DDR5 doubled the bandwidth and moved power regulation onto the module. Whether that is worth paying for depends entirely on which platform you are building.
DDR5 has been shipping since late 2021, and by 2026 it is the default on every current desktop platform. DDR4 has not disappeared — it has become something stranger, a mature product whose price is now driven by scarcity rather than by competition. Understanding the difference matters, because the two are not interchangeable: the notch position differs, and no motherboard accepts both.
The headline figure is bandwidth. A DDR4-3200 module moves 25.6 GB/s. A DDR5-6000 module moves 48 GB/s, and the fastest kits on the market push past 64 GB/s. That is a real doubling, and it is the number every product page leads with.
Three less-advertised changes matter more in practice.
Two independent channels per module. A DDR4 DIMM presents one 64-bit channel. A DDR5 DIMM presents two 32-bit channels. The total width is the same, but two narrower channels handle scattered, unpredictable access patterns far better than one wide one — which is most of what a desktop actually does.
On-module power regulation. DDR4 modules take regulated voltage from the motherboard. DDR5 modules carry their own voltage regulator. This makes the modules more expensive and moves a failure point onto the DIMM, but it means memory voltage is no longer hostage to the quality of the board's VRM.
On-die ECC. Every DDR5 module corrects single-bit errors within the chip. This is not the same as the full ECC a server uses — it does not protect the path between module and CPU, and it does not report to the operating system. It exists because DDR5 cell density made uncorrected error rates unacceptable.
The first thing anyone notices comparing spec sheets is that CAS latency went up. DDR4-3200 CL16 against DDR5-6000 CL30 looks like a regression.
It is not, because CAS latency is measured in clock cycles, not time. The conversion is straightforward:
True latency (ns) = CL × 2000 / data rate
DDR4-3200 CL16 gives 10.0 ns. DDR5-6000 CL30 gives exactly the same 10.0 ns — with 88% more bandwidth. DDR5-6000 CL28 gives 9.3 ns, which is genuinely faster. The CL number in isolation tells you nothing across generations, and comparing them directly is the single commonest mistake in RAM buying.
If you are building on AM5 or on Intel's current desktop socket, the question does not arise — those platforms take DDR5 and nothing else.
The interesting case is an existing DDR4 machine. Adding capacity to a DDR4 system is often still the best value upgrade available, because the alternative is a new board, a new CPU and new memory all at once. But DDR4 pricing has inverted as production winds down, and a 32 GB DDR4 kit can now cost more than its DDR5 equivalent. Check the current spread before assuming the older generation is the cheaper one.
For a platform-by-platform breakdown of what capacity to target, see how much RAM you actually need.
CAS latency is a count of clock cycles, not a unit of time, so a lower CL number does not mean a faster module once two kits run at different clock speeds -- converting to true latency is the only way to compare them honestly.
A RAM kit ships running at a conservative JEDEC default speed no matter what the box says, and it stays there until an XMP or EXPO profile is enabled in BIOS -- the two standards do the same job for Intel and AMD platforms respectively.