For most gaming PCs, the answer is to look at both numbers together and compare true latency in nanoseconds, not either figure alone. Higher speed lifts bandwidth, tighter timings lower the wait for each access, and the best kit for your system balances them within the limits of your CPU and motherboard. If you spend a fixed amount, getting enough capacity and running a matched pair in dual channel should come first; speed and latency tuning come after.
The explanation below comes from the memory standard’s documented behavior, manufacturer specification sheets and buyer discussions, assembled by our editorial team. The numbers are calculations, not measurements.
The core rule: speed is bandwidth, latency is the wait
Think of memory as a delivery route. Speed, expressed in MT/s (often labeled MHz on packaging), is how many deliveries can move per second. Latency is how long a single request waits before the first byte arrives. Games issue huge numbers of small requests, so both matter, but not equally for every game or processor.
The trap is that the two numbers rise together. Each new memory generation raises speed substantially while the CAS latency number, counted in clock cycles, also climbs. A bigger CL does not mean a slower kit if the clock is faster. That is why you must convert.
The true latency formula
True latency in nanoseconds is the CAS latency multiplied by 2000, divided by the transfer rate in MT/s. The factor of 2000 appears because DDR memory transfers twice per clock cycle: the clock in MHz is half the MT/s figure, and one cycle takes 1000 divided by that clock in nanoseconds.
Formula: true latency (ns) = CL x 2000 / MT/s.
For example, DDR5-6000 CL30 gives 30 x 2000 / 6000 = 10.0 ns. DDR5-6000 CL36 gives 36 x 2000 / 6000 = 12.0 ns. The first kit responds sooner on the primary timing, even though both have the same bandwidth.
This formula covers only the first timing, CAS. Real behavior also depends on other timings such as tRCD, tRP and tRAS, on the memory controller and on the CPU’s internal clocks. Use the formula as a quick filter, not a full model.
True latency across generations
| Kit (as sold) | CL | True latency (calculated) | Notes |
|---|---|---|---|
| DDR4-2666 | 19 | 14.3 ns | Common older default, low bandwidth |
| DDR4-3200 | 16 | 10.0 ns | Widely used value kit |
| DDR4-3600 | 16 | 8.9 ns | Often the DDR4 sweet spot |
| DDR5-4800 | 40 | 16.7 ns | Baseline DDR5, high latency |
| DDR5-5600 | 40 | 14.3 ns | More bandwidth, similar wait to old DDR4-2666 |
| DDR5-6000 | 30 | 10.0 ns | Popular target on AMD AM5 boards |
| DDR5-6000 | 36 | 12.0 ns | Same speed, looser timing |
| DDR5-7200 | 34 | 9.4 ns | Needs a board and CPU that handle it |
Look at what this shows: a DDR4-3600 CL16 kit has a lower first-word wait than most DDR5 kits, while the DDR5 kit gives more bandwidth. DDR5 wins on throughput, especially with fast kits, but DDR4 can match it on latency in simple comparisons. That is part of why people who already own a good DDR4 platform often postpone upgrading just for memory. Our piece on how much RAM you need for gaming covers capacity, the other half of the buying question.
Worked example: choosing between two kits
Assume you are building on a DDR5 platform and have two 32 GB kits in front of you, at similar prices. Kit A is DDR5-6000 CL30. Kit B is DDR5-7200 CL38. Assume your board and processor list both speeds as supported.
| Kit | Speed | CL | True latency | Relative bandwidth |
|---|---|---|---|---|
| A | 6000 | 30 | 10.0 ns | Baseline (1.00x) |
| B | 7200 | 38 | 10.6 ns | About 1.20x |
Kit B has about 20 percent more raw bandwidth but slightly higher first-word latency. Which wins depends on the platform. On a board where 7200 forces the memory controller into a less efficient ratio, kit A can match or beat it in games, and it will probably be easier to stabilize. On a platform that handles high speeds well, kit B may pull ahead in bandwidth-hungry tasks. If you do not know which describes your build, take the simpler kit A and check the board’s memory support list before buying anything faster.
Which to prioritize, by situation
| Your situation | Prioritize | Reason |
|---|---|---|
| AMD AM5 gaming build | The documented sweet-spot speed with the lowest CL you can afford | Many boards and CPUs are reported to run best at a particular speed in a 1:1 memory-controller ratio |
| Intel build with a board that supports high speeds | Higher speed, provided the board runs it stably | Some Intel platforms scale well with bandwidth |
| Integrated graphics (APU) | Speed and bandwidth | The GPU shares system memory |
| Older DDR4 platform | Tighter timings at 3200 to 3600 | Cheap, effective tuning |
| Content creation, large files | Capacity first, then speed | Running out of RAM costs more than a few nanoseconds |
| Tight budget | A matched dual-channel kit at the platform’s default sweet spot | Avoid paying a premium for tiny gains |
If you are weighing whole platforms, our comparison Ryzen 7 9800X3D vs Core Ultra 9 285K touches on how each handles memory, and the list of DDR5 6000 CL30 kits shows what the AM5 sweet spot looks like in kit form. On a DDR4 board, stay with DDR4 kits; the generations do not mix.
Primary timings are only the start
A kit label lists four numbers, such as 30-38-38-96. The first is CAS latency, the one our formula uses. The second and third, tRCD and tRP, set how long the memory takes to open a row and close one, and the fourth, tRAS, is the minimum time a row stays active. Two kits with the same CL can behave differently if the other numbers differ, which is why a CL30 kit with looser follow-on timings is not always identical to another CL30 kit. Behind those sit dozens of secondary and tertiary timings that the BIOS sets automatically from the profile. Manual tuning can squeeze out more, but it takes time, patience and stability checks, and the gains in games are typically modest. For most buyers, picking a well-regarded kit with a good primary-timing ratio and enabling its profile gets you the bulk of the benefit.
Capacity, channels and ranks still count
Speed and latency tuning sits on top of a basic setup. Two matched modules in dual channel give far more bandwidth than a single stick at any speed. Capacity matters because a game that needs more memory than you have will start swapping to the drive, and no timing trick makes up for that. Rank count also plays in: modules with two ranks per stick can raise efficiency at a given speed but may limit the top stable frequency, especially with four sticks installed. A 2 x 16 GB pair is a common starting point; how much capacity to buy is a separate question that depends on the games and apps you run.
Memory controller ratios in plain terms
On some platforms the processor’s memory controller and the memory itself have to run at a fixed ratio, such as 1:1, to give the best latency. Past a certain speed, the controller may switch to a half-speed ratio, and latency can rise even though bandwidth keeps climbing. This is why a faster kit does not always improve games on those chips, and why many guides name a single speed as the sweet spot for a given platform. The exact limit varies by processor and BIOS version, so look up your CPU’s documentation or reputable reviews instead of assuming a number.
Platform and hardware limits
The kit’s label is a promise from the memory maker, not from your motherboard. The processor’s memory controller, the board’s trace layout and the BIOS all set a ceiling. Populating four slots instead of two often lowers the stable speed. Dual-rank modules can also limit speed compared with single-rank. The memory support list on the board maker’s site is the safest reference.
You also need to turn the speed on. Without XMP or EXPO enabled in BIOS, the kit usually runs at its standard baseline speed. Every figure above assumes the profile is on and stable. A fast profile that crashes during gaming is worse than a slightly slower setting that does not.
Verification checklist
- Compute true latency for each candidate: CL x 2000 / MT/s.
- Confirm the speed is on the CPU and motherboard support lists, not only on the kit label.
- Choose a kit of two matched modules before a four-module setup.
- After installing, enable XMP or EXPO, then check the speed and timings in Windows or CPU-Z.
- Run a memory diagnostic or stress workload to catch instability before you rely on the machine.
- Keep capacity as the first priority; a kit that is too small hurts more than tuning can fix.
When to stop tuning
If the PC crashes, freezes or reboots after you raise speed or loosen limits, step back to the last stable profile. The potential gain is small, so a stable system matters more. If you are CPU-limited, the improvement from faster memory often sits in the single-digit percentage range, and if you are GPU-limited, it may be invisible. Spend leftover budget on a better graphics card or more capacity when memory is already at a sensible sweet spot.
Frequently asked questions
Is a higher MHz number always better?
No. The MHz or MT/s figure only tells you how many transfers happen per second, and a higher number usually comes with a looser CAS latency. Divide the two to get true latency in nanoseconds. A kit that is faster on paper but has much looser timings can end up with the same or worse real latency, so compare both numbers together.
What is a good CAS latency for DDR5?
There is no single good number, because the right CL depends on the speed. At DDR5-6000 many enthusiasts treat CL30 as the sweet spot and CL36 as ordinary, while at DDR5-4800 a CL of 40 is common. Convert each to nanoseconds with the formula to compare across speeds, and remember bandwidth also matters.
Does RAM speed matter more than capacity?
Usually not. Running out of memory causes much larger slowdowns than a modest speed difference, so make sure you have enough capacity first. Once capacity is covered, speed and latency tuning are refinements, often with single-digit percentage gains in CPU-limited games and little change when the graphics card is the bottleneck.
Will faster RAM improve my frame rates?
Sometimes. Games that are limited by the processor, often at lower resolution and high refresh rates, tend to respond most. Games limited by the graphics card barely change. Gains also depend on the platform, since some CPUs need the memory controller to stay in a certain ratio to benefit. Check reviews for your CPU and game types.
Do I have to enable XMP or EXPO to get the advertised speed?
Yes in nearly all cases. Memory usually starts at a standard baseline speed, not the speed on the label, until you turn on the profile in BIOS. Intel boards call it XMP and AMD boards call it EXPO or DOCP. After enabling it, confirm the speed in Windows or a utility like CPU-Z and watch for stability problems.
