Think of RAM as a desk beside a filing cabinet (your SSD). Capacity is the desk's size: how many open files fit before you walk to the cabinet. Speed (MT/s) is how fast papers slide across the desk, and CAS latency is the pause before you start reading one. Channels are the number of hands working at once. Profiles such as XMP and EXPO are pre-tested instructions for working faster. Buy enough desk first, then speed, and only then chase tighter timings.
The memory standard, such as DDR4 or DDR5. Each generation has a different notch and will not fit the other's slots.
When it matters: Match the board, not only the CPU. AM5 and Intel LGA1851 boards take only DDR5. LGA1700 boards take either DDR4 or DDR5, never both.
The stick's physical format: DIMM for desktops, SO-DIMM for laptops and mini PCs, with a set pin count for each.
When it matters: Trap: laptop SO-DIMMs do not fit desktop slots and the reverse. Check the pin count and format against your device.
Whether the stick is unbuffered (UDIMM, used in desktops) or registered (RDIMM, used in servers and workstations).
When it matters: Trap: registered and unbuffered sticks are not interchangeable. A consumer desktop will not boot with RDIMMs, and most workstation boards need them.
Skip it if: You are buying for a normal desktop and the listing says UDIMM.
Which CPU platform the kit is tuned and tested for, such as AMD AM5 or Intel.
When it matters: Kits tuned for one platform usually work on the other, sometimes at a lower speed. AM5 runs best around DDR5-6000; Intel Core Ultra 200S tolerates faster kits.
Skip it if: The kit lists both XMP and EXPO and your board's support list includes it.
Whether Intel or AMD has validated the kit on their platform.
When it matters: Adds confidence that the rated profile will work. Your board's memory support list is still the best check.
Skip it if: The kit is on your board's memory support list.
The combined size of every stick in the kit, in GB.
When it matters: 16 GB is the minimum for a gaming PC and 32 GB is the comfortable standard. Video editing and 3D work want 64 GB or more. Extra capacity beyond your use adds no speed.
How the capacity is split: the number of sticks and the size of each, such as 2 x 16 GB.
When it matters: Two sticks fill both channels on a mainstream desktop. Four DDR5 sticks usually run slower. Trap: two separate kits may not run together at their rated speed, so buy one kit.
How many sets of memory chips each stick has: single rank (1R) or dual rank (2R). Larger sticks are usually dual rank.
When it matters: Dual rank can be slightly faster at the same speed but is harder to run at very high speeds, especially with four sticks.
Skip it if: You buy a common two-stick kit and run its rated profile.
The kit's advertised speed in MT/s (million transfers per second), reached by enabling its XMP or EXPO profile.
When it matters: Higher speed usually beats tighter timings. Trap: often listed as MHz, which is technically wrong. Kits faster than your CPU supports may not run stable at the rated speed.
The standard speed the stick runs at by default, before you enable any profile.
When it matters: Trap: a new PC runs the RAM at this slower speed until you turn on XMP or EXPO in the BIOS. Many buyers never do.
Skip it if: You will enable the profile, or the kit runs at stock speed anyway.
The delay, in clock cycles, between asking for data and getting it. Written as CL30 and similar. Lower is better at the same speed.
When it matters: Compare CL only between kits of the same speed. DDR5's higher CL numbers do not make it slower than DDR4, because each cycle is shorter.
The four main delay settings, written like 30-36-36-76: CAS latency, tRCD, tRP and tRAS. Lower numbers mean less waiting.
When it matters: Lower is better only when comparing the same speed. The gains are small in most games. Paying a lot extra for tight timings is rarely worth it.
Skip it if: You are choosing between kits of different speeds: speed matters more.
The real delay in nanoseconds, worked out as CAS latency x 2000 / speed in MT/s. Lower is better.
When it matters: The fair way to compare kits of different speeds. DDR4-3600 CL16 and DDR5-6000 CL30 land close, but DDR5 moves far more data per second.
Whether the kit stores an Intel XMP (Extreme Memory Profile), a pre-tested speed and timing setting you enable in the BIOS.
When it matters: Most AMD boards can load XMP too. Enabling it is still an overclock: if a new PC crashes, test with it turned off first.
Whether the kit stores an AMD EXPO profile (Extended Profiles for Overclocking), AMD's version of XMP.
When it matters: Best choice for AM5 builds. EXPO kits often run on Intel too. Trap: first boot after enabling it can take a minute of black screen while memory trains.
Skip it if: You are on Intel and the kit also has XMP.
How many preset speed profiles the kit stores.
When it matters: A second, slower profile gives a fallback if the fastest one is unstable on your system.
Skip it if: Your rated profile runs stable.
The voltage the kit needs to run its rated profile, in volts.
When it matters: Higher voltage means more heat and less headroom. Standard DDR5 runs at 1.1 V; fast kits need more, and the board sets it when you enable the profile.
Skip it if: You use the kit's profile as-is.
The standard voltage for this memory generation at default speed: 1.1 V for DDR5, 1.2 V for DDR4.
When it matters: It tells you what the stick runs at before any profile. Every stick of one generation shares it.
Skip it if: You are comparing kits of the same generation.
The maker and type of the memory chips on the stick, such as SK Hynix or Samsung. The stick's brand rarely makes the chips.
When it matters: Chip type decides how far a kit can be tuned. Trap: some kits change chips between batches without changing the part number.
Skip it if: You run the kit at its rated profile and never tune.
The capacity of each memory chip, in gigabits (Gbit). Higher density allows larger sticks with the same number of chips.
When it matters: It explains odd sizes: 24 Gbit chips make 24 GB and 48 GB sticks possible. Newer, denser chips often reach higher speeds.
Skip it if: You are not tuning memory yourself.
Error correction built into every DDR5 chip. It fixes errors inside the chip only.
When it matters: Trap: it is not full ECC. It does not protect data on its way to the CPU. Every DDR5 stick has it, so it does not separate kits.
Skip it if: Always: it is standard on all DDR5.
Full error-correcting code memory, with extra chips that detect and fix errors between the stick and the CPU.
When it matters: Worth it for servers, NAS builds and long compute jobs. The CPU and board must support it. Trap: ECC UDIMMs and RDIMMs are different products.
Skip it if: You are building a gaming or everyday PC.
The metal cover on the stick that spreads heat and adds looks.
When it matters: Fast DDR5 runs warmer because of its onboard power chip, so a heatspreader helps at high speeds. At standard speeds it is mostly cosmetic.
Skip it if: You run memory at standard speeds.
How tall the stick stands above the slot, in mm. Low-profile sticks are close to the bare board height.
When it matters: Trap: tall sticks can block large tower coolers. Check the cooler's RAM clearance, especially in small cases.
Skip it if: You use a liquid cooler or a compact tower with high clearance.
Whether the stick has lighting and which software controls it.
When it matters: Trap: running several lighting apps together can cause conflicts or odd behavior. Check that the kit works with your board's software.
Skip it if: You do not care about lighting.
The kit's class: value, gaming, overclocking, laptop, or server.
When it matters: A quick filter. Value kits run slower but reliably; overclocking kits use better chips for tuning.
Skip it if: You already know the speed and capacity you need.