RAM specifications can look like a collection of unrelated numbers: DDR4 or DDR5, a transfer rate, capacity and a string of timings such as 16-18-18-38. Understanding what each describes helps you compare upgrades without assuming that the largest speed number will fix every slow computer.
RAM is working memory. It holds the data that running applications need, while an SSD or hard drive stores files over the longer term. Capacity determines how much work can fit in memory; transfer rate and timings describe aspects of accessing that work. Enough capacity, supported hardware and stable settings are the starting points for a useful upgrade.
MHz and MT/s measure different things
Megahertz, or MHz, means millions of clock cycles per second. Megatransfers per second, or MT/s, measures millions of transfers per second. DDR stands for Double Data Rate: data transfers occur on both edges of the memory bus clock, so the transfer rate is twice that clock frequency.
For example, DDR4-3200 means 3200 MT/s at a 1600 MHz memory bus clock. It does not mean a 3200 MHz physical clock that you should double again to get 6400 MT/s. Likewise, DDR5-6000 corresponds to a 3000 MHz bus clock. Retailers sometimes use MHz loosely for the advertised data rate; check the DDR model and specification rather than multiplying the listing blindly.
Kingston's explanation of MHz and MT/s covers this distinction. Its memory support FAQ also explains why diagnostic software can show half the marketed DDR transfer-rate number as the DRAM frequency. A 2400 MHz reading for memory running at 4800 MT/s is consistent, rather than proof of an underperforming kit.
Reading CL and the primary timings
In a timing sequence such as 16-18-18-38, the numbers commonly describe CL, tRCD, tRP and tRAS. Consult the module's specification because the full timing configuration includes more than these four values.
- CAS latency, CL: the delay in memory clock cycles between a read command and the start of returned data, for the relevant open row.
- tRCD: the delay between activating a row and a column access within it.
- tRP: the precharge interval used when closing a row before another activation.
- tRAS: the minimum interval a row must remain active before it can be precharged.
A smaller CL value is not automatically faster across different data rates: clock cycles have different durations. At the same rate, lower supported timings can reduce those delays. Across rates, convert the values to time before comparing, while remembering that the controller and other timings also affect actual access.
Calculate CAS delay in nanoseconds
For DDR memory, the arithmetic is CAS delay in nanoseconds = CL × 2000 ÷ transfer rate in MT/s. The factor of 2000 accounts for nanosecond units and the two transfers per clock cycle. A 3200 MT/s CL16 configuration gives 16 × 2000 ÷ 3200 = 10 ns.
The following figures are calculated examples, not measured application results. Bandwidth assumes a 64-bit data width and decimal GB/s.
| Example configuration | Transfer rate | CAS cycles | Calculated CAS delay | Theoretical 64-bit bandwidth |
|---|---|---|---|---|
| DDR4-3200 CL16 | 3200 MT/s | 16 | 10.00 ns | 25.6 GB/s |
| DDR4-3600 CL16 | 3600 MT/s | 16 | 8.89 ns | 28.8 GB/s |
| DDR4-3600 CL18 | 3600 MT/s | 18 | 10.00 ns | 28.8 GB/s |
| DDR5-5200 CL40 | 5200 MT/s | 40 | 15.38 ns | 41.6 GB/s |
| DDR5-6000 CL30 | 6000 MT/s | 30 | 10.00 ns | 48.0 GB/s |
| DDR5-6400 CL32 | 6400 MT/s | 32 | 10.00 ns | 51.2 GB/s |
DDR4-3200 CL16 and DDR5-6000 CL30 both produce a 10 ns CAS component, while the latter has more theoretical bandwidth. That does not make their complete memory latency identical. Row state, command scheduling, controller ratios, refresh and other system behaviour contribute to the time a processor actually waits. A benchmark can therefore report a much longer latency than this simple CAS calculation.
Kingston's timing guide provides the formula and timing definitions. Use the calculation to understand a specification, rather than as a substitute for testing your actual workload.
Bandwidth, channels and DDR5 architecture
A 64-bit data path transfers eight bytes at a time, giving the example bandwidth calculation MT/s × 8 ÷ 1000 = GB/s. This is an ideal limit; application throughput includes overhead and depends on the memory arrangement. Installing modules in the recommended slots can enable the processor's intended channel configuration. Follow the motherboard manual instead of assuming adjacent slots are the correct pair.
Kingston's DDR5 architecture document explains that a standard DDR5 module splits its 64-bit data width into two independent 32-bit subchannels. Those subchannels do not double that module's total data width. They should also not be confused with populating two motherboard memory channels.
DDR5 includes on-die ECC, which corrects certain errors inside a DRAM chip. It does not provide the same protection as a complete ECC memory path between the module and CPU. The extra check bits on a server ECC module are a separate feature: the 40-bit subchannel width described for such modules is not caused simply by ordinary DDR5 on-die ECC. For a server or NAS, verify CPU, motherboard and module ECC support together.
DDR5 also moves power management onto the module and increases burst length. These architectural changes improve how memory is organised, but they do not make DDR5 mechanically compatible with a DDR4 socket. The correct generation and module type remain essential.
Controller clocks: avoid a universal Gear recommendation
A memory controller can run at a different clock ratio from the DRAM. The available ratios and their behaviour depend on the processor and firmware. Raising the transfer rate can involve a different controller setting, so a larger MT/s number may not produce the lowest complete latency.
For example, Intel's 13th/14th-generation processor documentation lists DDR5 operating points in Gear 2. On AMD systems, MCLK, UCLK and FCLK describe separate memory, controller and fabric clocks. Do not apply an AMD 1:1 recommendation to every Intel DDR5 system, or assume one speed is a guaranteed limit for every Ryzen processor.
Rated profiles and reliable settings
Intel XMP and AMD EXPO provide memory profiles containing settings such as rate, timings and voltage. They are overclocking technologies, not universal guarantees that a kit's fastest profile will work in every computer. Check the system's support and warranty conditions before enabling settings outside the published limits.
Look up the exact kit part number in the motherboard's compatibility information, alongside the CPU's supported configuration. Module count and capacity matter too. Two separately purchased kits with matching labels are not necessarily validated together as one kit. If a profile causes boot failures or instability, return to default settings and investigate compatibility rather than copying a stranger's voltage settings.
Choosing an upgrade that helps your work
First check whether you run out of capacity during the tasks that feel slow. If a workload cannot fit comfortably in RAM, buying a smaller kit with tighter timings may be the wrong trade-off. Then consider transfer rate, latency and price within the configurations your system supports.
Gaming, compilation, editing and rendering respond differently to memory changes, and even similar applications can be limited by different components. Compare the same project or game settings before and after an upgrade; do not expect a fixed percentage gain or guaranteed elimination of stutters. Keep the original settings as a baseline and include stability testing before trusting important work to a new profile.
A balanced upgrade provides enough capacity, the correct memory type and reliable operation at a sensible price. MHz, MT/s and CL help explain the choice, but none of them alone decides the best RAM for every computer.
