The newest is not always better. Recently, SSD manufacturers have begun to compromise between speed and reliability to incorporate more storage space into their drives. Protocols like NVMe and PCIe become faster, but some SSDs go backwards.
QLC Flash is the problem
Here is the problem. Making SSDs is expensive, and few people want to pay $ 200 for a 512 GB SSD while you can get "2000 GB" mechanical hard drives for less than $ 50. Larger capacities are selling.
SSD drive manufacturers are increasing storage capacity while keeping costs down, but it hurts performance and endurance. Large SSDs may be cheaper, but every leap in SSD technology is a compromise. We are currently seeing the rise of QLC (Quad Level Cell) SSDs, capable of storing 4 bits of information per memory cell. QLC has not completely replaced the standard SSDs, but some using it are now on the market and are causing problems.
Specifically, SSD manufacturers need to find a way to incorporate more space into NAND flash chips of the same size (the data storage part of the SSD). Traditionally, this was done with a retractable process node, thus reducing the transistors inside the flash. But as Moore's law slows down, you have to be more creative.
The ingenious solution is the multi-level NAND flash. NAND flash memory is capable of storing a specific voltage level in a cell for an extended period of time. Traditional NAND flash memory stores two levels: on and off. This is called the SLC flash, and it's very fast. But since NAND essentially stores an analog voltage, you can represent several bits with slightly different voltage levels, like this:
The problem, as shown below, is that it is evolving exponentially. The SLC flash requires only a voltage or no voltage. The MLC flash requires four levels of voltage. TLC needs eight. And over the past year, the QLC flash has made a breakthrough on the market, requiring 16 distinct voltage levels.
This poses a lot of problems. As you add voltage levels, it becomes more and more difficult to distinguish the bits. This makes the QLC flash 25% denser than the CCM but significantly slower. The reading speed is not too much affected, but the writing speed takes a plunge. Most SSDs (using the latest NVMe protocol) oscillate around 1500 MB / s for continuous reading and writing (ie, loading or copying large files). But the QLC flash only manages between 80-160 MB / s for sustained writings, which is worse than a decent hard drive.
QLC SSDs break down much faster
All SSDs generally have poor write endurance compared to hard drives. Whenever you write in a cell of an SSD drive, it will slowly use it. Erasing a cell is supposed to rid it of electrons, but a few still remain around, so that a "0" cell gets closer to "1" over time. time. The controller compensates for this by applying a more positive voltage over time, which is fine if you have plenty of room for voltage. But QLC does not do it.
SLC has an average write an endurance of 100,000 program cycles / erasure (write operations). The MLC has between 35,000 and 10,000. TLC has about 5,000. But QLC has only a measly 1,000. This makes QLC unusable for frequent-access readers, such as your boot drive, which are written very frequently.
The end result: Do not buy a QLC reader to use for your system's system drive. They are far too unreliable to be sure that it will not deteriorate in a few years. We recommend that you use a large QLC disk as a replacement for a rotating hard disk and use a fast SLC, MLC or TLC disk as your operating system's primary disk. . This can be a problem in laptops where you do not have the option, but QLC is still very new and has not made its way into laptops yet.
Effective caching masks these problems
At this point, you may be wondering why the QLC is even a thing while it is objectively slower and breaks much faster than other types of flash. You obviously can not market a decommissioning, but the SDD manufacturers have found a way to hide the problem: caching.
QLC SSDs dedicate part of the drive to a cache. This cache ignores the fact that it is supposed to be QLC and works instead as the SLC flash. The cache will be 75% smaller than the actual disk space it takes, but it will be much faster.
Cache data can be written at the same speed as other high-end SSDs. They will then be slowly emptied by the controller and sorted into the QLC cells. But when this cache is full, the controller must write directly into the slow QLC cells, which results in a considerable drop in performance during long writes.
Take a look at this Tom's Hardware reference review of the Crucial P1 500GB, a QLC consumer SSD, which shows very clearly this problem:
The red line representing the Crucial P1 runs at optimal NVMe speeds, although a little slow compared to some high-end offers. But after about 75 GB of writing, the cache is full and you can see the actual speed of the QLC flash. The line drops to about 80 MB / s, slower than most hard drives for sustained writing.
The ADATA XPG SX8200, a TLC player, displays the same features, with the exception of the TLC raw flash after the release, it is even faster. Most other readers also use this caching method because it speeds up fast writes to the drive (which are the most common). But what you will notice the most is the durability of the writes: you will not notice if a small file copy takes 0.15 seconds against 0.21 seconds, but you will notice that if a large copy takes 10 minutes additional.
This could easily be considered a marginal scenario, but this cache does not always stay at 75 GB. As you fill the disk, the cache becomes smaller. according to Anandtech testsFor the Intel SSD 660p family, the 512GB model cache is reduced to just 6GB when the drive is nearly full, even with 128GB of available space.
This means that if you fill your SSD, then try to install a game of 20-30 GB of Steam, the first 6 GB would write on the drive very quickly, then you would start to see the same speeds of 80 MB / s the remaining files.
Admittedly, in this example, you are probably limited by the download speed, but in the case of updates (which must download and then replace the existing files, which effectively requires twice as much space), the problem would be much more apparent. You have finished downloading and must wait indefinitely for the installation.
So, should you avoid QLC?
You must absolutely avoid QLC disks with 512 GB (and less, once it becomes cheaper to produce) because they do not make much sense. You will fill them much faster and the cache will be smaller when it is full, which will slow it down considerably. In addition, they are currently not much cheaper alternatives.
Despite its weaknesses, the QLC flash is not really a problem when you examine the larger drives. The 2TB model of the 660p has a minimum of 24GB of cache once filled. It's still the QLC flash, but it's an acceptable compromise for a cheap 2 TB SSD that works very fast most of the time.
Because of their huge capacity, QLC-based SSDs can properly replace a rotating hard drive, as long as you perform regular backups in case it deprives you. This is great for something you rarely access but want to be very fast when you do it, and with a decent sized SLC cache, most write-intensive operations will be reasonably fast until you fill the disc.
Due to reliability issues, avoid using it as a boot drive or for anything that is written very often.
Much more needs to be done in other aspects of manufacturing: better controllers that can handle more flash chips, cheaper flash chips as process nodes mature, and perhaps even other technologies. The QLC flash does not become the norm anytime soon; currently, it's just another option. Just make sure that when buying a SSD, you check the technical specifications and that you pay close attention to the type of flash used to make them.