast memory, and this may limit the performance of the drive when accessing large amounts of data.
It is worth noting that the Crucial P1 drive mentioned at the beginning of the article uses exactly the same element base as the Intel SSD 660p. But in it, the manufacturer did not save on the amount of DRAM buffer and, moreover, did not set a goal to place all QLC 3D NAND chips on one side of the board. Therefore, the Micron solution differs from the Intel SSD 660p in design and, obviously, has some peculiarities in the performance profile. Although in general the Intel SSD 660p and Crucial P1 should be similar in performance.
Intel traditionally supplies its drives with a fairly functional SSD Toolbox utility, which is also compatible with the Intel SSD 660p. In terms of capabilities, it is in many ways similar to other similar programs. This utility not only allows you to get detailed information about the state of the drive, but also has a whole set of additional tools for configuring and optimizing it.
For example, it can be used to send a TRIM command package to the drive — in the SSD Toolbox interface, this function is called SSD Optimizer. Moreover, the service utility can perform this action not only once, but also offline, according to a schedule. In addition, Intel has provided another opportunity for software optimization — forced clearing of the contents of the SLC cache of the drive. This is a unique feature of the Intel SSD Toolbox, no service utilities from other manufacturers of drives can do this.
The SSD Toolbox has a diagnostic scan option during which you can check the status and health of your flash memory. Scanning is performed in both fast and full modes — the difference is whether the scan will affect the entire array of flash memory or only some part of it.
You can also use the SSD Toolbox to check if the firmware used by the drive is up to date and initiate a Secure Erase operation.
Another proprietary feature of the Intel service utility is the System Tuner tool. With it, you can see what critical settings regarding the disk subsystem are available in the operating system, as well as get recommendations for changing them.
⇡ # Testing methodology
Testing is carried out in the Microsoft Windows 10 Enterprise x64 Build 16299 operating system, which correctly recognizes and maintains modern solid state drives. This means that in the process of passing the tests, as in normal everyday use of the SSD, the TRIM command is supported and actively involved. Performance measurement is performed with drives in a «used» state, which is achieved by pre-filling them with data. Before each test, the drives are cleaned and maintained using the TRIM command. Between individual tests, a 15-minute pause is maintained, allotted for the correct development of garbage collection technology. All tests use randomized incompressible data.
The partition within which the speed of operations is tested has a size of 32 GB, and the duration of each test is forty seconds. Such parameters, in particular, will allow you to get more relevant results for those SSDs that use various SLC caching technologies.
Applications and tests used:
- Iometer 1.1.0
- Measuring the speed of sequential reading and writing data in blocks of 128 KB (the most typical block size for sequential operations in desktop tasks). Testing is carried out at different request queue depths, which makes it possible to evaluate both realistic and peak performance parameters.
- Measuring the speed and latency of random reads and writes in 4 KB blocks (this block size is used in the vast majority of real operations). The test is carried out twice — without a request queue and with a request queue with a depth of 4 commands (typical for desktop applications that actively work with a forked file system). The data blocks are aligned with the flash memory pages of the drives.
- Establishing the dependence of random read and write speeds when the drive is working with 4-kilobyte blocks on the depth of the request queue (in the range from one to 32 commands). The data blocks are aligned with the flash memory pages of the drives.
- Establishing the dependence of random read and write speeds when the drive is working with blocks of different sizes. Blocks from 512 bytes to 256 KB are used. The depth of the request queue during the test is 4 commands. The data blocks are aligned with the flash memory pages of the drives.
- Measuring performance under a mixed multi-threaded load and establishing its dependence on the ratio between read and write operations. The test is performed twice: for sequential read and write operations in 128 KB blocks, performed in two independent threads, and for random operations with 4 KB blocks, which are performed in four independent threads. In both cases, the ratio between reads and writes varies in 20 percent increments.
- Investigation of SSD performance degradation when processing a continuous stream of random write operations. Blocks of 4 KB and a queue depth of 32 commands are used. The data blocks are aligned with the flash memory pages of the drives. The duration of the test is two hours, instantaneous speed measurements are taken every second. At the end of the test, the ability of the drive to restore its performance to its original values is additionally checked due to the operation of the garbage collection technology and after the TRIM command has been processed.
- CrystalDiskMark 6.0.2
- Synthetic benchmark that provides typical SSD performance measured on a 1 GB area of the disk «on top» of the file system. From the entire set of parameters that can be evaluated using this utility, we pay attention to the speed of sequential read and write, as well as the performance of random reads and writes in 4-kilobyte blocks without a request queue and with a queue of 32 instructions deep.
- PCMark 8 Storage Benchmark 2.0
- A test based on emulating real disk load, which is typical for various popular applications. On the tested drive, a single partition is created in the NTFS file system for the entire available volume, and the Secondary Storage 2.0 test is carried out in PCMark 8. As test results, both the final performance and the speed of execution of individual test traces generated by various applications are taken into account.
- Real file load tests
- Measuring the speed of copying directories with files of different types. For copying, a standard Windows tool is used — the Robocopy utility, as a test set, a working directory is used, including office documents, photographs and illustrations, pdf-files and multimedia content with a total volume of 8 GB.
- Measuring the speed of archiving files. The test is carried out with the same working directory as the copying, and the 7-zip archiver version 9.22 beta is chosen as a tool for compressing files. The Deflate method is used to reduce the impact of processor performance.
- Research of archive unfolding speed. The test is carried out with an archive obtained by measuring the archiving speed.
- Evaluation of the speed of launching a game application. Measures the performance of the disk subsystem when executing a script captured when launching Far Cry 4 and loading a custom save level into it. To minimize the impact of processor and memory performance, all delays that occur due to their fault were removed from the test scenario.
- Evaluation of the startup speed of applications that form a typical working user environment. The performance of the disk subsystem is measured when executing a script captured when running an application package that consists of the Google Chrome browser, Microsoft Word text editor, Adobe Photoshop graphics editor, and Adobe Premiere Pro video editor with working files. To minimize the impact of processor and memory performance, all delays that occur due to their fault were removed from the test scenario.
⇡ # Test bench
With the release of Coffee Lake Refresh processors, we decided to once again update the test system, which is used to measure the performance of NVMe SSD models. Still, such drives are primarily bought by enthusiasts moving to new platforms, and therefore it is logical to use the latest platform in test tests.
As a result, a computer with an ASRock Z390 Taichi motherboard, a Core i7-9700K processor with an integrated Intel UHD Graphics 630 graphics core and 8 GB DDR4-2666 SDRAM is used as a test platform. Drives with an M.2 interface during testing are installed in the corresponding slot of the motherboard connected to the chipset. Drives in the form of PCI Express cards are installed in a PCI Express 3.0 x4 slot, which also works through the chipset.
The volume and speed of data transfer in benchmarks are indicated in binary units (1 KB = 1024 bytes).
A separate explanation should be made regarding the closing of the Meltdown and Specter processor vulnerabilities. Existing patches significantly reduce the performance of SSDs, so measurements are taken with deactivated OC patches designed to close these vulnerabilities.
⇡#List of test participants
The Intel SSD 660p is an inexpensive NVMe drive. Its direct competitors are not solutions of the Samsung 970 EVO Plus class, but much cheaper models. Therefore, along with the results of regular test participants on the charts, you will find performance indicators of affordable NVMe drives playing in the same price category, like the ADATA SX6000 Pro or Kingston A1000.
As a result, the list of tested models looks like this:
NVMe driver versions used:
- Intel Client NVMe Driver 22.214.171.1247;
- Microsoft Windows NVMe Driver 10.0.16299.371;
- Samsung NVM Express Driver 126.96.36.1992.
⇡#Sequential read and write performance
Nobody expected any breakthroughs in performance from the Intel SSD 660p, and this is exactly what we see in the results of measuring linear speeds. This drive is a typical budget solution that can only compete with Kingston A1000 level drives. However, it is worth recalling that 660p, unlike the Kingston product, works through four PCI Express 3.0 lanes, which, however, does not help it at all. The linear speeds of the Intel QLC model are such that it could well work on two lines.
⇡#Random read performance
Random read speed is clearly the weak point of the Intel SSD 660p. This is not surprising, such is QLC 3D NAND. The latencies that occur when accessing data in the case of four-bit memory are noticeably higher than in the case of TLC memory, which immediately affects the performance indicators for small-block reading. As a result, 660p is even slower than bufferless drives, despite the fact that this model has a full DRAM buffer.
⇡#Random write performance
But with random writes, the performance of the Intel SSD 660p looks noticeably better. The SLC cache implemented by Intel engineers really works very efficiently, and in the end, small-block writes with a small request queue depth are not at all a problem for the QLC drive in question.
⇡#Performance under mixed load
The Intel SSD 660p looks relatively good for a budget model even when working with a mixed load. Drives based on Silicon Motion controllers always perform well in such scenarios, and the model built on QLC 3D NAND was no exception. Of course, 660p cannot compete with the leaders, but against the background of cheap NVMe SSDs, it looks quite up to par. And this is a good sign, since most of the real workloads that occur in modern multitasking operating systems are of a mixed nature.
⇡#Performance in CrystalDiskMark
And now — shock content! Check out how the Intel SSD 660p scores in the popular CrystalDiskMark amateur test! For comparison, next to the screenshot of the results taken for our hero, we have placed the results of the popular Samsung 970 EVO.
How is it that in random reads without a request queue, the Intel SSD 660p seriously beats a solid mid-range model? Yes, it’s very simple: this is how a special implementation of SLC caching manifests itself. An Intel drive only flushes the SLC cache to QLC memory after a few seconds of drive inactivity. Therefore, CrystalDiskMark, which first writes a test file and then immediately reads data from it, willy-nilly shows the read speed from the SLC cache, and not from the main QLC memory array. And this means that the numbers shown in the screenshot are absolutely irrelevant, although they look very nice.
⇡#Performance in PCMark 8 Storage Benchmark 2.0
Thanks to good performance in mixed operations and intelligent SLC caching technology, the Intel SSD 660p boasts a good integral result in the PCMark 2.0 test, which measures the performance of disk subsystems in real applications. According to this benchmark, 660p should be considered the best choice among inexpensive NVMe SSDs, since both the Kingston A1000 and the bufferless models of Transcend and ADATA lag behind Intel’s QLC solution.
The integral result of PCMark 8 should be supplemented with the performance indicators given out by the drives during the passage of individual test tracks, which simulate various variants of a real load. The fact is that with a diverse load, flash drives can behave in some special way.
In many applications, the Intel SSD 660p looks even better than you might expect. In fact, the relatively low performance of this SSD is manifested only in Adobe programs. In other cases, 660p can sometimes even reach the speed provided by the ADATA XPG SX8200.
⇡ # Performance under real load
With file operations, the Intel SSD 660p copes poorly. Why this is so is easy to understand. Just remember that the QLC 3D NAND array has relatively slow read speeds, especially when it comes to random access to data (that is, operations on small files). Although despite all its shortcomings, the Intel SSD 660p is still better than the Kingston A1000, and noticeably.
However, when running programs or games, the Intel SSD 660p produces not too optimistic results. In such scenarios, it is about twice as slow as the newly announced Samsung 970 EVO Plus. And here one could even say that 660p is poorly suited for the role of a system drive, but there is one caveat: the cost of the considered Intel solution on QLC memory is about one and a half times lower.
⇡ # Features of TRIM processing
In general, TRIM processing does not cause any problems for modern drives, but with offline garbage collection, which is able to restore performance after loads without the help of the operating system, the situation is different. To check, we usually conduct a simple experiment: after filling the SSD with data and then deleting it with TRIM support turned off, we wait 15 minutes, during which the SSD can try to autonomously recover due to garbage collection, and measure the performance. Then the TRIM command is forcibly sent to the drive — and the speed is measured again, which makes it possible to verify that the SSD is able to fully restore its passport speed using TRIM.
In this test, the Intel SSD 660p performs fairly mundane. After issuing the TRIM command, it fully restores its original performance, as it should be. Offline garbage collection without TRIM does not work for this drive, which, however, is quite typical for inexpensive drives with an NVMe interface.
Another important point related to TRIM concerns how much load on the controller is the processing of this command. The fact is that this is happening not so already and imperceptibly for the user. When the operating system informs the drive that some sectors are being retired by the file system, the SSD controller must consolidate these sectors and clean up the freed flash memory pages for future operations. Such a regrouping requires rewriting and clearing areas of memory, and this not only takes a noticeable amount of time, but also seriously loads the controller with work.
As a result, after deleting large amounts of data from a disk, SSD owners often experience temporary slowdowns or even “freezes” of the drive. In practice, this can cause serious discomfort, because no one expects an SSD, whose main advantage is instant response, to freeze for a few seconds. Therefore, we added an additional study to the methodology, which allows us to track how imperceptibly for the user this or that SSD serves TRIM commands. The test method is very simple: immediately after deleting a large file — 32 GB in size — we check how the drive copes with random data read operations, controlling both the read speed and the waiting time that elapses from the moment of each data request to the response of the drive.
After deleting a 32 GB file, the drive takes about six seconds to clean up. During this period, he almost completely ceases to respond to external influences. Response time grows by three orders of magnitude, and performance drops to zero even when reading. This may serve as another confirmation of the thesis that such an SSD will not be very good as a working disk, which is usually loaded with heterogeneous operations.
⇡ # Checking the temperature regime
The problem of overheating is usually acute for high-speed NVMe drives in the M.2 form factor. However, the Intel SSD 660p is a performance product that is noticeably lower than what the flagships offer. Does this mean that such an SSD does not need any special cooling?
For a practical test, we monitored the temperature regime when the test drive was loaded with sequential operations with a request queue depth of 4 commands. The measurements were carried out on an open stand, no additional airflow was performed on the SSD.
While reading, the temperature regime of the Intel SSD 660p does not raise concerns. The drive warms up to a maximum of 60-65 degrees, and this temperature is far from the critical value, which is set to 80 degrees for this SSD.
When writing, the situation is not much different from reading. We managed to warm up the drive only up to 70 degrees, which means that we did not see any manifestations of temperature throttling. In other words, the Intel SSD 660p can be attributed to the number of NVMe drives that can be used without any additional cooling, at least when it comes to the 512 GB version. However, one should not be surprised about this, a simple principle applies here: lower performance means less heating.
The Intel SSD 660p is not the first 4-bit memory-based drive to hit our lab. Therefore, we were prepared for the fact that QLC 3D NAND is noticeably slower than the usual TLC 3D NAND. However, for an NVMe drive, the high speed of a flash array is more important, and Intel’s decision to release a QLC 3D NAND-based drive with a fast interface seemed a little strange at first.
But practice has shown that four-bit memory did not make the Intel SSD 660p a completely useless product. On the contrary, modern algorithms embedded in solid state drives successfully mask many of the disadvantages of slow memory. 660p has dynamic SLC caching, and thanks to it there are no serious performance issues, at least when it comes to write operations. As for reading, where caching generally can no longer help, then under such a load, the Intel SSD 660p really lags behind many competitors. But the gap is not catastrophic, and it can be put up with. Ultimately, the overall performance of the Intel SSD 660p is quite consistent with the level set by cheap bufferless NVMe SSDs.
At the same time, do not forget about the advantages that 660p also undoubtedly has: QLC 3D NAND has reduced the cost, so the model proposed by Intel is sold at a very attractive price. In fact, this is one of the most affordable NVMe SSDs on store shelves today. In terms of price, it can be compared with the average SATA SSD of the Samsung 860 EVO level, and in some places it costs even less. And it immediately puts everything in its place. The Intel SSD 660p is not designed to set performance records in the big leagues, it should be treated differently — as a deliberately faster alternative to solid state drives with an outdated interface.
And this means that the Intel SSD 660p will not gather dust in the windows. For example, until recently, the affordable Kingston A1000 NVMe drive was very popular, and the Intel SSD 660p is not only even cheaper, but also better in terms of performance. Therefore, we have no doubt that there will be buyers for 660p, and there will certainly be many of them.
For those who are worried about the endurance of QLC 3D NAND, which is naturally lower than that of TLC 3D NAND, we have some comforting news. First, the resource of a four-bit memory with a three-dimensional structure, subject to the use of modern error correction methods, is estimated at about 1000 rewrite cycles, which corresponds to the resource of the planar TLC memory that was widely used before. And this was quite enough for everyday use before, which means it should be enough now. Secondly, the Intel SSD 660p has been on the market for half a year, and during this time no noticeable critical problems have been identified with this model. There are no massive complaints about premature failures of QLC memory. And thirdly, the Intel SSD 660p has a five-year warranty, which confirms the manufacturer’s confidence in the reliability of the proposed product.
In other words, for everyday use as part of an average PC, the considered drive is quite suitable. The main thing is to clearly understand what kind of product it is, and not want too much from it.