Why do I need a PCI Express 4.0 SSD? Explain the example of Seagate FireCuda 520

Today we want to talk about one of our new products - the Seagate FireCuda 520 SSD-drive. But do not rush to flip the tape further with the thoughts “well, another laudatory review of the gadget from the brand” - we tried to make the material useful and interesting. Under the cut, we primarily focus not on the device itself, but on the PCIe 4.0 interface, which is used in it. And we will tell you what to expect from him, how good he is and to whom it can be potentially useful.



Let's be honest: the PCI Express 4.0 standard is not such a novelty. The first devices with its support appeared on the consumer market as early as last summer. Thank you for saying this to AMD: it was she who created the first platforms that are capable of accepting devices with PCI Express 4.0, and also made such devices herself - these are graphic cards based on GPUs with RDNA architecture.

The increase in throughput always gives rise to great expectations, but, as it turned out, video cards almost do not benefit from switching to a faster interface. At least when it comes to game loads. As numerous independent tests have shown, even the fastest cards supporting PCI Express 4.0, primarily the Radeon RX 5700 XT, work the same way when using a new and fast interface, or when connected to the classic PCI Express 3.0 bus.

But with SSDs is a completely different matter. The performance of productive NVMe SSDs working through PCI Express 3.0 (for example, Seagate FireCuda 510), under linear loads, clearly depends on the interface bandwidth. Therefore, the expansion of the bandwidth is simply obliged to positively affect the capabilities of the new generation disk subsystems.

A good illustration of the fact that there is little bandwidth is the fact that while we are talking about the first devices with support for PCI Express 4.0, the PCI Special Interest Group (PCI-SIG) has already approved the PCI Express 5.0 specification, which takes one more step in the direction of increasing the speeds of the interfaces through which modern processors communicate with external devices. But about it somehow another time, today it is PCI Express 4.0 that is on the agenda.

What good is PCI Express 4.0?


The PCIe (Peripheral Component Interconnect Express) specification standardizes how expansion cards, such as graphics cards, sound controllers, network adapters, and finally NVMe SSDs, communicate with the basic components that make up the PC platform. The higher the PCIe specification, the higher the bandwidth it provides. In addition, when it comes to PCIe slots, in addition to the specification version, they also talk about the number of lines, which is denoted as x1, x2, x4, x8 or x16. A larger number of lines also gives a multiple higher throughput due to bus expansion and represents another, extensive way to improve the speed characteristics of the interface. But if we talk about NVMe SSDs, then this approach is difficult to apply in them. Available in the compact form factor M.2 SSDs for PCs can use two or a maximum of four lines, while support for up to 16 lines can be allocated only full-size cards for PCIe slots. For this reason, the introduction of new versions of the PCIe standard is considered a key event for the market for productive SSDs.

All versions of the PCIe specification are backward compatible. Drives oriented to PCIe 4.0 can also work on platforms where only PCIe 3.0 is supported, and motherboards with PCIe 4.0 slots can easily install components that work in accordance with the PCIe 3.0 standard. However, in either case, the system will work with PCIe 3.0 speeds - the younger version of the standard, which is supported on both sides.

The main innovation inherent in PCIe 4.0 is the doubled bandwidth of one line. There are different options for numerical estimates of the changes, but if we talk about theoretical and peak values, the PCIe 4.0 specification assumes a maximum transfer speed of 1.97 GB / s on one line in each direction, while in PCIe 3.0 the maximum speed was limited by 0.98 GB / s In some sources, you may find twice as high indicators, but this is due to the fact that they indicate the total data transfer rate in both directions.



As we said above, such an increase in interface speed in practice is not very useful (or rather, almost completely useless) for graphic cards. At the same time, NVMe-drives working through four PCIe lanes get the opportunity to pump up to 7.88 GB / s on a bus from four lanes (in the ideal case), which opens up wide scope for them to improve performance.

In addition to increasing throughput, the PCIe 4.0 standard also offers other innovations. For example, it contains new opportunities to reduce power consumption, as well as more extensive functions for device virtualization. But the main direction in which the developers moved was still the increase in speeds, and almost everything was done primarily for its sake. For example, a number of improvements in the new version of the interface are aimed at improving the integrity of signals and the reliability of their transmission. In other words, for most consumers, PCIe 4.0 implies higher throughput and nothing more.

What is the platform with support for PCI Express 4.0?


Unfortunately, despite the fact that the PCI Express 4.0 specification itself was approved back in 2017, there are still not many real platforms with its support on the market. This means that if you want to use a new generation of high-performance solid-state drives, you will have to take care not only to find such a drive, but also to select a platform that can fully reveal its potential.

The fact is that only AMD has supported the new PCIe 4.0 interface so far, and even that is only fragmentary. It is implemented in terms of its processors based on the Zen 2 architecture, and more specifically, in the desktop Ryzen 3000 series and in the high-performance Threadripper 3000 series, but, for example, not in the Ryzen 4000 mobile series. Moreover, if PCIe 4.0 is supported in any Socket sTR4 motherboard for the third-generation Threadripper, Ryzen 3000 processors will only be able to interact with PCIe 4.0 peripherals in full speed mode in motherboards built on the basis of the X570 logic set, where signal lines are designed taking into account increased requirements for shielding and minimization of electrical noise.



The good news here is that potential Ryzen 3000 owners will soon be able to get at their disposal another class of more affordable motherboards with support for PCIe 4.0 video cards and drives. They will be built on the new B550 chipset, which is due out in the next couple of months.

As for Intel platforms, they do not yet support PCIe 4.0 at all. Moreover, the coming soon Comet Lake-S desktop processors, which will bring with them the new LGA 1200 processor socket, and the new chipsets of the four-hundredth series, PCIe 4.0 will not receive either. If we talk about the mass desktop systems of Intel, then support for this interface may appear only with the release of Rocket Lake processors, but this will happen around the beginning of next year. But this interface may get into mobile systems earlier: plans for PCIe 4.0 are announced for Tiger Lake processors, the formal announcement of which may take place this summer. In addition, it cannot be ruled out that high-performance desktops of the HEDT class will switch to PCIe 4.0 also this year: it will become possible,if Intel decides to offer Ice Lake-X in this segment - the analogs of server-based Ice Lake-SP.

As a result, despite the fact that PCIe 4.0 will be widely used in the medium term, right now there are not so many options for fast NVMe SSD supporters to choose a platform. The most obvious of these is the Socket AM4 system based on the Ryzen 3000 processor and the X570-based motherboard.

What about PCI Express 4.0 drives?


If you look at the assortment of NVMe SSDs with PCIe 4.0 support, which is presented on store shelves, you might get the feeling that the market is full of various options for high-speed solutions of the new generation. However, in reality this impression is misleading. Despite the fact that the PCIe 4.0 specification has existed for several years, hardware developers have not yet managed to bring a sufficient number of alternatives to the mass production stage.

The only controller that SSD manufacturers can now use for their products is the Phison PS5016-E16. Moreover, in reality, this controller cannot be called a full-fledged development of a new generation. It is rather a transitional solution based on another, earlier chip PS5012-E12, in which the function block responsible for the external bus was simply replaced.

For the end user, this means two things. First of all, all PCIe 4.0-enabled NVMe drives on the market are not too different from each other, at least when it comes to performance. And if you see that for a certain product higher passport speeds are suddenly announced, this is most likely due to the cunning of marketers, and not to any real advantages, because in the end, the same controller is used in both products. Secondly, today's PCIe 4.0 drives cannot yet boast of utilizing the full bandwidth of the new bus - the maximum speeds promised by the Phison PS5016-E16 chip are at 5 GB / s for linear reading and 4.4 GB / s for records.



An important consequence follows from the foregoing: in the future, NVMe SSDs can make another performance leap even without switching to the next version of the PCI Express specification. It is only necessary to wait for the appearance of newer controllers with a redesigned core adapted for the capabilities of PCIe 4.0. And such solutions are already being developed. The appearance of such a product is at least expected from Samsung, in addition, independent engineering teams work on more advanced controllers: Phison (PS5018-E18), Silicon Motion (SM2267), Marvell (88SS1321) and even not too famous Innogrit (IG5236).

The only trouble is that all this splendor can appear very soon. The development of controllers is a lengthy process, and serious delays often occur in the final stages - when preparing the firmware or during validation. In addition, the coronavirus pandemic has had a huge impact on the entire industry, which is why new product releases have been postponed to a later date.

In other words, you can wait for something better for a long time, and if a higher performance disk subsystem is needed now, it makes sense to dwell on what is already there - the drives on the Phison PS5016-E16 controller. Although they don’t choose the full bandwidth of the four PCIe 4.0 lanes, they can boast quite good performance in small-block operations, which, according to the developers, reaches 750 thousand IOPS. This is ensured both by the controller design, which is based on the dual-core 32-bit ARM Cortex R5 processor, and a set of proprietary tricks: dynamic SLC caching and CoXProcessor 2.0 technology - hardware acceleration of typical operation chains.

Why is Seagate FireCuda 520?


It was said above that all existing consumer NVMe drives with PCIe 4.0 support are built on the same foundation - the Phison PS5016-E16 controller. However, this does not mean that taking the first SSD for the PCIe 4.0 bus in the store is a good idea. Here, we would recommend paying attention to the Seagate FireCuda 520, but not at all because you are reading this article on the Seagate corporate blog.



The devil is in the details and, if you start to understand, the Seagate FireCuda 520 may be more attractive than many alternatives on the same Phison PS5016-E16 chip. There are several reasons for this, but they all boil down to one thing - the flash memory installed in FireCuda 520.

Formally, all drives with the Phison PS5016-E16 controller use the same flash memory: 96-layer BiCS4 (TLC 3D NAND) manufactured by Kioxia (formerly Toshiba Memory). However, in fact, this memory may vary. Depending on what priorities a particular manufacturer has chosen for itself, memory can relate to completely different gradations of quality. For example, in the products of third-tier companies, there is often a flash memory of a “media” purpose, which, generally speaking, is intended for flash drives and memory cards, but not for SSDs.

With Seagate drives, this is completely out of the question. The company does not buy flash memory on the open market, but has a long-term direct contract with Kioxia, which was concluded at the time when Toshiba got rid of the production of memory. Thanks to this, we get NAND chips, as they say, first-hand and have access to the best quality silicon.

This is inevitably reflected in the reliability parameters. Representatives of the Seagate FireCuda 520 series are equipped with a five-year warranty, and the installed resource allows you to overwrite the full capacity of the drive 1800 times, that is, on average once a day. These are very high endurance indicators, according to which the Seagate offer, for example, is three times higher than the popular Samsung 970 EVO Plus.

And then it's time to show how the Seagate FireCuda 520 looks from the outside. This is an M.2-board of the traditional form factor 2280 with microcircuits located on both sides of it.



There are no special cooling facilities that other manufacturers like to pile on their drives, due to the fact that almost one hundred percent of motherboards with PCIe 4.0 support have their own cooling systems for M.2 slots.

Otherwise, the drive is similar to other products based on the Phison PS5016-E16 controller, but with a noticeable difference - the controller chip is labeled Seagate. This is due to the fact that the controllers for the FireCuda 520 were also purchased not on the open market, but made on special order. However, for the end user, this does not mean much, but what really matters is the use of modified firmware, which contains certain optimizations that distinguish the Seagate drive from other SSDs with similar hardware.



It is clear that firmware can hardly somehow significantly change the speed characteristics of the controller, nevertheless, it allows something. For example, FireCuda 520 boasts the implementation of dynamic SLC caching, while the drives on Phison controllers, which were released earlier, used a static SLC cache of a rather limited volume. A new approach allows you to record on FireCuda 520 with high speed much larger amounts of information.

It works very simply: any data coming to the drive is written to the TLC flash memory in a very fast single-bit SLC mode. The cells used in this way are transferred to the TLC state either later, when the user no longer accesses the drive, or as necessary, if the pool of empty cells is exhausted during the recording process. In other words, a third of the free space on FireCuda 520 can be continuously filled at maximum speed, then productivity will decrease. But it’s worth a little wait, as a third of the remaining free space can again be used in high-speed mode.

Here, for example, looks like a linear recording graph on a clean one on FireCuda 520 with a capacity of 2 TB.



The first 667 GB is recorded at a speed of 4.1 GB / s, then the speed drops drastically to 0.53 GB / s, but you should understand that during normal use of the drive you will not encounter such behavior - for this you need a long and continuous record huge amounts of information.

In addition to firmware, FireCuda 520 is also interesting with bundled software. The proprietary utility SeaTools SSD is much more convenient for monitoring the status of SSDs than third-party programs. In addition, it allows you to update firmware, test performance and perform some additional operations, such as advanced diagnostics or Secure Erase.





It is also worth mentioning that FireCuda 520 owners can download the DiscWizard program from Seagate website for smooth migration from past disk drives with the transfer of all data and the operating system.

So is it really fast?


It remains to back up everything said about the advantages of the PCI Express 4.0 interface and the drive with its support with some practical results. And with this there is no particular difficulty, because the FireCuda 520 really has a noticeably higher performance, which is not available for previous generation drives. Despite the fact that there are well-grounded claims against the Phison PS5016-E16 controller related to the fact that it still does not utilize the full PCIe 4.0 bandwidth, the Seagate FireCuda 520 speed indicators are obviously higher than those for PCIe 3.0 drives.

In the following table, the characteristics of the Seagate FireCuda 520 are compared with the characteristics of the FireCuda 510 - Seagate's previous flagship NVMe SSD model, which is designed for the PCIe 3.0 x4 interface. For example, the comparison is limited to the most capacious and high-speed SSD options with a capacity of 2 TB, but if you compare the modifications of other capacities with each other, the picture will be approximately the same.



However, passport characteristics are one thing, but real life is another. Therefore, we just took these two drives - FireCuda 520 2 TB and FireCuda 510 2 TB - and compared in tests.

FireCuda 520 2 TB

FireCuda 510 2 TB

CrystalDiskMark results require some comment. The new PCIe 4.0 SSD turned out to be noticeably faster than its predecessor in linear speeds: the advantage reaches almost one and a half times the size and can be traced both in deep and in minimal queues of requests. FireCuda 520 surpasses the previous version of the NVMe SSD Seagate even in small-block operations, although there is no such impressive breakthrough here: it all comes down to the fact that the controller logic remains old. Thus, the FireCuda 520 will shine primarily with successive loads. As for operations with arbitrary blocks of small size, the PCI Express 4.0 interface, of course, cannot do anything similar to Optane from a flash drive.

But the fact that high-speed linear operations are a very powerful trump card FireCuda 520, it is impossible to deny. This can be seen in more detail in the results of ATTO Disk Benchmark: as soon as the blocks exchanged for data acquire a volume of 128 Kbytes or more, it is impossible to keep up with the FireCuda 520 even in theory (even Optane is not capable of this), because the data exchange speeds go beyond set by the bandwidth of the PCIe 3.0 x4 interface.

FireCuda 520 2 TB

FireCuda 510 2 TB

In synthetic tests everything turns out more than convincingly, but what in real life? PCMark 10 can answer this question - it has scripts that reproduce the typical load on the drives during the user's daily work.

And in this case, the FireCuda 520 is faster than its predecessor by up to 30%. Moreover, this advantage is expressed not only in the increase in the speed of disk operations, but also in a noticeable decrease in the reaction time of the disk subsystem. This pattern can be seen when using SSD as the only and universal drive (see Full System Drive Benchmark). And in the case when the SSD plays the role of exclusively the system drive on which the OS and software are installed (see Quick System Drive Benchmark). And even when the SSD is placed under a “file wash” (see Data Drive Benchmark), although this, frankly, happens very infrequently.





The speed advantages of the FireCuda 520 are easy to track with regular file copying. The diagram below shows the results of the DiskBench test when copying a working directory with different files with a total volume of about 20 GB inside the drive. Of course, such an increase as in synthetic tests is not observed here, but the transition to PCIe 4.0 gives its additional 25-30% to the performance without question.



For a change, you can also look at how much faster the PCIe 4.0 drive allows you to download gaming applications. For example, below is the loading time of the level in Final Fantasy XIV StormBlood (the choice of this game is due to the convenient monitoring tools built into it). Here, the gain that the FireCuda 520 provides against the background of the FireCuda 510 is a second or so, which is not so significant, but still noticeable.



But under the loads inherent to workstations, PCI Express 4.0, as they say, must have. The fact is that computers aimed at professional content creation are equipped with very powerful multi-core processors and fast memory. And in this case, bottlenecks in the system can easily arise in the disk subsystem. For example, many video professionals used to prefer RAID arrays from SSDs, but now they can satisfy their needs by choosing FireCuda 520, which can receive data at speeds of over 4 GB / s alone.

All these considerations can easily be supported by the results of the SPECworkstation 3 test, which very clearly shows the significance of a drive with a modern interface: FireCuda 520 copes with heavy professional disk load scenarios on average 22% faster compared to FireCuda 510.



But special attention should be paid to the performance of General Operation (the usual speed of working with files during archiving and copying, as well as during software development) and Product Development (shows the speed of work in CAD / CAM systems and when solving problems of computational fluid dynamics). Here, the potential inherent in FireCuda 520 is revealed especially convincingly.

Summary


The above examples are enough to doubt that PCIe 4.0-drives really allow you to get higher performance and better responsiveness when solving resource-intensive tasks. Therefore, when building a high-performance system on AMD Ryzen 3000 or Threadripper 3000 multi-core processors, the use of the most modern NVMe SSDs should not be neglected. Seagate FireCuda 520 may be the right choice: there’s absolutely nothing faster in stores at the moment.



Naturally, a PCIe 4.0 drive will cost a little more than the FireCuda 510, but the reasons for this are well understood. And most importantly, the price of the FireCuda 520 is quite marketable, because this SSD costs almost the same as alternative PCIe 4.0 drives from third-tier manufacturers.

: Ryzen 9 3900X, ASRock X570 Creator 16 DDR4-3200 SDRAM (16-16-16-32). Windows 10 Professional 1909 NVMe- Standard NVM Express Controller 10.0.18362.1.

All Articles