RAM speed and latency explained
MT/s tells you how much data moves per second. CAS latency tells you how long each transfer takes. Both numbers matter — and you need both to compare kits fairly.
What MT/s means
The speed rating in MT/s (megatransfers per second) is the effective data rate — it is double the actual clock frequency. DDR4-3600 runs at a 1800 MHz clock; DDR5-6000 runs at 3000 MHz. Higher MT/s means more data can move per second, increasing peak memory bandwidth. This matters most for workloads that saturate the memory bus: video editing, 3D rendering, large-data compiling, and some games running at very high frame rates.
CAS latency (CL) and primary timings
CAS latency (CL or tCL) is the number of clock cycles between a read request and when the data is available. Primary timings are usually listed as four numbers: CL-tRCD-tRP-tRAS (e.g., 30-40-40-77). Lower numbers mean shorter waits per access. The absolute latency in nanoseconds is calculated as: CL ÷ (speed_MHz) × 2000 = nanoseconds. So DDR5-6000 CL30 = 30 / 3000 × 2000 ≈ 10 ns, and DDR4-3600 CL18 = 18 / 1800 × 2000 = 10 ns — equal absolute latency despite different generational labels.
The speed vs latency trade-off
Manufacturers often loosen timings to hit higher MT/s numbers. A kit marketed as DDR5-7200 CL36 may have worse absolute latency than DDR5-6000 CL30, even though the data rate is higher. For latency-sensitive workloads (competitive gaming at high FPS, low-level system calls), absolute latency matters. For bandwidth-hungry workloads, the raw MT/s number matters more. Check both numbers before comparing kits.
Secondary and tertiary timings
Beyond the primary four, dozens of additional sub-timings (tRFC, tWR, tFAW, etc.) affect real-world performance. Stock XMP/EXPO kits set conservative secondary timings. Manual tuners can tighten them for extra performance, but this requires testing stability with tools like TestMem5 or MemTest86. For most users, letting XMP set everything is the right call.
Practical rule
When comparing kits, calculate the absolute latency (CL ÷ clock × 2000) for each. A kit with lower absolute latency and comparable bandwidth is better for latency-sensitive tasks. If bandwidth matters more (encoding, rendering), prioritize MT/s. For gaming, balance both — DDR5-6000 CL30 and DDR5-6400 CL32 are very close and both excellent choices.