RAM

RAM Timings, Latency and Memory Benchmarks

Learn how RAM timings and CL translate to latency in nanoseconds, how to read the four timings, and how to check speed, benchmarks, and stability for your kit.

On this page
  1. Latency in nanoseconds
  2. Reading the four timings
  3. Bandwidth versus latency
  4. Platform rules that limit speed
  5. AMD fast timing and XMP/EXPO
  6. Benchmarking your RAM
  7. Testing stability after tuning
  8. Tuning safely
  9. Common misconceptions

Two RAM kits can have the same speed and still respond differently, because latency depends on both the speed and the timings. This guide shows how to compare kits in nanoseconds, how to read the four timings on a label, and how to test a kit once you have tuned it.

Latency in nanoseconds

The first number on a kit label is CAS latency (CL), counted in clock cycles. Because the clock is half the MT/s figure, you convert CL to time with this formula:

First-word latency (ns) = CL × 2000 ÷ MT/s

Kit Latency (ns)
DDR4-3200 CL16 10.0
DDR4-3600 CL16 8.9
DDR4-3600 CL18 10.0
DDR5-4800 CL40 (JEDEC default) 16.7
DDR5-6000 CL30 10.0
DDR5-6000 CL36 12.0
DDR5-6400 CL32 10.0
DDR5-8000 CL38 9.5

The table shows why CL alone is not enough. DDR5-6000 CL30 and DDR5-6400 CL32 are both 10 ns. DDR5-6000 CL36 is 12 ns, which is 20% slower to the first word than the CL30 kit at the same speed.

Reading the four timings

Kits often list four numbers, such as 30-38-38-96:

  • CL (CAS latency): cycles from a read command to the first data.
  • tRCD: cycles to open a row before reading a column.
  • tRP: cycles to close a row before opening another.
  • tRAS: the minimum number of cycles a row must stay open.

tRCD and tRP count whenever an access needs a row that is not already open. Two kits with the same CL can differ by a nanosecond or more here. Tighter values are better, but each one must stay stable at the speed you run.

tRFC (the refresh cycle time) is a secondary timing with a large effect on performance. XMP and EXPO profiles often set it loose, and it is one of the first settings tuners change.

Bandwidth versus latency

Latency and bandwidth are separate. Bandwidth is how much data moves per second, and it rises with speed and channel count. Dual-channel DDR4-3200 tops out at 51.2 GB/s, and dual-channel DDR5-6000 reaches 96 GB/s. A higher-speed kit with a higher CL can have more bandwidth but similar first-word latency.

Platform rules that limit speed

The memory controller and the board set the practical limits.

  • AMD AM5 (Ryzen 7000 and 9000): the sweet spot is DDR5-6000 CL30, run with the memory controller clock (UCLK) at a 1:1 ratio with the memory clock (MCLK), and FCLK near 2000 MHz. Above about 6400 MT/s, the controller drops to a 1:2 ratio, and latency rises.
  • AMD AM4 (Ryzen 3000 and 5000): the sweet spot is DDR4-3600 CL16 at a 1:1 FCLK ratio (FCLK 1800 MHz). Going past the FCLK limit forces a 2:1 mode, which costs latency.
  • Intel: memory runs in a gear mode. Gear 1 is 1:1, Gear 2 is 1:2, and Gear 4 is 1:4. DDR5 runs in Gear 2 on 12th to 14th gen.

AMD fast timing and XMP/EXPO

XMP (Intel) and EXPO (AMD) profiles set the rated speed and timings in one step. Enable one in the BIOS, then check the result in CPU-Z or your BIOS.

AMD “fast timing” options are board-vendor presets that tighten the secondary timings beyond what the profile sets. The names vary by manufacturer, for example “High Bandwidth” or “Low Latency” on some Gigabyte boards, and “Memory Try It!” on some MSI boards. The presets are a starting point, not a guarantee, so test the result before you rely on it.

Benchmarking your RAM

A benchmark measures the memory’s speed and latency on your system. Common choices:

  • AIDA64 Cache & Memory Benchmark: read, write, copy, and latency numbers. It is a paid tool with a trial. Typical latency is roughly 55 to 65 ns on tuned Zen 3 with DDR4, about 60 to 75 ns on AM5 with DDR5-6000 CL30, and about 60 to 80 ns on Intel DDR5. Your numbers will vary.
  • Intel Memory Latency Checker: a measurement tool from Intel.
  • y-cruncher: a compute workload that is also sensitive to memory speed and timings.
  • ZenTimings: shows the AMD timings the system is actually using. The Ryzen DRAM Calculator, a common timing tool, is DDR4-only.

Compare results only under the same conditions: the same BIOS, the same profile, and no background load.

Testing stability after tuning

A benchmark pass does not prove that the tuning is stable. Run dedicated stress tests for hours, not minutes:

  • TestMem5 with a configuration suited to your platform.
  • OCCT memory test.
  • y-cruncher, using its stress test mode.
  • HCI MemTest and Karhu RAM Test (paid) are also common.

MemTest86 is a useful first check for hardware faults, but passing it does not prove that an overclock is stable. Use OS-based stress tests on top of it. For the basic error test, see how to test RAM for errors with MemTest86.

Tuning safely

Change one setting at a time, and back up your data before you start. Raise DRAM voltage in small steps. Many enthusiasts keep DDR4 daily voltages below about 1.45 V and DDR5 VDD around 1.40 to 1.45 V with good airflow, but the safe limit depends on the memory chips and the CPU. Check your kit’s and your CPU’s documented limits before you go higher. Unstable RAM can corrupt operating system files.

Gaming gains from tuned RAM are modest. Expect roughly 5 to 15% in CPU-limited 1080p games, and little difference at 4K, where the GPU is usually the limit. Results vary by game and CPU. X3D chips with large cache see less benefit.

Common misconceptions

  • “Lower CL is always faster.” CL only means something next to the speed. Compare nanoseconds.
  • “CL is the only timing that matters.” tRFC and the other subtimings can matter as much.
  • “A benchmark pass proves stability.” Run stress tests for hours.
  • “The DRAM Calculator works for DDR5.” It is a DDR4 tool.

The complete guide to ram

PC RAM Guide: Module Types, Specs and How Much You Need
Read the guide

Keep exploring

Guides from other topics

All topics →