At matched capacity on 2026-09-22, DDR5 against DDR4 per gigabyte, cheapest kits: 32GB +44.6%, 64GB +113.5%. The claim holds at 0 of 2 sizes. The board decides the generation. Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy.
Live prices updated · refreshed through the day
Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy. Part of this market's demand is fear of higher prices, and buying on fear is how a buyer ends up owning capacity they did not need at the top of a price cycle. If your machine works and the memory in it is enough for what you run, the right amount to buy today is none.
On 2026-09-22, measured by us at matched capacity: 32GB, cheapest kit against cheapest kit: DDR5 costs 44.6% more per gigabyte; 64GB, cheapest kit against cheapest kit: DDR5 costs 113.5% more per gigabyte. So the claim that DDR4 now costs about as much per gigabyte as DDR5 does not hold at any matched capacity we can test: DDR5 is dearer per gigabyte at all of them. Using the median kit instead of the cheapest, it holds at 0 of 2.
The claim going around is that DDR4, the older memory generation, now costs about as much per gigabyte as DDR5. It is usually delivered as a fact about the whole market and it is usually illustrated with a single pair of kits. Both habits make it untestable, so the first job is to say what would make it true or false. We take the claim to mean: for a kit of the same total capacity, the DDR4 price per gigabyte is within a stated distance of the DDR5 price per gigabyte. We call a gap of 10 percent or less 'about the same', and we print that definition here so nobody has to guess where our line is.
Everything is measured at MATCHED capacity. A 32GB DDR4 kit is compared with a 32GB DDR5 kit and a 64GB with a 64GB, never a small kit with a large one. That rule is not pedantry. Comparing across capacities measures the mix of sizes on the shelf as much as the price of the memory, and the panel below shows the same comparison with capacity left unmatched so you can see how different the answer becomes.
Two instruments are used, and each is labelled wherever it appears. The shelf method compares the single cheapest in-stock kit of each generation at a capacity. The typical method compares the median kit. Both are levels, observations of the shelf today, and they are printed with the number of kits behind each so a comparison resting on two listings does not look like one resting on forty.
| Whose figure | What it is | Value / date |
|---|---|---|
| MEASURED BY US | 32GB DDR4, cheapest new kit per GB - 27 kits | $5.97/GB, 2026-09-22 |
| MEASURED BY US | 32GB DDR5, cheapest new kit per GB - 113 kits | $8.63/GB, 2026-09-22 |
| MEASURED BY US | 64GB DDR4, cheapest new kit per GB - 12 kits | $6.44/GB, 2026-09-22 |
| MEASURED BY US | 64GB DDR5, cheapest new kit per GB - 65 kits | $13.75/GB, 2026-09-22 |
| REPORTED BY TRENDFORCE DRAMEXCHANGE SPOT PRICING | The DDR4 1Gx8 3200 (8Gb) DRAM chip reached a record $42.50 in spot pricing on 10 August 2026, the highest since DRAMeXchange began tracking it in June 2016. |
Change over time, by the matched-model method: DDR4 desktop has no reading yet; DDR5 desktop has no reading yet. A shelf floor moving between two dates would measure catalogue membership as much as price, so no direction is printed from it.
Whatever the numbers say, the buying decision is not made by the price gap. DDR4 and DDR5 modules are keyed differently and will not fit each other's slots, so the generation is a property of your motherboard. If your board takes DDR4, DDR5 is not an option at any price without replacing the board and, on many platforms, the processor with it. If your board takes DDR5, DDR4 is not an option at any price. The physical facts are set out in DDR4 vs DDR5 and we link them rather than restate them.
So DDR4 only makes sense on a platform you already own. That is the framing the sources give and it is the honest one: a shrinking price gap between the generations tells you nothing about whether to move platforms, because the memory is a small part of what a platform costs. Nobody should change motherboards to save or spend a few percent per gigabyte on memory, in either direction. If you are choosing a platform from scratch, choose it on the platform and read the memory price as one line of a larger bill.
That is also why this page does not end in a recommendation to buy either generation. Its job is to say whether a claim is true, at which sizes, on which date.
The reported figure in the box below is a chip-level spot price, and it points the same broad way: DDR4 has become dearer, and by more than most people expect of an older product. What none of them can tell you is the per-gigabyte relationship between two finished kits at a given size on one retailer today, which is the only thing a buyer can act on. That is a shelf measurement, and it is the one this page makes.
Two features of the shelf make the headline story hard to reproduce from a single listing. First, the two generations are not sold in the same sizes: we track DDR4 desktop kits at 16GB, 32GB, 64GB and DDR5 at 32GB, 64GB, 96GB, 128GB, so only 32GB and 64GB can be matched, and the small DDR4 kits that pull an unmatched average down have no DDR5 counterpart to be compared with. Second, a cheapest-kit comparison is sensitive to one listing: one DDR4 kit priced a little under the pack moves that generation's floor at that size, which is why the median comparison sits beside it.
Our record for these kits starts on the date in the index panel and nothing earlier is reconstructed, so this page can say where the gap is and cannot yet say whether it is closing or widening. That question needs the matched-model index, which compares each kit against its own earlier price and is not affected by kits appearing or leaving the shelf. When it has two complete days it prints a reading here, per generation, with the number of matched kits, and the two generations' readings can be set side by side. Until then, any page that tells you the gap is closing is describing a direction from a single observation.
The other thing that would change it is supply. Reported statements from the two memory makers who have said publicly how long they expect the constraint to last are set out in the outlook, and they differ in kind; they are on the RAM forecast page rather than here, because this page is about a price relationship and the forecast page is about direction.
On 2026-09-22, measured by us at matched capacity: 32GB, cheapest kit against cheapest kit: DDR5 costs 44.6% more per gigabyte; 64GB, cheapest kit against cheapest kit: DDR5 costs 113.5% more per gigabyte. So the claim that DDR4 now costs about as much per gigabyte as DDR5 does not hold at any matched capacity we can test: DDR5 is dearer per gigabyte at all of them. Using the median kit instead of the cheapest, it holds at 0 of 2. That is the live answer on 2026-09-22, at matched capacity, comparing the cheapest kit of each generation; the page also shows the median comparison.
Because comparing across capacities measures the mix of sizes on the shelf, not the price of the memory. We track different sizes in each generation, so an unmatched average compares small DDR4 kits with large DDR5 kits. The page shows the unmatched figure next to the matched ones so the difference is visible.
That is not a price decision. The generation is set by your motherboard and the two do not fit each other's slots. DDR4 only makes sense on a platform you already own; if you are choosing a platform, the memory price is a small part of the bill.
Answering that needs a change over time, and the honest method for a change is the matched-model index, which compares each kit only against its own earlier price. Our record starts on the date in the index panel, so until it has two complete days we do not print a direction.
Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy. We do not predict prices; the outlook, with the dated statements from SK hynix and Micron side by side, is on the RAM price forecast.
It does show one, clearly labelled as reported and dated. But a reported market-wide or chip-level figure cannot tell you the relationship between two finished kits at one size on one retailer today, which is the thing this page tests on our own data.