Until now, the evolution of solid-state drives has been based on improving storage capacity and access speed. However, in the years to come, thanks to the use of CXL technology in SSDswe are going to see how the latter will completely change the way in which applications and operating systems will use our computers.
One of the most interesting things that was seen during this year’s Hot Chips was the presentation of a Samsung NVMe-type solid state drive with the capabilities that we are going to discuss below. Said unit at the moment is for servers or what today are known as data centers. At the moment the CXL or Compute Express Link is not implemented in any processor on the market, but its advantages are such that there is no doubt that its adoption will be massive in many systems in the future.
RAM and storage at the same level
The CXL protocol, Compute Express Linkis in principle a PCI Express 5.0 extension which, among other things, has the ability to all devices connected to that port are fully consistent in memory access. This means that any changes made to a part of the system’s global memory, DRAM, or local memory, such as the graphics card’s VRAM, will be seen by all devices that are connected to that port.
The main advantage of this is the fact that stop having to explicitly code copy processes from one device to another. Although at the level of hardware operation this will not change. Since we are physically talking about separate memories, so it will continue to function as a non-unified memory architecture, but the big difference is to make the work of application programmers easier and one of the components that will benefit most from these future changes is NVMe SSD storage, especially with the advent of PCIe Gen 5 compatible drives.
The future of NVMe SSDs lies with CXL
However, the CXL protocol in NVMe SSDs goes beyond this, as you can see in the graph above, and the access latency is much lower than a conventional drive. This will remind many of what Intel promised with Optane, but this time we are dealing with a non-proprietary standard that does not require the use of special flash memory chips, only an update to the computer’s PCIe interface.
Now, let’s review what M.2-type SSDs are, they usually have their corresponding flash memory chips where the data is stored, the flash controller that is nothing more than a memory ditto with increased capacities and finally one or two DRAM memory chips, which can be DDR or LPDDR. As well, thanks to the CXL said memory would become part of the system RAM. In other words, The process of copying to the system memory would not be necessary and both the processor and the graphics card can access directlywithout waiting for the data to be copied into their corresponding work memories.
The big drawback
However, we cannot forget that the fact of accessing apart from the RAM memory through the PCI Express port, and even if we do it consistently, adds latency to the equation. That is, the data and instructions that are executed from the memory of the SSD will not be executed with the same ease. However, it should be clarified that the CPU will continue to access RAM through its conventional mechanism, precisely the CXL is not based on the processor accessing everything by taking that path.
And, on the other hand, and to finish, in the vast majority of cases it is much faster to access the SSD RAM directly than not having to carry out the entire process of copying from one memory to another, as is currently the case.