RAM Speed and Latency Explained: MHz and CL

What RAM speed really means: MHz or MT/s

RAM speed is the number of data transfers a module makes per second, written in megatransfers per second (MT/s). Shops and box labels usually write MHz instead, so a kit sold as "DDR5 6000MHz" really runs at 6000 MT/s. The difference matters once you start comparing kits on latency.

DDR memory moves data twice per clock cycle, once on the rising edge and once on the falling edge. That is where the name double data rate comes from. A DDR5 6000 kit therefore has an actual memory clock of 3000 MHz, and a DDR4 3200 kit runs at 1600 MHz. Some system tools report that real clock, which is why they can seem to show half the speed on the box.

Each generation has its own range of standard speeds. Kingston lists 3200 MT/s as the top JEDEC speed for DDR4 and 4800 MT/s as the starting point for DDR5. Engadget reports that JEDEC raised the official DDR5 ceiling to 8,800 MT/s in 2024. Anything sold above the standard speeds is a factory overclock that relies on a memory profile, more on that below. If you are new to the basics, start with our guide to what RAM is and how much you need.

What RAM latency (CL) means

CAS latency, shortened to CL, is the number of clock cycles the memory waits between receiving a request and sending the first piece of data back. A kit labeled CL30 waits 30 cycles, a CL16 kit waits 16. Lower is better at the same speed.

The catch is that CL counts cycles, not time. A cycle on a fast kit is shorter than a cycle on a slow kit, so a DDR5 kit with CL30 is not twice as slow as a DDR4 kit with CL16. To compare them you need to turn cycles into nanoseconds.

The full timing string on a spec sheet, such as 30-36-36-76, lists CL first and then a few secondary timings. For buying decisions the first number and the speed are enough. The secondary timings matter to people who tune memory by hand.

True latency: the number that lets you compare kits

Crucial gives the conversion: true latency is the length of one clock cycle times the CAS latency. Because DDR transfers twice per cycle, that works out to a simple rule you can do on a phone calculator: CL x 2000 divided by the speed in MT/s gives the latency in nanoseconds.

Two quick examples. DDR4 3200 CL16 is 16 x 2000 / 3200 = 10 ns. DDR5 6000 CL30 is 30 x 2000 / 6000, also 10 ns. Both kits answer a request in the same time, but the DDR5 kit moves almost twice as much data per second once it starts.

Crucial also points out that CL numbers went up as speeds went up, so true latency in nanoseconds has stayed roughly the same across generations. When two kits land on the same latency in nanoseconds, the faster one wins. Engadget adds that standard JEDEC timings keep DDR5 near 14 ns from DDR5 3200 all the way to DDR5 8800. Lower latency than that comes from kits sold with tighter timings and a memory profile.

The table below shows common speed and CL pairs we see in US shops, their true latency and the cheapest desktop kit of that pair this week, in any capacity.

Latency is only half the picture. A DDR4 3200 CL16 kit and a DDR5 6000 CL30 kit tie on response time, yet the DDR5 kit has far more bandwidth. That is one reason a DDR4 board is a poor place to spend big on fast memory. Our DDR5 vs DDR4 comparison covers the rest of the differences, including why the two never fit the same slot.

Why a new kit runs slower than the box says

Most fast kits do not run at their rated speed out of the box. The speed and timings on the label are stored in a profile on the module: XMP for Intel systems and EXPO for AMD AM5 systems. Intel describes XMP as pre-tested overclocking profiles that you switch on in the BIOS, and AMD describes EXPO as its DDR5 equivalent with profiles tuned for Ryzen.

Until you enable that profile, the board falls back to a standard JEDEC speed. Newegg notes that a DDR5 6000 kit typically starts at 4800 or 5600 MT/s. So if you paid for DDR5 6000 CL30 and never opened the BIOS, you are running a slower kit than the one you bought.

Both Intel and AMD mark these profiles as overclocking, which falls outside the official processor spec. If a system becomes unstable after enabling the profile, check the motherboard's memory support list for your kit before you blame the memory.

Which RAM speed fits your processor

The right speed depends on the platform more than on the kit. Here is what the makers list and what the press recommends.

AMD Ryzen 7000 and 9000 (AM5). AMD lists DDR5 5600 as the supported speed for Ryzen 9000 and DDR5 5200 for Ryzen 7000. Even so, DDR5 6000 CL30 is the common choice for AM5 builds. Engadget explains why: the memory controller can run in step with the memory clock up to roughly 3,000 MHz, which is DDR5 6000. Faster kits push the controller to half the memory clock, which adds delay. Engadget cites Tom's Hardware testing where DDR5 6000 CL30 finished ahead of a DDR5 6400 CL32 kit. Keep in mind that DDR5 6000 is a community sweet spot, not an AMD guarantee, and that it relies on EXPO.

Intel Core Ultra 200S (LGA 1851). Intel lists DDR5 up to 6400 MT/s for the Core Ultra 9 285K. Engadget adds that 6400 applies to CUDIMMs, modules with their own clock driver, and that regular sticks are rated at DDR5 5600. Newegg's 2026 buying guide recommends DDR5 6000 to 6400 for Core Ultra builds.

Intel 12th to 14th gen and AMD AM4. Intel's 12th to 14th gen Core chips support both DDR4 and DDR5, and the motherboard decides which one you need. AM4 boards with Ryzen 5000 are DDR4 only. On DDR4, 3200 MT/s is the top standard speed, and CL16 at that speed matches the 10 ns latency of a good DDR5 kit. Our DDR4 price page lists those kits.

The next table shows the cheapest 32GB desktop kit at each common speed this week, so you can see what a step up in speed costs right now.

For every DDR5 6000 kit in one list, sorted by price, see our DDR5 6000 CL30 price page. If you are still deciding on size rather than speed, how much RAM you need is the better place to start.

Does RAM speed matter for gaming?

Less than the price gap between kits suggests. Engadget's October 2026 look at DDR5 speed found that no game requires a particular DDR5 speed, and that any game that runs on DDR5 runs fine on the slowest DDR5 4800 kits. Moving from the slowest kits to a mid range kit can add a little performance. Kits at DDR5 7200 and 8400 only marginally improve frame rates in processor bound games, at a much higher price.

The same article cites Tom's Hardware testing on Intel, where the lowest scoring 32GB kit finished about 4 percent behind the fastest in heavy productivity work. That is a real but small gap.

For most gaming builds the order of priorities is simple: enough capacity first, then a two stick kit, then a sensible speed for your platform such as DDR5 6000 CL30 on AM5. Paying extra beyond that buys very little. Our best RAM for gaming comparison puts the common picks side by side with this week's prices.

Common mistakes

• Comparing CL across generations. CL40 on DDR5 and CL16 on DDR4 are not comparable as raw numbers. Convert both to nanoseconds first.
• Reading MHz as the clock speed. The MHz on the box is the data rate. The real clock is half of it.
• Forgetting XMP or EXPO. Without the profile, a DDR5 6000 kit runs at a slower default speed and you lose what you paid for.
• Buying the fastest kit for an AM5 build. Above DDR5 6000 the AMD memory controller drops to half the memory clock, so a pricier kit can end up slower.
• Filling four slots. Four modules can force the board to a much lower speed than two. Buy the capacity you need as one two stick kit.
• Paying for speed instead of capacity. If your system runs out of memory, more gigabytes help far more than a faster kit of the same size.