DDR5 for gaming: speed, capacity & what CL30 means

CL30 means the memory waits thirty clock cycles between being asked for data and handing it over. Lower is better. On its own, though, the number tells you nothing, because a cycle on fast memory is shorter than a cycle on slow memory. CL30 at 6000MHz and CL36 at 6000MHz are meaningfully different. CL30 at 6000 and CL16 at 3200 are, in the way that matters, identical.

DDR5-6000 CL30 is the answer for most gaming builds. It's the sweet spot on current AMD systems and a sensible choice on Intel ones. The rest of this is why, and when to ignore it.

The number that actually matters

CAS latency counts clock cycles. It isn't a measure of time, so to get time out of it you divide. Multiply the CL number by 2000, then divide by the memory speed, and the result is true latency in nanoseconds.

  • DDR5-6000 CL30 gives 10ns
  • DDR5-6000 CL36 gives 12ns
  • DDR5-5600 CL36 gives about 12.9ns
  • DDR4-3200 CL16 gives 10ns

Compare the first entry with the last. A good DDR5-6000 kit and a good DDR4-3200 kit respond in the same time, despite the DDR5 kit's CL number being nearly double, which is why those higher CL figures caused so much confusion when DDR5 launched and why comparing CL numbers across different speeds is meaningless.

What DDR5 adds is bandwidth. Same response time, roughly twice as much data moving per second. That's the generational improvement, and it also explains why early DDR5 felt underwhelming. The bandwidth was there from day one. The latency took a couple of years of maturing kits to catch up with good DDR4.

Where the sum earns its keep is comparing two kits at the same price. One is 6000 CL30, the other 6400 CL36. Run the numbers rather than assuming the bigger speed wins: the 6000 CL30 kit responds quicker, 10ns against 11.25ns, while the 6400 kit moves slightly more data. For gaming, the first usually edges it.

DDR4 or DDR5 is rarely your decision

The two are physically different and not interchangeable. A board takes one or the other, and the slots are keyed so the wrong one won't fit. The question isn't really which memory to buy, then. It's which platform you're on.

DDR5 only: AMD's AM5 socket, used by Ryzen 7000 and 9000 chips, and Intel's LGA1851 socket used by Core Ultra.

DDR4 only: AMD's older AM4 socket, still very much alive at the budget end with Ryzen 5000 chips.

If you're buying a complete PC, the memory type follows from the processor you pick. If you're upgrading, check what your board takes before buying anything, because a DDR5 kit will not go into a DDR4 board however firmly you press.

What speed to buy

On AMD AM5, DDR5-6000 is the target, and that isn't just a rule of thumb. The memory controller runs in step with the memory up to around that point. Push higher and it drops into a less efficient mode where it runs at half rate, and you lose more in latency than you gain in bandwidth. A 6000 CL30 kit will usually beat a 7200 kit on the same system, which is not what the numbers on the box suggest.

On Intel Core Ultra, the controller behaves differently and higher speeds are more useful. The chips officially support 6400, with the 2026 refresh parts rated for 7200. There's also a newer type of module with a small clock driver built onto the stick, designed to keep signals clean at high speeds, which mainly benefits Intel systems. Worth knowing if you're chasing the top end. Not necessary otherwise.

On DDR4 systems, 3200 or 3600 with CL16 timings is the sensible target. Don't pay much for anything beyond that.

One thing applies to all of it. Memory runs at a slow default speed until you enable its profile in the BIOS, called EXPO on AMD and XMP on Intel. Buying a fast kit and never turning it on is the most common way people waste this money entirely.

What the difference is worth

Less than the marketing implies, and it depends what processor you have. For most gaming, the gap between sensible memory and expensive memory is a few percent. Real, measurable, and not worth reordering your budget over. If the choice is a better memory kit or a better graphics card, take the graphics card every time.

The interesting exception runs the other way. An AMD X3D processor makes memory speed matter to you less, not more, because the large cache on those chips absorbs a lot of the traffic that would otherwise leave the processor. Avoiding trips out to memory is the entire point of that cache. So the buyers most likely to reach for an expensive kit are often the ones with least to gain from it.

Where memory speed does show up more clearly is in the same places cache does. Simulation games, strategy titles with large maps, anything with a big working set. Same underlying reason.

One stick or two

A DDR4 stick presents a single 64-bit channel to the processor. A DDR5 stick presents two independent 32-bit sub-channels, which is a real architectural change and is why you'll see single DDR5 sticks described as already being dual channel. They aren't, in the sense that matters to you. Two sub-channels of 32 bits is still 64 bits in total, the same width as one DDR4 stick. Fit a second stick and you get 128 bits. The sub-channel split makes that width easier to use efficiently. It doesn't double it.

So the old rule is unchanged: buy two sticks, not one.

What it's worth depends entirely on what's limiting you, and this is where most of the confusion online comes from, because people quote figures from very different setups. If your graphics card is the bottleneck, which it usually is at 1440p and above, moving from one stick to two may change surprisingly little, because the frames aren't waiting on memory. Where it shows up is anywhere the processor is doing the work: 1080p at high frame rates, simulation and strategy titles, big open worlds, and in the 1% lows rather than the average. Those are the same conditions that make memory speed matter at all, and for the same underlying reason.

The exception is integrated graphics, and it's not a small one. A machine with no separate graphics card borrows system memory to act as video memory, so the memory bus is doing two jobs at once. Halve its width and you halve what both jobs get. On an integrated build, running single channel is one of the largest self-inflicted performance losses available, and it's common, because a single stick is an easy place to save a few pounds when a machine is being built down to a price.

Two practical points. Sticks bought separately often won't run at their rated speed together, so buy a matched kit rather than adding to one later. And if you're upgrading an existing machine, open it and count the populated slots before you buy anything, because one large stick and two small ones are different problems with different fixes.

There's a fuller explanation of how channels work in our guide to single, dual and quad channel memory.

Capacity, briefly

This piece is about speed and timings, so the short version on capacity: 32GB is the sensible target for a new gaming build, 16GB still works, and 64GB is for work rather than games. What matters more than either figure is buying it as two matching sticks rather than one large one, for the reasons above.

What to look for

For a new AM5 build, a 32GB kit at DDR5-6000 CL30, bought as a matched pair.

For Intel Core Ultra, that same kit is a perfectly good choice, and you can go faster if you want to without wasting money the way you would on AMD.

For a DDR4 budget build, 16GB or 32GB at 3200 CL16.

Whatever you buy, check the profile is enabled once it's in the machine.

You can see the kits we stock in RAM and memory, and specify memory alongside the rest of the build in the PC configurator. If you'd rather not pick components individually, the full gaming PC range ships with memory already matched to the platform.