Best value RAM in 2026: eight DDR4 and DDR5 kits compared, and the generation that inverted during the shortage

Any memory guide written this year has to start with the reason value has become hard to find, and this one is unusually well documented, so here it is with the dates and the attributions rather than as a vague warning.

On 2 February 2026, The Register reported that TrendForce had revised its own forecast sharply upward: from an expected 55 to 60% sequential rise in DRAM contract prices for the first quarter, to 90 to 95%, with PC DRAM expected to roughly double from the previous quarter and LPDDR memory seeing what the analyst called the steepest increases in their history. On 2 June 2026, the same title reported the outcome: contract prices for conventional DRAM had gone up by as much as 98% in the first quarter, with a further 58 to 63% forecast for the second.

The cause is not consumer demand. TrendForce's explanation is that incremental supply is being prioritised for high-capacity server modules destined for AI datacentres, which limits what reaches PC and smartphone makers. Hyperscale buyers accept the price rises, and everyone else follows in order to secure an allocation at all. On how long it lasts, the estimates diverge: the chairman of one of the three manufacturers has warned it could run to 2030, while other analysts put the end nearer 2027.

For a decade the rule was simple: the previous generation is the cheap one. In 2026 that has stopped being true for system memory, and it is the single most useful thing to know before you decide whether to upgrade an older machine or replace it.

Measured across the eight kits in this comparison, the DDR4 kits now ask more per gigabyte than the DDR5 kits do. The 32 GB DDR4 kit in this guide costs more per gigabyte than the 32 GB DDR5 kit that tops it, and it is a slower, older technology with lower bandwidth. That is not a quirk of one listing: it is what happens when a generation goes out of production during a shortage. Samsung and SK hynix have both been winding DDR4 down while the fabs that could make it are turned over to the memory the datacentres are buying, and a component nobody is making any more does not get cheaper.

What that changes, concretely. The traditional advice, keep the old board and put cheap memory in it, has quietly reversed. If your machine takes DDR4 and you need memory now, buy it now, because it is the generation moving fastest and there is no cheaper wave behind it. But if you were weighing an upgrade against a new platform, the memory is no longer the argument for staying: the gap between the two generations is narrowing from the wrong end.

This is the decision that gives away the most performance in the entire purchase and it does not cost a penny extra: at the same capacity, always two modules rather than one.

A motherboard talks to memory over two channels. With a single module it works in single channel and uses half the available road; with two matched modules it works in dual channel and uses all of it. On anything that moves a lot of memory, and above all on a machine using integrated graphics where the graphics share that same memory, the difference is more noticeable than raising the frequency would be.

That is why a 16 GB kit made of two 8 GB sticks sits ahead of a single 16 GB stick in this guide, and why the single Patriot module is here with one very specific job: adding capacity to a machine that already has a matching stick in it.

Searching for gaming memory has a short answer this year: 32 GB in two modules, DDR5 at 6000, with the lowest latency you can get. In games latency matters as much as frequency, and the CL30 of the Kingston kit in this guide is the best figure on the list, ahead of kits that ask considerably more. Sixteen gigabytes will still start a game, but the recommended requirements published for the big releases of the last two years already ask for that figure for the game alone, before the browser and the voice chat sitting behind it. And dual channel is not negotiable: two modules, always, which is the section above.

It is not a choice, it is a check. DDR4 and DDR5 do not go in the same slot and cannot be mixed: the notch on the module sits in a different place on purpose, so that it cannot be forced. What decides it is the motherboard, not the budget.

If your board is DDR4, the question has changed sign this year, and not in the direction you would expect. Moving from a 32 GB DDR4 kit to the DDR5 equivalent used to be a clear step up in outlay; the gap is closing, and it is closing because DDR4 is climbing rather than because DDR5 is falling. Adding memory to an older board is still sensible if the machine has years left in it. If you were already planning to retire it, do not pay a scarcity premium for a technology going out of production.

If you are building from scratch, DDR5 is what you buy and there is no argument. And if the machine is for work rather than games, the figure that matters is capacity rather than frequency: 32 GB at a moderate speed is a better machine than 16 GB at a fast one.

Everything published points one way. TrendForce's figures, reported through the first half of 2026, describe consecutive quarterly rises in conventional DRAM contract prices rather than any turn, and the reason it gives is structural: incremental supply is going to high-capacity server modules for AI datacentres, which limits availability for PC makers. There is no consensus on when it ends. One manufacturer's chairman has publicly suggested the shortage could persist to 2030; other analysts put it closer to the end of 2027.

So the rule stands and it is short. If you need memory, buy it. If you do not need it, do not buy out of fear. What makes no sense is the third option, waiting for a fall that no published source is forecasting. The sources, with who published them and when, are at the foot of this guide.