New generation of working memory – DDR5 represents a big step forward

The higher memory speeds are much needed to provide sufficient memory bandwidth for the rapidly growing number of processor cores. Since the beginning of 2017, the number of cores in processors has increased significantly in both the mainstream desktop market and the high-end desktop market. Processor manufacturer AMD then brought a larger number of powerful cores to various segments with its Ryzen CPUs, which competitor Intel followed step by step. At that time, DDR4 was already the mainstream working memory and it is to this day.

The officially supported speeds of DDR4 are higher than a few years ago due to the processor families that have been released since then, but the bottom line is that the CPU cores had to make do with less and less bandwidth. At the beginning of 2017, the number of cores and threads in a high-end consumer processor was 4 and 8 respectively; less than three years later, that was already 16 cores and 32 threads. Against that quadrupling in the number of cores and threads, the most optimistic comparison of DDR4-2133 to 3200MT/s, a 50 percent increase, is relatively small.

In the segment of high-end desktops and the server market, the number of cores has also increased much faster than the memory bandwidth in recent years, with a growth from 28 to 64 cores per socket. Thanks to clever use of also considerably increased cache sizes, the CPU cores could be provided with more data, but nowadays that is no longer enough to allow the large number of processor cores to use their full potential in all scenarios.

Not only processors with a large number of CPU cores can benefit from more memory bandwidth. Integrated GPUs also generally perform better when faster memory is used. It also only makes sense for processor manufacturers to place more GPU calculation cores on processors when more bandwidth is available, so with the arrival of DDR5 we can also expect faster Ryzen APUs.

Prefetch buffer

One of the main differences in DDR5 is the doubled prefetch buffer compared to DDR4. The prefetch concerns the number of data words processed per clock cycle. With DDR4 and DDR3 these are 8 bit, with DDR5 this has increased to 16 bit. One of the advantages of the higher prefetch is that the internal clock speed of the memory can remain limited, so that no higher demands are placed on the memory chips and the energy consumption does not increase. For example, the memory works more efficiently because more memory can be read or written per clock tick. Following this is the burst lengthalso doubled, allowing a larger portion of a row and column in each memory cell to be read per clock tick. The result of these and other improvements is that DDR5 should provide around 36 percent more effective memory bandwidth than DDR4 at the same memory speed.

Memory capacity: 128GB per dimm

Where the maximum capacity of DDR4 modules is 32GB due to the upper limit of 16Gbit chips, according to the specification of DDR5 this is four times higher with the new generation. With 64Gbit chips, 128GB can be placed per dimm, although memory modules with this capacity will not appear at the beginning of the DDR5 life cycle. In the long term, server memory should even reach 2TB per module thanks to die-stacking ; when exactly is still unclear.

Always dual channel

With the working memory that is now in your PC, such as DDR4, it is common for the processor to use each dimm as one 64-bit data channel, so that with two memory modules in a dual-channel setup you effectively have a 128-bit memory bus to your CPU. On DDR5, two 32-bit data channels are used on each dimm, which can work independently of each other and thus effectively function as a dual channel per dimm. These two channels also have their own error-correcting code , each of which ends up at 8bit (7+1). That is a lot less than the 24bit bus used for this on DDR4, which results in more complex DDR5 memory controllers.

Lower voltage and power management

According to the official specifications, DDR5 works with an I/O voltage of 1.1 volts. That is slightly lower than the 1.2 volts specified for DDR4, which can result in lower energy consumption with DDR5, among other things. In practice, however, faster DDR5 modules will operate at higher voltages, just as higher clocked DDR4 memory typically operates at 1.35 volts.

Another and perhaps more important adjustment compared to DDR4 is that the new generation uses pmics, full power management integrated circuits. On DDR5 memory, voltage regulation will be handled on the module itself. Each memory module receives a voltage of 12 volts from the motherboard, which is converted by the pmic to the correct voltage for the components on the memory module. That means motherboards with DDR5 memory slots get a little easier as far as power is concerned. There is no need to place a vrm there that is over-specified on a maximum number of modules at high voltage. DDR5 handles this with the pmic itself per memory stick. As a result, the complexity of design and manufacturing increases compared to previous generations of memory. As a result, DDR5 memory will be more expensive to manufacture,

Appearance and layout

DDR5 memory modules, like DDR4, will have 288 contact points. However, the modules of these two generations are not interchangeable, not only because these contact points have different functions on the new generation, but also because the recess at the bottom is in a slightly different place. As with previous generations of RAM, this prevents users from combining incompatible hardware.