That the (backwards compatible) NVMe SSDs with PCIe 4.0 will slowly but surely push the drives with PCIe 3.0 out of the market is only a question of time and the price that you will then have to pay for simpler models. That’s where today’s article comes in and preempts the review of the brand new MSI Spatium 450, which I’ll be publishing shortly. Already during the teardown and the then exposed components, I decided in the course of the research that it would be better to make a separate article from the gained knowledge in advance, because the topic is quite interesting and you also have to be sensitized for the advantages and disadvantages of such solutions as a buyer.
Certain additional costs still stand in the way of widespread use. However, in order to be able to reduce the costs not only marginally but significantly, the controller and the NAND flash have to be made cheaper, the DRAM has to be removed and the boards have to be made much simpler overall. Thus, there are three points that have to lead to cost savings after the cheaper QLC storage has proven to be a kind of technical dead end. Too susceptible and unstable, combined with a rather questionable long-term durability, the QLC modules are certainly still somewhat suitable as a data grave, but not for anything more. To use something like this as a system drive would also be to perform an execution in installments. However, as a customer, you can’t do anything with that either.
Of course, the theoretically possible data rates of PCIe 4.0 can also be used for so-called flagship solutions with significantly higher performance (with up to 7000 MB/s and more), but that is not the issue today. Because the way for the broad mass of users will probably be to replace the (expensive) top models of the PCIe 3.0 era with “butter-and-bread” models with PCIe 4.0, which are cheaper and yet (with certain restrictions) also faster and do not rely on the unloved QLC memory.
Abandonment of the DRAM buffer
Buffer-free products like the Samsung 980 or the WD Blue SN550 have been on the market for months and show that it is possible to do without DRAM with certain restrictions. However, the PCIe 4.0 bus with increased bandwidth, which should theoretically let the SSD controller work more efficiently with the PC’s RAM, does not always help. This is because it is also important to remember that the lack of a dedicated DRAM forces the controller to use only HMB (Host Memory Buffer) technology to work with the address translation table.
Unfortunately, this also means that the drive will fork a certain part of the PC’s memory (up to 64 MB) for its own purposes during operation in order to buffer calls to the flash memory array. As a result, the SSD’s speed decreases like an avalanche under high load in certain scenarios, which especially affects the read process very strongly. If you have to process more than 32 GB of information in one go, it can get quite tight. The drop in random reading can be more than half, or it can be mitigated or circumvented by tricks in the controller. However, it is a fact that an unbuffered SSD is rather poorly suited for many intensive workloads. You have to scrutinize yourself, the area of application and the applications used.
But we are still in the consumer sector. For a better understanding, I have created a more practical load here in the graphic by reading 64 GB of video stream. Well, the theoretically possible (and also measured with CrystalDiskMark) 3700 MB/s are definitely not possible, but with around 3000 MB/s you are still quite well in the race despite the missing buffer. So, the effect is not that bad as long as you really use PCIe 4.0 and thus also connect the involved RAM fast enough. Even though the SSDs are compatible, the PCIe 3.0 then shrinks the whole thing almost to half.
22 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Urgestein
Mitglied
Urgestein
1
Mitglied
Mitglied
1
Veteran
Moderator
Mitglied
1
1
Moderator
Veteran
1
Mitglied
Mitglied
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →