CAS Latency Chart: DDR3, DDR4 and DDR5 Kits Compared

According to Wikipedia's CAS latency article, CL is the delay in clock cycles between the READ command and the moment data is available. To get a time, multiply CL by the length of one clock cycle.

DDR memory moves data twice per clock cycle, so the clock runs at half the data rate. A DDR5-6000 kit has a 3000 MHz clock, and each cycle takes 0.333 ns. At CL30 that's 30 cycles, or 10.00 ns.

As a formula, first word latency in nanoseconds is CL × 2000 ÷ the data rate in MT/s. The 2000 is there because a cycle lasts 1000 divided by the clock in MHz, and the clock is half the data rate.

A bigger CL number doesn't mean a slower kit when the clock runs faster. DDR4-3200 CL16 has a cycle time of 0.625 ns, so its CL takes 10.00 ns. DDR5-6000 CL30 has a CL of 30 against 16, but each of its cycles lasts only 0.333 ns. It lands on 10.00 ns too.

DDR4-3600 CL18 and DDR5-6400 CL32 reach the same 10.00 ns, so these four kits sit in a row in the table. Kits that tie go by data rate, with the highest first.

The reverse holds as well. Wikipedia's Memory timings article says that without the clock frequency, it's impossible to say whether one set of timings is faster than another. Don't compare CL numbers across kits that run at different speeds.

CL is the wait for the first word of data once a row is already open. The CAS latency article says a completely random access also has to close any open row and open the one it wants before the CAS latency starts. Two timings, tRP and tRCD, set those steps, and the chart doesn't rank them.

They can differ between kits with the same CL time. DDR4-3200 CL16 has a tRCD of 18 cycles, and DDR5-6000 CL30 has a tRCD of 38 cycles. Opening a row takes 12.67 ns on the DDR5 kit and 11.25 ns on the DDR4 kit, even though both reach their first word in 10.00 ns. Open a kit in the calculator to see all of its timings.

Bandwidth is a separate number. A 64-bit DDR5-6000 channel peaks at 48.0 GB/s, which is 88% above the 25.6 GB/s of DDR4-3200. The Memory timings article says higher bandwidth can improve systems with several processors or threads, and integrated graphics that use system memory, even when latency goes up.

The same article says different workloads have different memory access patterns, so it's hard to say how much a latency gap changes real performance. It also says the architecture of the CPU can change the latency.

Start with the generation your board takes. According to Wikipedia's DDR4 article, DDR4 isn't compatible with any earlier type of RAM because of its signaling voltage and physical interface, so your CPU and motherboard usually decide this for you. Within a generation, look for a low first word latency, then compare prices.

Each row shows the speed and timings a kit is sold at. Faster kits reach them through an XMP or EXPO profile. The Memory timings article says some modules store these faster, tested profiles next to the standard JEDEC ones, and the BIOS lets you choose between them. If your BIOS runs a kit on a JEDEC profile, you get that profile's timings, and the row here won't match what the kit runs at.

For a closer look at one kit, open it in the RAM Latency Calculator. The DDR4 vs DDR5 latency page compares two popular kits, and the RAM latency guide for PC builders covers how to choose.