If you ask which interface should be used for an NVMe-enabled SSD, then anyone (who knows what NVMe is) will answer: of course PCIe 3.0 x4! True, with the justification, he is likely to have difficulties. At best, we get the answer that such drives support PCIe 3.0 x4, and the interface bandwidth matters. It has something, but all the talk about it began only when it became crowded for some drives in some operations within the framework of the "regular" SATA. But between its 600 MB/s and the (equally theoretical) 4 GB/s of the PCIe 3.0 x4 interface is just an abyss filled with a lot of options! What if one PCIe 3.0 line is enough, since it is already one and a half times more than SATA600? Fuels are added to the fire by controller manufacturers who threaten to switch to PCIe 3.0 x2 in budget products, as well as the fact that many users do not have such and such. More precisely, theoretically there are, but you can release them only by reconfiguring the system or even changing something in it, which you don’t want to do. But I want to buy a top-end solid-state drive, but there are fears that there will be no benefit from this at all (even moral satisfaction from the results of test utilities).
But is it so or not? In other words, is it really necessary to focus exclusively on the supported mode of operation - or is it still possible in practice give up principles? That is what we decided to check today. Let the check be quick and not claiming to be exhaustive, but the information received should be enough (as it seems to us) at least to think ... In the meantime, let's briefly get acquainted with the theory.
PCI Express: existing standards and their bandwidth
Let's start with what PCIe is and how fast this interface works. Often it is called a "bus", which is somewhat ideologically incorrect: as such, there is no bus to which all devices are connected. In fact, there is a set of point-to-point connections (similar to many other serial interfaces) with a controller in the middle and devices attached to it (each of which can itself be a next level hub).
First PCI version Express appeared almost 15 years ago. Orientation to use inside the computer (often within the same board) made it possible to make the standard high-speed: 2.5 gigatransactions per second. Since the interface is serial and full duplex, a single PCIe lane (x1; actually an atomic unit) provides data transfer at speeds up to 5 Gbps. However, in each direction - only half of this, i.e. 2.5 Gb / s, and this is the full speed of the interface, and not "useful": to improve reliability, each byte is encoded with 10 bits, so the theoretical bandwidth of one PCIe line 1.x is approximately 250 MB/s each way. In practice, it is still necessary to transfer service information, and as a result, it is more correct to talk about ≈200 MB / s of user data transfer. Which, however, at that time not only covered the needs of most devices, but also provided a solid supply: suffice it to recall that the predecessor of PCIe in the segment of mass system interfaces, namely the PCI bus, provided a throughput of 133 MB / s. And even if we consider not only mass implementation, but also all PCI options, then the maximum was 533 MB / s, and for the entire bus, i.e. such a PS was divided into all devices connected to it. Here, 250 MB / s (since PCI usually gives full, not useful bandwidth) per line - in exclusive use. And for devices that need more, the possibility of aggregating several lines into a single interface was initially provided, by powers of two - from 2 to 32, that is. standardized the x32 variant could already transfer up to 8 GB / s in each direction. In personal computers, x32 was not used due to the complexity of creating and breeding the corresponding controllers and devices, so the variant with 16 lines became the maximum. It was used (and is still used) mainly by video cards, since most devices do not need so much. In general, a considerable number of them and one line is enough, but some successfully use both x4 and x8: just on the storage topic - RAID controllers or SSDs.
Time did not stand still, and about 10 years ago, the second version of PCIe appeared. The improvements were not only about speeds, but a step forward was also taken in this regard - the interface began to provide 5 gigatransactions per second while maintaining the same coding scheme, i.e., the throughput doubled. And it doubled again in 2010: PCIe 3.0 provides 8 (instead of 10) gigatransactions per second, but the redundancy has decreased - now 130 is used to encode 128 bits, and not 160, as before. In principle, the PCIe 4.0 version with the next doubling of speeds is already ready to appear on paper, but in the near future we are unlikely to see it massively in hardware. In fact, PCIe 3.0 is still used in many platforms in conjunction with PCIe 2.0, because the performance of the latter is simply ... not needed for many applications. And where it is needed, the good old method of line aggregation works. Only each of them has become four times faster over the past years, i.e. PCIe 3.0 x4 is PCIe 1.0 x16, the fastest slot in mid-zero computers. This option is supported by top SSD controllers, and it is recommended to use it. It is clear that if such an opportunity exists - a lot is not enough. And if she is not? Will there be any problems, and if so, which ones? This is the question we have to deal with.
Test Methodology
Run tests with different versions The PCIe standard is not difficult: almost all controllers allow you to use not only the one they support, but also all earlier ones. It's more difficult with the number of lanes: we wanted to directly test variants with one or two PCIe lanes. The Asus H97-Pro Gamer board we usually use on Intel chipset H97 complete set does not support, but in addition to the “processor” x16 slot (which is usually used), it has another one that works in PCIe 2.0 x2 or x4 modes. We took advantage of this trio, adding to it the PCIe 2.0 mode of the “processor” slot in order to assess whether there is a difference. Still, in this case, there are no extraneous “intermediaries” between the processor and the SSD, but when working with the “chipset” slot, there is: the chipset itself, which is actually connected to the processor by the same PCIe 2.0 x4. We could add a few more modes of operation, but we were still going to conduct the main part of the study on a different system.
The fact is that we decided to take this opportunity and at the same time check one "urban legend", namely, the belief about the usefulness of using top-end processors for testing drives. So we took the eight-core Core i7-5960X - a relative of the Core i3-4170 usually used in tests (these are Haswell and Haswell-E), but which has four times as many cores. In addition, the Asus Sabertooth X99 board found in the bins is useful to us today by the presence PCI slot e x4, actually able to work as x1 or x2. In this system, we tested three x4 variants (PCIe 1.0/2.0/3.0) from the processor and chipset PCIe 1.0 x1, PCIe 1.0 x2, PCIe 2.0 x1 and PCIe 2.0 x2 (in all cases, chipset configurations are marked on the diagrams with the icon (c)). Does it make sense now to turn to the first version of PCIe, given the fact that there is hardly a single board that supports only this version of the standard and can boot from an NVMe device? From a practical point of view, no, but to check a priori the expected ratio of PCIe 1.1 x4 = PCIe 2.0 x2 and the like, it will come in handy for us. If the test shows that the bus scalability corresponds to the theory, it means that it does not matter that we have not yet been able to obtain practically significant ways to connect PCIe 3.0 x1 / x2: the first one will be exactly identical to PCIe 1.1 x4 or PCIe 2.0 x2, and the second - PCIe 2.0 x4 . And we have them.
In terms of software, we limited ourselves only to Anvil’s Storage Utilities 1.1.0: it measures various low-level characteristics of drives quite well, but we don’t need anything else. On the contrary: any influence of other components of the system is extremely undesirable, so low-level synthetics have no alternative for our purposes.
As a "working body" we used a 240 GB Patriot Hellfire. As it was found during testing, this is not a performance record holder, but its speed characteristics are quite consistent with the results. best SSD same class and same capacity. Yes, and there are already slower devices on the market, and there will be more of them. In principle, it will be possible to repeat the tests with something faster, however, as it seems to us, there is no need for this - the results are predictable. But let's not get ahead of ourselves, but let's see what we got.
Test results
When testing Hellfire, we noticed that the maximum speed on sequential operations can only be “squeezed” out of it by a multi-threaded load, so this should also be taken into account for the future: the theoretical throughput is theoretical, because “real” data, received in different programs according to different scenarios, they will depend more not on it, but on these same programs and scenarios - in the case, of course, when force majeure circumstances do not interfere :) Just such circumstances we are now observing: it was already said above that PCIe 1 .x x1 is ≈200 MB/s, and that's exactly what we're seeing. Two PCIe 1.x lanes or one PCIe 2.0 lane is twice as fast, and that's exactly what we're seeing. Four PCIe 1.x lanes, two PCIe 2.0 lanes, or one PCIe 3.0 lane is twice as fast, which was confirmed for the first two options, so the third is unlikely to be different. That is, in principle, scalability, as expected, is ideal: the operations are linear, Flash copes with them well, so the interface matters. Flash stops do well to PCIe 2.0 x4 for writing (so PCIe 3.0 x2 will do). Reading "may" more, but the last step already gives one and a half, and not two (as it potentially should be) increase. We also note that there is no noticeable difference between the chipset and processor controllers, and also between the platforms. However, LGA2011-3 is a little ahead, but only a little.
Everything is smooth and beautiful. But templates do not tear: the maximum in these tests is only a little more than 500 MB / s, and even SATA600 or (in the appendix to today's testing) PCIe 1.0 x4 / PCIe 2.0 x2 / PCIe 3.0 x1. That's right: do not be afraid of the release of budget controllers for PCIe x2 or the presence of only so many lines (and the version of the 2.0 standard) in the M.2 slots on some boards, when more is not needed. Sometimes so much is not needed: the maximum results are achieved with a queue of 16 commands, which is not typical for mass software. More often there is a queue with 1-4 commands, and for this you can get by with one line of the very first PCIe and even the very first SATA. There are overheads and such, though, so a quick interface is useful. However, too fast - perhaps not harmful.
And in this test, the platforms behave differently, and with a single command queue, they behave fundamentally differently. The "trouble" is not at all that many cores are bad. They are still not used here, except perhaps one, and not so much that the boost mode unfolds with might and main. So we have a difference of about 20% in the frequency of the cores and one and a half times in the cache memory - in Haswell-E it operates at a lower frequency, and not synchronously with the cores. In general, the top platform can only be useful for kicking out the maximum "yops" through the most multi-threaded mode with a large command queue depth. The only pity is that from the point of view practical work this is a very spherical synthetic in a vacuum :)
On the record, the state of affairs has not fundamentally changed - in every sense. But, funny, on both systems, the PCIe 2.0 x4 mode in the “processor” slot turned out to be the fastest. On both! And with multiple checks/rechecks. At this point, you might wonder if you need these are your new standards Or is it better not to rush anywhere at all ...
When working with blocks of different sizes, the theoretical idyll breaks down that increasing the speed of the interface still makes sense. The resulting numbers are such that a couple of PCIe 2.0 lanes would be enough, but in reality, in this case, the performance is lower than that of PCIe 3.0 x4, albeit not at times. And in general, here the budget platform "scores" the top one to a much greater extent. But just such operations are mainly found in application software, i.e. this diagram is the closest to reality. As a result, there is nothing surprising that thick interfaces and trendy protocols do not give any “wow effect”. More precisely, those who are passing from mechanics will be given, but exactly the same as any solid-state drive with any interface will provide it.
Total
To make it easier to perceive the picture of the hospital as a whole, we used the score given by the program (total - for reading and writing), normalizing it according to the PCIe 2.0 x4 "chipset" mode: this moment it is he who is the most widely available, since it is found even on LGA1155 or AMD platforms without the need to "offend" the video card. In addition, it is equivalent to PCIe 3.0 x2, which budget controllers are preparing to master. Yes, and on the new AMD platform AM4, again, it is this mode that can be obtained without affecting the discrete video card.
So what do we see? The use of PCIe 3.0 x4, if possible, is certainly preferable, but not necessary: it brings literally 10% additional performance to middle-class NVMe drives (in its initially top segment). And even then - due to operations, in general, not so often encountered in practice. Why is this option implemented in this case? Firstly, there was such an opportunity, but the pocket does not pull the stock. Secondly, there are drives and faster than our test Patriot Hellfire. Thirdly, there are such areas of activity where loads that are “atypical” for a desktop system are just quite typical. And it is there that the performance of the storage system is most critical, or at least the ability to make part of it very fast. But this does not apply to ordinary personal computers.
In them, as we can see, the use of PCIe 2.0 x2 (or, accordingly, PCIe 3.0 x1) does not lead to a dramatic decrease in performance - only by 15-20%. And this is despite the fact that in this case we limited the potential capabilities of the controller by four times! For many operations, this throughput is enough. Here, one PCIe 2.0 lane is no longer enough, so it makes sense for controllers to support exactly PCIe 3.0 - and in conditions of a severe shortage of lanes in modern system this will work well. In addition, x4 width is useful - even if there is no support for modern PCIe versions in the system, it will still allow you to work at normal speed (albeit slower than it could potentially), if there is a more or less wide slot.
In principle, a large number of scenarios in which the flash memory itself turns out to be the bottleneck (yes, this is possible and inherent not only in mechanics) leads to the fact that the four lanes of the third PCIe version on this drive overtake the first one by about 3.5 times - the theoretical throughput of these two cases differs by 16 times. From which, of course, it does not follow that you need to rush to master very slow interfaces - their time has gone forever. It's just that many of the features of fast interfaces can only be implemented in the future. Or in conditions that an ordinary user of an ordinary computer will never directly encounter in his life (with the exception of those who like to measure themselves with what they know). Actually, that's all.
). One of the criteria for this idea was the free use of USB 3.0 ports at home. Well, since am2 platform motherboards were not equipped with such ports, there is only one way out - to use a controller.
There were no questions about delivery - the track was international and everything was tracked correctly.
But the packaging (more precisely, its absence) - discouraged. The seller is either naive to the point of recklessness or is the owner of a whole bunch of steel eggs. Because the antistatic bag with the controller packs simply in the mail. Without a hint of packaging / sealing material. As a result, the controller was delivered to me with a broken capacitor (+ one more on parole). And in the comments through one, buyers write about a scratched disk with drivers or a bent mounting frame.
As for the driver disk. Specifically, in my case, I can not check its serviceability / performance. Because I don't have the equipment to read it. But there are a few caveats to this:
1) Under Windows 10, the controller does not require driver installation. (I confirm!)
2) In the reviews they write that the seller sends the driver on request. (I can’t comment, I didn’t apply. There was no need)
3) The driver (for XP, VISTA, 7 and 8) can always be downloaded on the chip model - VL805. 
Frankly speaking, apart from the jamb with capacitors, there is no more place to find fault with the installation. Soldered neat and clean.
Geometrically, the board is small in size and can easily fit even in a skinny system unit, but there is no corresponding mounting frame for this. (The ideal, in my opinion, would be the possibility of choosing when placing an order. But it’s true, dreams ...) 
The VL805 chip "can do" four USB 3.0 ports. Two of which (A F) are directed outward, and two (19 pin) are directed inside the system unit. Next to the latter is a 4-pin molex connector for external power supply. 
For speed tests, I used the following external drives:
1) Memory card connected using a USB 3.0 card reader ().
[Further on all the screenshots on the left are measurements on the old port 2.0. And on the right on the new one - 3.0] 
2) 2.5 inch HDD western digital WD Scorpio Blue 320 GB () paired with USB 3.0 "pocket" 
3) SSD OCZ Vertex 2 () paired with the above-mentioned external box from AgeStar. 
What can I say about this. The values of the numbers turned out to be expected, because USB 2.0 is an old and slow port. And the difference in speed between it and a USB 3.0 port will be all the more noticeable the faster the drive you intend to use.
If we talk specifically about this considered board, then due to the carelessness of the seller, it may very well be that when buying it you will have to work with a soldering iron and / or download a driver from the Internet. On the other hand, it's one of the cheapest options to equip your older PC with a faster interface.
Now I say goodbye. Be kind!
I plan to buy +21 Add to favorites Liked the review +24 +41USB is the most popular modern interface for connecting external peripherals. Specifically, the USB 2.0 standard provides data transfer rates up to 480 Mbps. It is also considered obsolete in light of the appearance of faster and more advanced versions - 3.0 and 3.1. Nonetheless, USB capabilities 2.0 is enough for peripherals that do not require high data transfer speed or powerful power - keyboards, mice, webcams, flash drives, etc.; and this interface is inexpensive. So connectors of this type are still widely used in computer technology and the situation is unlikely to change in the near future.
USB 3.2 gen1
USB is the most popular modern interface for connecting external peripherals. USB 3.2 gen1 (previously labeled USB 3.1 gen1 and USB 3.0) is the next generation of this interface after 2.0. In this version, the data transfer rate has been increased to 4.8 Gb / s, and the amount of power that can be supplied to the connector has also increased.
USB 3.2 gen2
USB is the most popular modern interface for connecting external peripherals. Version 3.2 gen2 (formerly USB 3.1 gen2 and USB 3.1) has transfer speeds up to 10 Gbps.
USB C 3.2 gen1
eSATA is a specialized interface for connecting external drives. It provides a data transfer rate of 2.4 Gbps (5 times faster than the USB 2.0 standard). Also, connecting via eSATA is convenient because the drive does not take USB port, which may be useful for other peripherals. On the other hand, there are relatively few peripheral devices with such a connector.
COM port
The term "COM port" usually refers to the RS-232 interface. This is a service connector, nowadays used mainly for connecting various specialized equipment - CNC machines, sources uninterruptible power supply, programmable logic controllers, some models of routers, etc. In addition, this interface can be used for direct connection between two computers, as well as to control the settings of TVs, projectors, audio receivers and other audio and video equipment. A COM port is rarely installed in modern computers during the initial assembly, so a PCI controller is usually required to use this interface.
LPT port
LPT is an outdated interface for connecting peripheral devices to a computer. It was used mainly for printers, scanners and external drives, but technically it can also be used with other devices - from modems and joysticks to highly specialized equipment and a direct connection between two computers. New peripherals for LPT are practically not produced, however, there are quite a few old devices that work; in light of this, PCI controllers with similar ports continue to be produced.
Note that the LPT is quite large, so it is problematic to provide more than two such connectors in a PCI controller. However, in practice, one port is often enough.
SAS (SFF-8088)
SFF-8088 is one of the types of connectors used to connect external drives via SAS interface. This connection provides data transfer rates up to 12 Gb / s, which is especially convenient when working with large amounts of data. At the same time, relatively few peripherals with such a connector are produced, so connectors of this type are quite rare.
USB 2.0
The number of internal USB 2.0 ports provided in the controller.
See "USB 2.0" above for details on the interface itself. And the internal ports are located directly on the controller board and are directed, in accordance with the name, inside the case. They are well suited for peripherals that need to be kept permanently connected to the system and can be hidden in the case; Examples of such devices include Wi-Fi and Bluetooth adapters. The advantage of connecting to the internal USB is that the device does not stick out of the case and does not take up external ports that can be useful for other peripherals.
USB 3.2 gen1
The number of internal USB 3.2 gen1 (formerly USB 3.1 gen1 and USB 3.0) ports available on the controller.
For details on the interface itself, see "USB 3.2 gen1" above. And the internal ports are located directly on the controller board and are directed, in accordance with the name, inside the case. They are well suited for peripherals that need to be kept permanently connected to the system and can be hidden in the case; Wi-Fi and Bluetooth adapters are examples of such devices. The advantage of connecting to the internal USB is that the device does not stick out of the case and does not take up external ports that can be useful for other peripherals.
USB 3.2 gen2
The number of internal USB 3.2 gen2 (formerly USB 3.1 gen2 and USB 3.1) ports available on the controller.
For details on the interface itself, see "USB 3.2 gen2" above. And the internal ports are located directly on the controller board and are directed, in accordance with the name, inside the case. They are well suited for peripherals that need to be kept permanently connected to the system and can be hidden in the case; Wi-Fi and Bluetooth adapters are examples of such devices. The advantage of connecting to the internal USB is that the device does not stick out of the case and does not take up external ports that can be useful for other peripherals.
SATA
When evaluating the number of connectors, it should be borne in mind that one connector can be output to two USB ports.
Cache size
The amount of cache memory provided in the controller.
Cache memory is used in RAID controllers (see "Type"). It serves to store the data that is most often used during the operation of the device: the cache provides high-speed access to this data, thus improving the overall performance of the controller. The larger the cache, the more data can be stored in it and the faster the device can work; on the other hand, large amounts of memory have a corresponding effect on the cost.
Extra food
Connector type for connecting additional power, for which the controller is designed.
- Molex. A characteristic four-pin power connector, which has a rather large size. Quite versatile, used to power a wide variety of system components
- SATA. Power connector released at the same time as the corresponding data interface (see above) specifically for hard drives; however, it can also be used for other components. Has a 15 pin plug.
- Molex/SATA. Ability to connect to a power controller using any of the connectors described above. This design is the most versatile, it minimizes the likelihood that a suitable connector is not found in the power supply. On the other hand, such versatility affects the size and price of the device.
Occupied slots
The number of standard slots on the rear panel that the controller occupies. This information necessary in order to assess whether there is enough space in the case to install the board. Usually controllers occupy , or slots.
low profile
This feature means that the controller board has a small height; and the height in this case is how much the board protrudes above the "motherboard" in which it is installed.
Low-profile components are mainly designed for use in compact form factors where there is no room for full-sized boards. However, nothing prevents the installation of such a fee in more large body.
Board length
The total length of the controller is from the bar fixed on the back wall of the PC case to the opposite end of the board. This information allows you to assess whether there is enough space in the case to install this component.
In the past, there were only two types of SSDs that the mainstream buyer was interested in: either high-speed premium models like the Samsung 850 PRO, or value-for-money offerings like the Crucial BX100 or SanDisk Ultra II. That is, the segmentation of the SSD market was extremely weak, and although competition between manufacturers was unfolding in the areas of performance and price, the gap between top and bottom solutions remained quite small. This state of affairs was partly due to the fact that SSD technology itself significantly improves the user experience with a computer, and therefore implementation issues fade into the background for many. For the same reason, consumer SSDs have been incorporated into the old infrastructure, which initially focused on mechanical hard drives. This greatly facilitated their implementation, however, concluded the SSD in a rather narrow framework, which in many respects hinders both the growth in throughput and the reduction in the latency of the disk subsystem.
But until a certain time, this state of affairs suited everyone. SSD technology was new, and users moving to SSDs were happy with their purchase, even though they were essentially getting products that actually performed far beyond their capabilities, with artificial performance barriers hindering their performance. However, to date, SSD, perhaps, can already be considered the real mainstream. Any self-respecting owner of a personal computer, if he does not have at least one SSD in his system, is very serious about acquiring it in the very near future. And under these conditions, manufacturers are simply forced to think about finally deploying full-fledged competition: destroying all barriers and moving on to the production of wider product lines that fundamentally differ in their proposed characteristics. Fortunately, all the necessary ground has been prepared for this, and, first of all, most SSD developers have the desire and opportunity to start releasing products that work not through a legacy SATA interface, but through a much more efficient PCI Express bus.
Since SATA bandwidth is limited to 6 Gb/s, the maximum speed of the flagship SATA SSDs does not exceed 500 MB/s. However, modern flash drives are capable of much more: after all, if you think about it, they have more in common with system memory than with mechanical hard drives. As for the PCI Express bus, now it is actively used as a transport layer when connecting graphic cards and other additional controllers that need high-speed data exchange, such as Thunderbolt. One PCI Express Gen 2 lane delivers up to 500 MB/s of bandwidth, while a PCI Express 3.0 lane can reach speeds of up to 985 MB/s. Thus, an interface card installed in a PCIe x4 slot (with four lanes) can exchange data at speeds up to 2 GB / s in the case of PCI Express 2.0 and up to almost 4 GB / s when using PCI Express third generation. These are excellent indicators, which are quite suitable for modern solid-state drives.
From what has been said, it naturally follows that in addition to SATA SSDs, high-speed drives using the PCI Express bus should gradually find distribution on the market. And it really is happening. In stores, you can find several models of consumer SSDs from leading manufacturers, made in the form of expansion cards or M.2 cards that use different variants of the PCI Express bus. We decided to put them together and compare them in terms of performance and other parameters.
Test participants
Intel SSD 750 400 GBIn the SSD market Intel adheres to a rather non-standard strategy and does not pay too much attention to the development of SSDs for the consumer segment, concentrating on products for servers. However, its proposals do not become uninteresting, especially when it comes to a solid-state drive for the PCI Express bus. In this case, Intel decided to adapt its most advanced server platform for use in a high-performance client SSD. This is how the Intel SSD 750 400 GB was born, which received not only impressive performance characteristics and a number of server-level technologies responsible for reliability, but also support for the newfangled NVMe interface, which a few words should be said separately.
If we talk about specific improvements to NVMe, then the reduction in overhead costs deserves mention first of all. For example, transferring the most typical 4-kilobyte blocks in the new protocol requires only one command instead of two. And the entire set of control instructions has been simplified so much that their processing at the driver level reduces the processor load and the resulting delays by at least half. The second important innovation is support for deep pipelining and multitasking, which consists in the ability to create multiple request queues in parallel instead of the previously existing single queue for 32 commands. The NVMe interface protocol is capable of serving up to 65536 queues, and each of them can contain up to 65536 commands. In fact, any restrictions are eliminated at all, and this is very important for server environments, where a huge amount of simultaneous I / O operations can be assigned to the disk subsystem.
But despite working through the NVMe interface, the Intel SSD 750 is still not a server, but a consumer drive. Yes, almost the same hardware platform as in this drive is used in server-class SSDs Intel DC P3500, P3600 and P3700, but the Intel SSD 750 uses a cheaper ordinary MLC NAND, and besides, the firmware is modified. The manufacturer believes that thanks to such changes, the resulting product will appeal to enthusiasts, since it combines high power, fundamentally new interface NVMe and not too intimidating cost.
The Intel SSD 750 is a half-height PCIe x4 card that can use four 3.0 lanes and achieve sequential transfer rates up to 2.4 GB/s and random operations up to 440K IOPS. True, the most capacious modification of 1.2 TB is the most productive, while the 400 GB version we received for tests is a little slower.
The drive board is completely covered with armor. On the front side, this is an aluminum heatsink, and on the reverse side, there is a decorative metal plate that does not actually come into contact with the microcircuits. It should be noted that the use of a radiator here is a necessity. The main controller of the Intel SSD generates a lot of heat, and at high load, even a drive equipped with such cooling can warm up to temperatures of the order of 50-55 degrees. But thanks to the pre-installed cooling, there is no hint of throttling - performance remains constant even during continuous and intensive use.
The Intel SSD 750 is based on the Intel CH29AE41AB0 server-level controller, which operates at a frequency of 400 MHz and has eighteen (!) Channels for connecting flash memory. Considering that most consumer SSD controllers have either eight or four channels, it becomes clear that the Intel SSD 750 can indeed pump significantly more data over the bus than conventional SSD models.
As for the flash memory used, the Intel SSD 750 does not innovate in this area. It is based on the usual Intel-made MLC NAND, released according to the 20-nm process technology and having both 64 and 128 Gb cores interspersed. It should be noted that most other SSD manufacturers abandoned such memory a long time ago, switching to chips made according to thinner standards. And Intel itself has begun transferring not only its consumer, but also server drives to 16-nm memory. However, despite all this, the Intel SSD 750 uses older memory, which supposedly has a higher resource.
The server origin of the Intel SSD 750 can also be traced in the fact that the total flash memory capacity of this SSD is 480 GiB, of which only about 78 percent is available to the user. The rest is allocated to the replacement fund, garbage collection and data protection technologies. The Intel SSD 750 implements the traditional flagship RAID 5-like scheme at the level of MLC NAND chips, which allows you to successfully restore data even if one of the chips completely fails. In addition, the Intel SSD provides complete data protection from power outages. The Intel SSD 750 has two electrolytic capacitors, and their capacity is enough for a regular shutdown of the drive in offline mode.
Kingston HyperX Predator 480 GB
Kingston HyperX Predator is a much more traditional solution compared to the Intel SSD 750. Firstly, it works through the AHCI protocol, not NVMe, and secondly, this SSD requires the more common PCI Express 2.0 bus to connect to the system. All this makes the Kingston version somewhat slower - peak speeds for sequential operations do not exceed 1400 MB / s, and random ones - 160 thousand IOPS. But HyperX Predator does not impose any special requirements on the system - it is compatible with any, including old platforms.
Along with this, the drive has a not quite simple two-component design. The SSD itself is an M.2 form factor board, which is supplemented with a PCI Express adapter that allows you to connect M.2 drives through regular full-size PCIe slots. The adapter is made in the form of a half-height PCIe x4 card that uses all four PCI Express lanes. Thanks to this design, Kingston sells its HyperX Predator in two versions: as a PCIe SSD for desktops and as an M.2 drive for mobile systems (in this case, an adapter is not included in the delivery).
Kingston HyperX Predator is based on the Marvell Altaplus controller (88SS9293), which, on the one hand, supports four PCI Express 2.0 lanes, and on the other hand, has eight channels for connecting flash memory. This is Marvell's fastest mass-produced PCI Express SSD controller to date. However, Marvell will soon have faster followers with support for NVMe and PCI Express 3.0, which the Altaplus chip does not have.
Since Kingston itself does not produce controllers or memory, assembling its SSDs from the element base purchased from other manufacturers, there is nothing strange in the fact that the HyperX Predator PCIe SSD is based not only on a third-party controller, but also on 128-gigabit 19- nm MLC NAND chips from Toshiba. Such memory has a low purchase price and is now installed in many products of Kingston (and other companies), and primarily in consumer models.
However, the use of such memory has created a paradox: despite the fact that, according to its formal positioning, the Kingston HyperX Predator PCIe SSD is a premium product, it only has a three-year warranty, and the stated mean time between failures is much less than that of the flagship SATA SSDs. other manufacturers.
No special data protection technologies are provided in Kingston HyperX Predator either. But the drive has a relatively large area hidden from the user's eyes, the size of which is 13 percent of the total capacity of the drive. The spare flash memory included in it is used for garbage collection and wear leveling, but is primarily spent on replacing failed memory cells.
It only remains to add that the HyperX Predator design does not provide any special means for removing heat from the controller. Unlike most other high-performance solutions, this drive does not have a heatsink. However, this SSD is not prone to overheating at all - its maximum heat dissipation is only slightly higher than 8 watts.
OCZ Revodrive 350 480 GB
The OCZ Revodrive 350 is rightfully one of the oldest consumer SSDs with PCI interface express. Back in the days when none of the other manufacturers even thought about releasing client PCIe SSDs, in model range OCZ had the RevoDrive 3 (X2), a prototype of the current Revodrive 350. However, the OCZ PCIe drive's lingering roots make it a bit of an odd proposition compared to today's competitors. While most manufacturers of high-performance PC drives use modern controllers with native support for the PCI Express bus, the Revodrive 350 has a very intricate and clearly suboptimal architecture. It is based on two or four (depending on the volume) SandForce SF-2200 controllers, which are assembled in a zero-level RAID array.
If we talk about the 480 GB OCZ Revodrive 350 model that took part in this test, then it is actually based on four SATA SSDs with a capacity of 120 GB each, each of which is based on its own SF-2282 chip (an analogue of the widespread SF-2281) . Then these elements are combined into a single four-component RAID 0 array. However, for this purpose, a not quite familiar RAID controller is used, but a proprietary virtualization processor (VCA 2.0) OCZ ICT-0262. However, it is very similar to the fact that this name hides a flipped Marvell 88SE9548 chip, which is a four-port SAS / SATA 6 Gb / s RAID controller with a PCI Express 2.0 x8 interface. But even so, the OCZ engineers wrote their own firmware and driver for this controller.
The uniqueness of the software component of RevoDrive 350 lies in the fact that it does not implement quite a classic RAID 0, but some kind of it with interactive load balancing. Instead of splitting the data stream into blocks of a fixed size and sequentially transferring them to different SF-2282 controllers, VCA 2.0 technology involves the analysis and flexible redistribution of I / O operations depending on the current occupancy of the flash memory controllers. Therefore, the RevoDrive 350 looks like a solid state drive to the user. You can’t enter its BIOS, and it’s impossible to find out that a RAID array is hidden in the bowels of this SSD without a detailed acquaintance with the hardware stuffing. What's more, unlike conventional RAID arrays, RevoDrive 350 supports all typical SSD features: SMART monitoring, TRIM and Secure Erase.
RevoDrive 350 is available as boards with PCI Express 2.0 x8 interface. Despite the fact that all eight lines of the interface are actually used, the declared performance indicators are noticeably lower than their total theoretical throughput. The maximum speed of sequential operations is limited to 1800 MB / s, and the performance of arbitrary operations does not exceed 140 thousand IOPS.
It's worth noting that the OCZ RevoDrive 350 is a full-height PCI Express x8 card, which means it's physically larger than all the other SSDs we tested, and therefore can't be installed in low-profile systems. The front surface of the RevoDrive 350 board is covered with a decorative metal casing, which also acts as a heatsink for the base RAID controller chip. The SF-2282 controllers are located on the reverse side of the board and are devoid of any cooling.
To form the flash memory array, OCZ used chips from its parent company, Toshiba. The chips used are manufactured using a 19-nm process technology and have a capacity of 64 Gbps. The total amount of flash memory in the RevoDrive 350 480 GB is 512 GB, but 13% is reserved for internal needs - wear leveling and garbage collection.
It is worth noting that the architecture of the RevoDrive 350 is not unique. There are several more models of similar SSDs on the market that work on the principle of a “RAID array of SATA SSDs based on SandForce controllers”. However, all such solutions, like the OCZ PCIe drive under consideration, have an unpleasant drawback - their write performance degrades over time. This is due to the peculiarities of the internal algorithms of SandForce controllers, the TRIM operation for which does not return the write speed to the original level.
The indisputable fact that the RevoDrive 350 is one step below the next generation PCI Express drives is also emphasized by the fact that this drive is given only a three-year warranty, and its guaranteed write resource is only 54 TB - several times less than that of competitors. Moreover, despite the fact that RevoDrive 350 is based on the same design as the server Z-Drive 4500, it does not have any protection against power surges. However, all this does not prevent OCZ, with its inherent audacity, to position the RevoDrive 350 as a premium solution at the level of the Intel SSD 750.
Plextor M6e Black Edition 256 GB
It should be noted right away that the Plextor M6e Black Edition drive is a direct successor of the well-known M6e model. The similarity of the novelty to its predecessor can be traced in almost everything, if we talk about the technical, not the aesthetic component. The new SSD also has a two-piece design, including the actual drive in M.2 2280 format and an adapter that allows you to install it in any regular PCIe x4 slot (or faster). It is also based on the eight-channel Marvell 88SS9183 controller, which communicates with the outside world via two PCI Express 2.0 lines. Just like in the previous version, the M6e Black Edition uses Toshiba's MLC flash memory.
And this means that despite the fact that the M6e Black Edition assembled looks like a half-height PCI Express x4 card, in fact, this SSD uses only two PCI Express 2.0 lanes. Hence the not-too-impressive speeds, which are only slightly faster than traditional SATA SSDs. Passport performance on sequential operations is limited to 770 MB / s, and on arbitrary - 105 thousand IOPS. It is worth noting that the Plextor M6e Black Edition works according to the legacy AHCI protocol, and this ensures its wide compatibility with various systems.
Despite the fact that the Plextor M6e Black Edition, like the Kingston HyperX Predator, is a combination of a PCI Express adapter and a "core" in the M.2 format of the board, it is impossible to determine this from the front side. The entire drive is hidden under a figured black aluminum casing, in the center of which a red heatsink is embedded, which should remove heat from the controller and memory chips. The designers' calculation is clear: a similar color solution is widely used in various gaming hardware, so the Plextor M6e Black Edition will look harmoniously next to many gaming motherboards and video cards from most leading manufacturers.
The flash memory array in the Plextor M6e Black Edition is powered by Toshiba's second-generation 19nm MLC NAND chips with a capacity of 64Gbps. The reserve used for the replacement fund and the operation of internal wear-leveling and garbage collection algorithms is allocated 7 percent of the total. Everything else is available to the user.
Due to the use of a rather weak Marvell 88SS9183 controller with an external PCI Express 2.0 x2 bus, the Plextor M6e Black Edition drive should be considered a rather slow PCIe SSD. However, this does not prevent the manufacturer from referring this product to the upper price category. On the one hand, it is still faster than a SATA SSD, and on the other hand, it has good reliability characteristics: it has a long time between failures and is covered by a five-year warranty. However, no special technologies that can protect the M6e Black Edition from power surges or increase its resource are implemented in it.
Samsung SM951 256 GB
Samsung SM951 is the most elusive drive in today's testing. The fact is that initially this is a product for computer assemblers, so it is rather faded in retail sales. Nevertheless, if desired, it is still possible to buy it, so we did not refuse to consider the SM951. Moreover, judging by the characteristics, this is a very high-speed model. It is designed to work on the PCI Express 3.0 x4 bus, uses the AHCI protocol and promises impressive speeds: up to 2150 MB / s in sequential operations and up to 90,000 IOPS in random operations. But most importantly, despite all this, the Samsung SM951 is cheaper than many other PCIe SSDs, so looking for it on sale may have a very specific business case.
Another feature of the Samsung SM951 is that it comes in M.2 form. Initially, this solution is aimed at mobile systems, so no adapters for full-size PCIe slots are included with the drive. However, this can hardly be considered a serious drawback - most flagship motherboards also have M.2 interface slots on board. In addition, the necessary adapter boards are widely available on the market. The Samsung SM951 itself is an M.2 2280 form factor board, the connector of which has an M type key, indicating the need for an SSD in four PCI Express lanes.
The Samsung SM951 is based on the exceptionally powerful Samsung UBX controller, developed by the manufacturer specifically for PCI Express SSDs. It is based on three cores with ARM architecture and in theory is able to work with both AHCI and NVMe commands. In the SSD in question, only AHCI mode is enabled in the controller. But an NVMe version of this controller will soon be seen in a new consumer SSD that Samsung is due to launch this fall.
Due to the OEM focus, no warranty period or predicted endurance is reported for the drive in question. These parameters must be declared by the assemblers of the systems in which the SM951 will be installed, or by the sellers. However, it should be noted that 3D V-NAND, which is now actively promoted by Samsung in consumer SSDs as a faster and more reliable type of flash memory, is not used in the SM951. Instead, it uses the usual planar Toggle Mode 2.0 MLC NAND, produced, presumably, using 16nm technology (some sources suggest a 19nm process technology). This means that the SM951 should not be expected to have the same high endurance as the flagship 850 PRO SATA drive. In this parameter, the SM951 is closer to the usual mid-range models, in addition, only 7 percent of the flash memory array is allocated for redundancy in this SSD. The Samsung SM951 does not have any special server-level technologies to protect data from power failures. In other words, the emphasis in this model is placed solely on the speed of work, and everything else is cut off to reduce the cost.
It is worth noting one more thing. Under high load, the Samsung SM951 exhibits quite serious heating, which in the end can even lead to the inclusion of throttling. Therefore, in high-performance systems for SM951, it is desirable to organize at least airflow, or better, close it with a radiator.
Comparative characteristics of tested SSDs
Compatibility Issues
Like every new technology, PCI Express SSDs cannot yet boast 100% trouble-free performance with any platform, especially older ones. Therefore, you have to choose the right SSD not only based on consumer characteristics, but also with an eye to compatibility. Here it is important to keep in mind two points.First of all, different SSDs can use different numbers of PCI Express lanes and different generations this tire - 2.0 or 3.0. Therefore, before buying a PCIe drive, you need to make sure that the system where you plan to install it has a free slot with the required bandwidth. Of course, faster PCIe SSDs are backwards compatible with slower slots, but in this case, purchasing a high-speed SSD does not make much sense - it simply cannot reach its full potential.
The Plextor M6e Black Edition has the widest compatibility in this sense - it requires only two PCI Express 2.0 lanes, and such a free slot is sure to be found on almost any motherboard. Kingston HyperX Predator already needs four PCI Express 2.0 lanes: many motherboards also have such PCIe slots, but some cheap platforms may not have extra slots with four or more PCI Express lanes. This is especially true for motherboards built on low-level chipsets, the total number of lines in which can be cut down to six. Therefore, before purchasing a Kingston HyperX Predator, be sure to check that the system has a free slot with four or more PCI Express lanes.
The OCZ Revodrive 350 takes things a step further - it already requires eight PCI Express lanes. Such slots are usually implemented not by the chipset, but by the processor. Therefore, the best place to use such a drive is LGA 2011/2011-3 platforms, where the PCI Express processor controller has an excessive number of lanes, which allows servicing more than one video card. In systems with LGA 1155/1150/1151 processors, OCZ Revodrive 350 will be appropriate only if the graphics integrated into the CPU are used. Otherwise, in favor of a solid state drive, you will have to take away half the lines from the GPU by switching it to PCI Express x8 mode.
The Intel SSD 750 and Samsung SM951 are somewhat similar to the OCZ Revodrive 350: they are also preferable to use in CPU-powered PCI Express slots. However, the reason here is not the number of lanes - they require only four PCI Express lanes, but the generation of this interface: both of these drives are able to use the increased bandwidth of PCI Express 3.0. However, there is an exception: the latest Intel chipsets of the hundredth series, designed for processors of the Skylake family, received PCI support Express 3.0, so in the latest LGA 1151 boards, they can be installed without a twinge of conscience in chipset PCIe slots, to which at least four lines are connected.
The compatibility problem has a second part. To all the limitations associated with the bandwidth of various variations of PCI Express slots, there are also restrictions associated with the protocols used. The most trouble-free in this sense are SSDs that work through AHCI. Due to the fact that they emulate the behavior of a conventional SATA controller, they can work with any, even old, platforms: they are seen in the BIOS of any motherboards, they can be boot disks, and no additional drivers are required for their operation in the operating system. In other words, Kingston HyperX Predator and Plextor M6e Black Edition are two of the most hassle-free PCIe SSDs.
What about the other pair of AHCI drives? With them, the situation is a little more complicated. OCZ Revodrive 350 works in the operating system through its own driver, but even so, there are no problems with making this drive bootable. The situation is worse with Samsung SM951. Although this SSD communicates with the system via the legacy AHCI protocol, it does not have its own BIOS, and therefore it must be initialized motherboard BIOS fees. Unfortunately, support for this SSD is not available in all motherboards, especially older ones. Therefore, with full confidence we can only talk about its compatibility with boards based on the latest Intel chipsets of the ninetieth and hundredth series. In other cases, it may simply not be seen by the motherboard. Of course, this does not prevent using the Samsung SM951 in an operating system where it is easily initialized by the AHCI driver, but in this case, you will have to forget about the possibility of booting from a high-speed SSD.
But the biggest inconvenience can be caused by the Intel SSD 750, which works through the new NVMe interface. Drivers that are required to support SSDs using this protocol are present only in the latest operating systems. So, in Linux, NVMe support appeared in kernel version 3.1; The "native" NVMe driver is available in Microsoft systems starting with Windows 8.1 and Windows Server 2012R2; and in OS X, compatibility with NVMe drives was added in version 10.10.3. In addition, NVMe SSD is not supported by all motherboards. In order for such drives to be used as bootable drives, the motherboard BIOS must also have the appropriate driver. However, manufacturers have built the necessary functionality only in the most latest versions firmware released for the most recent models of motherboards. So download support operating system NVMe drives are available only on the most modern enthusiast boards based on the Intel Z97, Z170 and X99 chipsets. In older and cheaper platforms, users will only be able to use NVMe SSDs as second drives in a limited set of operating systems.
Despite the fact that we tried to describe all possible combinations of platforms and PCI Express drives, the main conclusion from what has been said is that the compatibility of PCIe SSDs with motherboards is far from being as obvious as in the case of SATA SSDs. Therefore, before purchasing any high-speed SSD that works through PCI Express, be sure to check its compatibility with a specific motherboard on the manufacturer's website.
Test configuration, tools and testing methodology
Testing is carried out in the operating room Microsoft system Windows 8.1 Professional x64 with Update correctly recognizes and services modern SSDs. This means that in the process of passing the tests, as in the usual everyday use SSD, the TRIM command is supported and actively enabled. 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, unless otherwise noted, use randomized, incompressible data.Applications and tests used:
Iometer 1.1.0
Measuring the speed of sequential reading and writing data in blocks of 256 KB (the most typical block size for sequential operations in desktop tasks). Estimates of speeds are performed within a minute, after which an average is calculated.
Measurement of random read and write speed in 4 KB blocks (this block size is used in the vast majority of real operations). The test is run 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. Speeds are evaluated for three minutes, after which an average is calculated.
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. Speeds are evaluated for three minutes, after which an average is calculated.
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. Speeds are evaluated for three minutes, after which an average is calculated.
Measuring performance under a mixed multi-threaded load and establishing its dependence on the ratio between read and write operations. The test is carried out twice: for sequential reads and writes in 128 KB blocks, performed in two independent threads, and for random operations with 4 KB blocks, which are performed in four threads. In both cases, the ratio between reads and writes varies in 20 percent increments. Speeds are evaluated for three minutes, after which an average is calculated.
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 5.0.2
Synthetic benchmark showing typical SSD performance measured on a 1 GB "on top" disk area 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.
PC Mark 8 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 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.
File Copy Tests
This test measures the speed of copying directories with files different type, as well as the speed of archiving and unzipping files inside the drive. For copying, a standard Windows tool is used - the Robocopy utility, for archiving and unzipping - the 7-zip archiver version 9.22 beta. Three sets of files are involved in the tests: ISO - a set that includes several disk images with software distributions; Program - a set that is a pre-installed software package; Work is a set of work files that includes office documents, photographs and illustrations, pdf files and multimedia content. Each of the sets has a total file size of 8 GB.
A computer with a motherboard is used as a test platform. ASUS board Z97 Pro, Core processor i5-4690K with integrated graphics Intel core HD Graphics 4600 and 16 GB DDR3-2133 SDRAM. Drives with a SATA interface are connected to the SATA 6 Gb / s controller built into the motherboard chipset and operate in AHCI mode. PCI Express drives are installed in the first full-speed PCI Express 3.0 x16 slot. The drivers used are Intel Rapid Storage Technology (RST) 13.5.2.1000 and Intel Windows NVMe driver 1.2.0.1002.
The volume and speed of data transfer in benchmarks are indicated in binary units (1 KB = 1024 bytes).
In addition to the five main characters of this test - client SSDs with PCI Express interface, we added the fastest SATA SSD to the company - Samsung 850 PRO.
As a result, the list of tested models took the following form:
Intel SSD 750 400 GB (SSDPEDMW400G4, firmware 8EV10135);
Kingston HyperX Predator PCIe 480GB (SHPM2280P2H/480G, Firmware OC34L5TA);
OCZ RevoDrive 350 480 GB (RVD350-FHPX28-480G, firmware 2.50);
Plextor M6e Black Edition 256 GB (PX-256M6e-BK, firmware 1.05);
Samsung 850 Pro 256 GB (MZ-7KE256, firmware EXM01B6Q);
Samsung SM951 256 GB (MZHPV256HDGL-00000, firmware BXW2500Q).
Performance
Sequential read and write operations
The new generation of solid state drives transferred to the PCI Express bus should be the first to stand out high speeds sequential read and write. And that's exactly what we see on the graph. All PCIe SSDs outperform the best SATA SSD, the Samsung 850 PRO. However, even such simple load how sequential read and write shows huge differences between SSDs various manufacturers. Moreover, the variant of the used PCI Express bus is not of decisive importance. The best performance here can be given by the Samsung SM951 PCI Express 3.0 x4 drive, and in second place is the Kingston HyperX Predator, which works via PCI Express 2.0 x4. The progressive NVMe drive Intel SSD 750 was only in third place.
Random reads
If we talk about random reading, as you can see from the diagrams, PCIe SSDs are not particularly different in speed from traditional SATA SSDs. Moreover, this applies not only to AHCI drives, but also to the product that works with the NVMe channel. Actually better than the Samsung 850 PRO performance with random read operations on small request queues, only three participants in this test can demonstrate: Samsung SM951, Intel SSD 750 and Kingston HyperX Predator.
Although deep request queue operations for personal computers are not typical, we will still see how the performance of the SSD in question depends on the depth of the request queue when reading 4-kilobyte blocks.
The graph clearly shows how solutions that work through PCI Express 3.0 x4 can outperform all other SSDs. The curves corresponding to the Samsung SM951 and Intel SSD 750 are significantly higher than the curves of other drives. Another conclusion can be drawn from the above diagram: OCZ RevoDrive 350 is a shamefully slow solid state drive. On random read operations, it is about half behind the SATA SSD, which is due to its RAID architecture and the use of outdated second-generation SandForce controllers.
In addition to this, we suggest looking at how the speed of random reading depends on the size of the data block:
Here the picture is slightly different. As the block size grows, operations begin to look like sequential ones, so not only the architecture and power of the SSD controller, but also the bandwidth of the bus they use, begins to play a role. On larger block sizes, Samsung SM951, Intel SSD 750, and Kingston HyperX Predator provide the best performance.
Random Writes
Somewhere, the advantages of the NVMe interface, which provides low latencies, and the Intel SSD 750 controller with a high level of parallelism should have manifested themselves. In addition, the capacious DRAM buffer available in this SSD allows you to organize very efficient data caching. And as a result, the Intel SSD 750 delivers unsurpassed random write performance even when the request queue has a minimum depth.
To see more clearly what happens to random write performance as request queue depth increases, see the following graph, which shows 4K random write performance vs. request queue depth:
Scaling Intel performance SSD 750 occurs until the queue depth reaches 8 commands. This is typical behavior for consumer SSDs. What sets Intel apart, however, is that its random write speeds are significantly faster than any other SSD, including the fastest PCIe models like the Samsung SM951 or the Kingston HyperX Predator. In other words, under random write load, the Intel SSD 750 offers fundamentally better performance than any other SSD. In other words, the transition to using the NVMe interface allows you to pump up the speed of random recording. And this is certainly an important characteristic, but first of all for server drives. Actually, the Intel SSD 750 is just a close relative of such models as the Intel DC P3500, P3600 and P3700.
The following graph shows random write performance versus data block size.
As block sizes increase, the Intel SSD 750 loses its undeniable advantage. The Samsung SM951 and Kingston HyperX Predator are starting to produce approximately the same performance.
As the cost of solid-state drives is no longer used as exclusively system drives and become ordinary work drives. In such situations, the SSD receives not only a refined load in the form of writes or reads, but also mixed requests when read and write operations are initiated different applications and must be processed at the same time. However, full-duplex operation for modern SSD controllers remains a significant problem. When mixing reads and writes in the same queue, the speed of most consumer-grade SSDs sags noticeably. This was the reason for a separate study, in which we check how SSDs perform when it is necessary to process sequential operations interspersed. The next pair of charts show the most typical case for desktops, where the ratio of the number of reads and writes is 4 to 1.
Under sequential mixed loads with predominant read operations, which is typical for ordinary personal computers, the Samsung SM951 and Kingston HyperX Predator give the best performance. Random mixed load turns out to be more difficult for SSDs and leaves Samsung SM951 in the lead, but Intel SSD 750 moves to second place. At the same time, Plextor M6e Black Edition, Kingston HyperX Predator and OCZ RevoDrive 350 generally turn out to be noticeably worse than a regular SATA SSD.
The next couple of graphs give a more detailed picture of mixed load performance, showing the speed of an SSD versus the ratio of reads and writes to it.
All of the above is well confirmed in the above graphs. In a mixed workload with sequential operations, the Samsung SM951 shows the best performance, which feels like a fish in water in any work with serial data. For arbitrary mixed operations, the situation is slightly different. Both Samsung drives, both PCI Express 3.0 x4 SM951 and regular SATA 850 PRO, perform very well in this test. nice results, bypassing the performance of almost all other SSDs. Resist them in individual cases only the Intel SSD 750 can, which, thanks to the NVMe instruction system, is perfectly optimized for working with random writes. And when mixed-trade workflow rises to 80 percent or more records, it jumps ahead.
Results in CrystalDiskMark
CrystalDiskMark is a popular and simple test application that runs "on top" of the file system, which allows you to get results that are easily replicated by ordinary users. The performance figures obtained in it should complement the detailed graphs that we built based on tests in IOMeter.
These four charts are only theoretical value, showing peak performance that is not achievable in typical client tasks. A request queue depth of 32 commands never occurs on personal computers, but in special tests it allows you to get maximum performance. And in this case, the leading performance by a wide margin is given by the Intel SSD 750, which has an architecture inherited from server drives, where a large depth of the request queue is quite in the order of things.
But these four diagrams are already of practical interest - they display the performance under load, which is typical for personal computers. And here the Samsung SM951 gives out the best performance, which lags behind the Intel SSD 750 only with random 4-kilobyte writes.
PCMark 8 2.0 Real Use Cases
The Futuremark PCMark 8 2.0 test package is interesting in that it is not synthetic in nature, but on the contrary, it is based on how real applications work. During its passage, real scenarios-traces of using a disk in common desktop tasks are reproduced, and the speed of their execution is measured. Current version This test simulates a load that is taken from real-life Battlefield 3 and World of Warcraft gaming applications and software packages from Abobe and Microsoft: After Effects, Illustrator, InDesign, Photoshop, Excel, PowerPoint, and Word. The final result is calculated as the average speed that the drives show when passing the test tracks.
PCMark 8 2.0 test, which evaluates the performance of storage systems in real applications, clearly tells us that there are only two PCIe drives, the speed of which is fundamentally higher than that of the usual models with a SATA interface. These are Samsung SM951 and Intel SSD 750, which also win in many other tests. Other PCIe SSDs, such as Plextor M6e Black Edition and Kingston HyperX Predator, are more than one and a half times behind the leaders. Well, the OCZ ReveDrive 350 demonstrates frankly poor performance. It is more than twice as slow as the best PCIe SSDs and is inferior in speed even to the Samsung 850 PRO, which works via a SATA interface.
The integral result of PCMark 8 should be supplemented with performance indicators issued by flash drives when passing individual test tracks that simulate various real load scenarios. The fact is that under different loads, flash drives often behave a little differently.
Whatever application we are talking about, in any case, one of the SSDs with PCI Express 3.0 x4 interface gives the highest performance: either Samsung SM951 or Intel SSD 750. Interestingly, other PCIe SSDs in some cases generally only give speeds at the SATA SSD level. . In fact, the advantage of the same Kingston HyperX Predator and Plextor M6e Black Edition over the Samsung 850 PRO can only be seen in Adobe Photoshop, Battlefield 3 and Microsoft Word.
Copying files
Keeping in mind that solid-state drives are being introduced into personal computers more and more, we decided to add to our methodology the measurement of performance during normal file operations - when copying and working with archivers - that are performed "inside" the drive. This is a typical disk activity that occurs if the SSD does not play the role of a system drive, but a regular disk.
In the copy tests, the leaders are still the same Samsung SM951 and Intel SSD 750. However, if we are talking about large sequential files, then Kingston HyperX Predator can compete with them. I must say that with simple copying, almost all PCIe SSDs turn out to be faster Samsung 850PRO. There is only one exception - Plextor M6e Black Edition. And the OCZ RevoDrive 350, which has consistently found itself in the position of a hopeless underdog in the rest of the tests, unexpectedly bypasses not only the SATA SSD, but also the slowest PCIe SSD.
The second group of tests was carried out during archiving and unzipping the directory with working files. The fundamental difference in this case is that half of the operations are performed with disparate files, and the second half with one big file archive.
The situation is similar when working with archives. The only difference is that here Samsung SM951 manages to confidently break away from all competitors.
How TRIM and background garbage collection work
When testing various SSDs, we always check how they process the TRIM command and whether they are able to collect garbage and restore their performance without support from the operating system, that is, in a situation where the TRIM command is not transmitted. Such testing was carried out this time as well. The scheme of this test is standard: after creating a long continuous load on writing data, which leads to degradation of the write speed, we disable TRIM support and wait 15 minutes, during which the SSD can try to recover on its own due to its own garbage collection algorithm, but without outside help operating system, and measure the speed. Then the TRIM command is forcibly sent to the drive - and after a short pause, the speed is measured again.The results of such testing are shown in the following table, which for each tested model indicates whether it responds to TRIM by clearing an unused part of the flash memory and whether it can prepare clean flash memory pages for future operations if the TRIM command is not given to it. For drives that turned out to be able to carry out garbage collection without the TRIM command, we also indicated the amount of flash memory that was independently released by the SSD controller for future operations. For the case of operating the drive in an environment without TRIM support, this is just the amount of data that can be stored on the drive at a high initial speed after idle time.
Despite the fact that high-quality support for the TRIM command has become the industry standard, some manufacturers consider it acceptable to sell drives in which this command is not fully processed. Such a negative example is demonstrated by OCZ Revodrive 350. Formally, it understands TRIM, and even tries to do something when receiving this command, but there is no need to talk about a full return of the write speed to its original values. And there is nothing strange in this: the Revodrive 350 is based on SandForce controllers, which are notable for their irreversible performance degradation. Accordingly, it is also present in Revodrive 350.
All other PCIe SSDs work with TRIM just like their SATA counterparts. That is, ideally: in operating systems that issue this command to drives, performance remains at a consistently high level.
However, we want more - a high-quality drive should be able to carry out garbage collection without issuing a TRIM command. And here the Plextor M6e Black Edition stands out - a drive that is able to independently free up much more flash memory for upcoming operations than its competitors. Although, of course, offline garbage collection works to some extent on all the SSDs we tested, with the exception of the Samsung SM951. In other words, under normal use in today's environments, the performance of the Samsung SM951 will not degrade, but in cases where TRIM is not supported, this SSD is not recommended.
conclusions
We should probably start summing up by stating the fact that consumer SSDs with PCI Express interface are no longer exotic and not some kind of experimental products, but a whole market segment in which the fastest solid state drives for enthusiasts play. Naturally, this also means that there have been no problems with PCIe SSDs for a long time: they support all the functions that SATA SSDs have, but at the same time they are more productive and sometimes have some new interesting technologies.At the same time, the client PCIe SSD market is not so crowded, and so far only companies with high engineering potential have been able to enter the cohort of manufacturers of such solid state drives. This is due to the fact that independent developers of mass-produced SSD controllers do not yet have designer solutions that allow them to start producing PCIe drives with minimal engineering effort. Therefore, each of the PCIe SSDs currently on store shelves is distinctive and unique in its own way.
In this test, we were able to bring together five of the most popular and most common PCIe SSDs targeted for use in personal computers. And according to the results of acquaintance with them, it becomes clear that buyers who want to switch to using solid-state drives with a progressive interface will not face any serious choice torment yet. In most cases, the choice will be unambiguous, the tested models differ so much in their consumer qualities.
In general, the most attractive PCIe SSD model turned out to be Samsung SM951. This is a brilliant PCI Express 3.0 x4 solution from one of the market leaders, which not only proved to be able to provide the highest performance in typical general workloads, but is also significantly cheaper than all other PCIe drives.
However, the Samsung SM951 is still not perfect. Firstly, it does not contain any special technologies aimed at improving reliability, but we would still like to have them in premium-level products. Secondly, this SSD is quite difficult to find on sale in Russia - it is not supplied to our country through official channels. Fortunately, we can offer to pay attention to a good alternative - Intel SSD 750. This SSD also runs via PCI Express 3.0 x4, and is only slightly behind the Samsung SM951. But it is a direct relative of server models, and therefore has high reliability and works on the NVMe protocol, which allows it to demonstrate unsurpassed speed on random write operations.
In principle, against the background of Samsung SM951 and Intel SSD 750, other PCIe SSDs look rather weak. However, there are still situations when they will have to prefer some other PCIe SSD model. The fact is that advanced Samsung and Intel drives are only compatible with modern motherboards built on Intel's ninetieth or hundredth series chipsets. In older systems, they can only work as a “second disk”, and loading the operating system from them will be impossible. Therefore, neither Samsung SM951 nor Intel SSD 750 are suitable for upgrading platforms of previous generations, and the choice will have to be made on the drive Kingston HyperX Predator, which, on the one hand, can provide good performance, and on the other hand, is guaranteed not to have any compatibility problems with older platforms.
WiFi modules and other similar devices. The development of this bus was started by Intel in 2002. Now the non-profit organization PCI Special Interest Group is developing new versions of this bus.
At the moment, the PCI Express bus has completely replaced such obsolete buses as AGP, PCI and PCI-X. The PCI Express bus is located at the bottom of the motherboard in a horizontal position.
PCI Express is a bus that was developed from the PCI bus. The main differences between PCI Express and PCI lie at the physical level. While PCI uses a common bus, PCI Express uses a star topology. Each device is connected to a common switch with a separate connection.
The software model of PCI Express largely repeats the PCI model. Therefore, most existing PCI controllers can be easily modified to use the PCI Express bus.
PCI Express and PCI slots on the motherboard
In addition, the PCI Express bus supports new features such as:
- Hot plug devices;
- Guaranteed data exchange rate;
- Energy management;
- Control of the integrity of transmitted information;
How the PCI Express Bus Works
The PCI Express bus uses a bidirectional serial connection to connect devices. Moreover, such a connection can have one (x1) or several (x2, x4, x8, x12, x16 and x32) separate lines. The more such lines are used, the more data transfer speed the PCI Express bus can provide. Depending on the number of lines supported, the sort size on the motherboard will differ. There are slots with one (x1), four (x4) and sixteen (x16) lines.
Visual demonstration of the dimensions of the PCI Express slot
At the same time, any PCI Express device can work in any slot if the slot has the same or more lanes. This allows you to install a PCI Express card with a x1 slot in a x16 slot on the motherboard.
The throughput of PCI Express depends on the number of lanes and the bus version.
|
One way / two ways in Gbps |
|||||||
|
Number of lines |
|||||||
| PCIe 1.0 | 2/4 | 4/8 | 8/16 | 16/32 | 24/48 | 32/64 | 64/128 |
| PCIe 2.0 | 4/8 | 8/16 | 16/32 | 32/64 | 48/96 | 64/128 | 128/256 |
| PCIe 3.0 | 8/16 | 16/32 | 32/64 | 64/128 | 96/192 | 128/256 | 256/512 |
| PCIe 4.0 | 16/32 | 32/64 | 64/128 | 128/256 | 192/384 | 256/512 | 512/1024 |
Examples of PCI Express Devices
First of all, PCI Express is used to connect discrete graphics cards. Since the advent of this bus, absolutely all video cards use it.
video card GIGABYTE GeForce GTX 770
However, this is not all that the PCI Express bus can do. It is used by manufacturers of other components.
SUS Xonar DX sound card
OCZ Z-Drive R4 Enterprise SSD