How Much RAM Do You Actually Need in 2026?
16 GB is no longer comfortable, 32 GB is the sensible default, and 64 GB is a workload decision rather than a future-proofing one.
Capacity avoids stutter caused by running out of memory; speed affects how efficiently the CPU feeds data to the GPU. They solve different problems, and buying enough of the first one before spending on the second is the right order.
For a gaming build, buy capacity first and speed second β not because speed doesn't matter, but because the two solve entirely different problems, and getting capacity wrong causes a much worse experience than getting speed slightly wrong. Running short on capacity forces the system to swap to disk, which is a categorically bigger hit than any difference between two reasonably fast memory kits.
Capacity determines whether everything you're running fits in memory at once β the game itself, the browser and chat client most people leave open in the background, a recording or streaming tool, and the operating system's own overhead. When it doesn't fit, the system pages to disk, which is orders of magnitude slower than RAM regardless of what memory speed you paid for.
Speed and timing determine how efficiently the CPU can feed data through to the rest of the system once everything already fits in memory. This matters more for gaming specifically than for most other desktop workloads because a game's frame pipeline depends on the CPU keeping the GPU supplied with data, and memory bandwidth and latency both factor into how well it can do that. But this effect only shows up once the capacity question is already settled β a fast kit that's too small still pages to disk exactly like a slow one would.
32 GB, as a 2Γ16 GB kit, comfortably covers a current game alongside the browser tabs, chat app and recording tool that realistically run alongside it, and it happens to sit at the best price-per-gigabyte point on the market because it's the highest-volume configuration manufacturers build. 16 GB still runs a game on its own, but stops being comfortable once the realistic background load around it is accounted for β see how much RAM you actually need for the fuller breakdown across use cases, not just gaming.
Two modules rather than four is worth sticking to here specifically. A two-module kit holds its rated speed far more reliably than four modules on the same board, and reliably holding the rated speed matters more for the CPU-feeding-GPU pipeline than nominally having more slots filled.
Once 32 GB is off the table as a variable, the speed decision is platform-specific rather than universal. On AM5, the meaningful jump is reaching 6000 MT/s with CL30 timings, because that's the point where the memory controller runs synchronously with the platform's fabric β going faster than that can actually run worse once the controller drops to a divided ratio. On Intel's current desktop platform the tolerance is wider, and DDR5-6400 and above remain a reasonable target without the same ratio penalty.
Whichever platform you're on, use the true-latency conversion rather than comparing CAS numbers directly across speed grades:
True latency (ns) = CL Γ 2000 / data rate
A DDR5-6000 CL30 kit and a DDR4-3200 CL16 kit both work out to 10.0 ns β the CAS number alone would suggest DDR4 is faster, and it isn't. See CAS latency and true latency explained for the full breakdown of why the raw CL figure misleads.
Following that order avoids the two most common mistakes: buying a fast but undersized kit that still pages to disk, and overpaying for a speed grade the platform can't actually use efficiently. See best RAM for gaming for live kit picks built around exactly this priority order.
16 GB is no longer comfortable, 32 GB is the sensible default, and 64 GB is a workload decision rather than a future-proofing one.
Whether a laptop can be upgraded at all depends entirely on which of three physical memory formats it uses, and that has to be confirmed before comparing a single spec sheet.
On a fixed budget, RAM is one of the few components where the cheap, sensible choice and the expensive, marginal upgrade are easy to tell apart -- which makes it a good place to save, not a good place to cut.