"Best DDR5 RAM" lists usually name a handful of kits. We rank every DDR5 kit we track by price per gigabyte instead, because most DDR5 kits at the same speed and latency perform the same, and the price is what differs. The median DDR5 kit costs £17.37 per gigabyte today, against £14.97/GB 90 days ago (+16%) and £4.61/GB a year ago (+277%). This guide explains how to tell a good-value DDR5 kit from an expensive one: which speeds and latencies are worth paying for, which capacities are cheapest per gigabyte, and which features on the box mean less than they suggest.
| Measure | DDR5 median £/GB |
|---|---|
| Today | £17.37 |
| 90 days ago | £14.97 |
| Change over 90 days | +16% |
| A year ago | £4.61 |
| Change over 12 months | +277% |
These are medians across the DDR5 kits in the ramprices.uk price index (4,248 kits across all generations, updated 22 October 2026), calculated from Amazon.co.uk prices divided by each kit's total capacity. For comparison, the DDR4 median is £8.22/GB. The DDR5 page ranks every kit live.
A DDR5 kit is described by its data rate in MT/s (often written as MHz on the box) and its CAS latency (CL), the number of clock cycles before data starts to arrive. Higher MT/s is better; lower CL is better. Because CL is counted in cycles and faster memory has shorter cycles, the two must be combined to compare kits fairly. The formula for first-word latency in nanoseconds is:
true latency (ns) = CL × 2000 ÷ MT/s
| Kit | Working | True latency |
|---|---|---|
| DDR5-4800 CL40 | 40 × 2000 ÷ 4800 | 16.7ns |
| DDR5-5600 CL40 | 40 × 2000 ÷ 5600 | 14.3ns |
| DDR5-5600 CL36 | 36 × 2000 ÷ 5600 | 12.9ns |
| DDR5-6000 CL36 | 36 × 2000 ÷ 6000 | 12.0ns |
| DDR5-6000 CL30 | 30 × 2000 ÷ 6000 | 10.0ns |
| DDR5-6400 CL32 | 32 × 2000 ÷ 6400 | 10.0ns |
| DDR4-3200 CL16 (for reference) | 16 × 2000 ÷ 3200 | 10.0ns |
Two things follow. DDR5-6000 CL30 and 6400 CL32 have the same latency, and the 6400 kit has a little more bandwidth. And a cheap DDR5-4800 CL40 kit, the JEDEC baseline, has considerably higher latency than a good DDR4 kit; DDR5's advantage comes from bandwidth, not latency. True latency is only one measure, and secondary timings also matter, but it is the quickest way to compare two listings.
DDR5 added non-binary module sizes, 24GB and 48GB, built from 24Gb chips. That gives four common two-module kits:
| Kit | Total | Notes |
|---|---|---|
| 2x16GB | 32GB | The default for gaming. Single-rank modules, which run high speeds easily. |
| 2x24GB | 48GB | Single-rank 24Gb modules. A useful step above 32GB without the dual-rank load. |
| 2x32GB | 64GB | Usually dual-rank. For editing, development and VMs alongside gaming. |
| 2x48GB | 96GB | Dual-rank 24Gb modules. Workstation capacity in two slots. |
Which is cheapest per gigabyte changes with the market, so check the ranking rather than assuming. As a rule the larger kits are not penalised per gigabyte the way they once were, and through 2026 the densest modules moved most in price. The 16GB DDR5 page covers single modules and 2x8GB kits, which are usually the dearest per gigabyte and, in the 2x8GB case, built from narrower modules with less bandwidth.
DDR5 is much harder on the memory controller with four modules than with two. With four sticks fitted, many systems will not hold the kit's rated speed and need to drop considerably below 6000, and AMD's and Intel's official supported speeds for two modules per channel are lower than for one. If you need 64GB, buy 2x32GB; if you need 96GB, buy 2x48GB; and leave the other two slots empty. For 128GB in two slots, 64GB modules exist but need a board and BIOS that support them, which the manual will state.
Enthusiasts discuss which DRAM chips a kit uses, and kits built on SK hynix chips (known as A-die and M-die) have a reputation for overclocking well. That is useful background, but it is a poor buying criterion for most people. Module makers can change the chips inside a kit without changing the part number, listings rarely state the chips, and at the 6000 CL30 sweet spot most kits will run their rated profile regardless. Buy the specification, not the rumour.
EXPO is AMD's memory profile format and XMP is Intel's. Both store a tested speed, timings and voltage on the module so the board can apply them in one setting. Many DDR5 kits carry both. On AM5 an EXPO kit is the straightforward choice, but most AM5 boards can also read XMP profiles. On Intel boards, use XMP. A kit with neither runs at JEDEC speed only, which is fine for office machines but not what you want for gaming.
Every DDR5 module has on-die ECC, and some listings mention it as a feature. It corrects bit errors inside each DRAM chip, which the denser chips need in order to be manufactured economically. It does not check data as it travels between the module and the CPU, and it does not report errors to the operating system. Real ECC memory has extra chips on the module and needs a supporting CPU and board; consumer DDR5 is not ECC memory. If you need ECC, see the ECC RAM page and the server page.
DDR5 moved voltage regulation from the motherboard onto each module, in a power management chip (PMIC). This gives cleaner power at high speeds and is one reason DDR5 modules cost more to make than DDR4 ones. For a buyer it mostly explains why DDR5 sticks run warmer around one corner and why every kit, including the plainest, has some form of heatspreader or label covering the chips. It is not a feature you need to shop for.
That kit is the best DDR5 RAM for your money today. If the platform question is still open, the DDR4 vs DDR5 guide covers when the older generation is the better buy, and the gaming RAM guide covers how much you need.