Two identical-looking DDR5 kits sit on your desk. Both are 32 GB. Both have heat spreaders with aggressive angles. One costs $40 more. The sticker on the first says DDR5-6000 CL30. The second says DDR5-6400 CL36. If you're like most people, you grab the second one because 6400 beats 6000, right?
Wrong. And this is the exact mistake I made when I built my first DDR5 system back when the platform was still rough around the edges. I paid extra for the higher number and got a kit that responded slower in real workloads than the cheaper one my friend bought. That afternoon taught me more about RAM than three years of reading spec sheets.
Here's what nobody puts on the box: the MHz figure is a throughput number, not a speed number in the way your brain reads it. What actually determines how quickly your CPU gets data back is latency, and latency is a calculation, not a sticker.
Key Takeaways
- RAM "speed" (6000 MT/s) is bandwidth. RAM "latency" (CL30) is response time. They are different animals.
- Real latency in nanoseconds = CL × 2000 ÷ data rate. That's the only number worth comparing across kits.
- A 6000 CL30 kit beats a 6400 CL36 kit in real latency, despite the lower number on the box.
- DDR5 kits running under ~10 ns are good. DDR4 kits under ~9 ns are good. Anything above 12 ns is leaving performance on the table.
- If your 6000 MHz RAM runs at 3000 MHz, it's almost always because EXPO/XMP was never enabled in BIOS.
RAM speed and latency explained: the two numbers that actually matter
When a manufacturer prints "DDR5-6000" on a module, that 6000 is measured in MT/s (megatransfers per second), not MHz, though the industry still sloppily calls it MHz. It tells you how many million data transfers the stick can push per second. Higher equals more bandwidth. That's it.
CAS Latency, or CL, is the number of clock cycles the memory waits before it answers a request. Lower equals faster response. But here's the trap: cycles happen at different rates depending on the data rate. A cycle at 6000 MT/s is shorter than a cycle at 3200 MT/s. So comparing CL30 to CL30 across two different speeds tells you nothing.
The formula nobody puts on the box
To compare kits honestly, convert everything to nanoseconds:
true latency (ns) = CL × 2000 ÷ data rate (MT/s)
- DDR5-6000 CL30 → 30 × 2000 ÷ 6000 = 10.0 ns
- DDR5-6400 CL36 → 36 × 2000 ÷ 6400 = 11.25 ns
- DDR4-3200 CL16 → 16 × 2000 ÷ 3200 = 10.0 ns
Notice anything? The DDR4 kit from years ago matches the fancy DDR5 kit on latency. What DDR5 buys you is bandwidth, not responsiveness. This is the single most misunderstood point in PC building, and I'll happily argue about it over a beer.
First word latency, put simply
When people say "first word latency," they mean exactly what the formula above describes: the delay between the CPU asking and the memory starting to answer. Once that first word arrives, the rest streams out at the bandwidth rate. Think of it as a librarian. Latency is how long it takes her to find the right shelf. Bandwidth is how fast she can hand you the books once she's there. A fast librarian at a small shelf beats a slow one at a warehouse.
What is a good latency speed for RAM?
For DDR5 in 2026, aim for 10 ns or lower. That means DDR5-6000 CL30, DDR5-6400 CL32, or DDR5-7200 CL34 if your motherboard and CPU memory controller can actually run it. Above 12 ns and you're paying for bandwidth you probably aren't using.
DDR4 is a different target: under 9 ns is excellent, which means DDR4-3200 CL14 or DDR4-3600 CL16. DDR4-3200 CL22 (the JEDEC default, what you get if you never touch BIOS) sits at 13.75 ns. That's not a small gap.
| Kit | Data rate | CL | True latency | Verdict |
|---|---|---|---|---|
| DDR5-5600 CL46 (JEDEC default) | 5600 MT/s | 46 | 16.4 ns | Slow, fixable in BIOS |
| DDR5-6000 CL30 | 6000 MT/s | 30 | 10.0 ns | Excellent |
| DDR5-6400 CL32 | 6400 MT/s | 32 | 10.0 ns | Excellent, more bandwidth |
| DDR5-7200 CL34 | 7200 MT/s | 34 | 9.4 ns | Top tier, CPU-hungry |
| DDR4-3200 CL16 | 3200 MT/s | 16 | 10.0 ns | Solid for older builds |
Is 30 or 36 CAS latency better?
It depends entirely on the data rate attached to it. CL30 at 6000 MT/s gives you 10 ns. CL36 at 6400 MT/s gives you 11.25 ns. So CL30 wins on latency, but CL36 at the higher speed wins on bandwidth. Which matters more comes down to your workload, and there's no universal answer.
For gaming, latency usually edges out bandwidth on the CPU-bound titles. For video export, compression, and large database workloads, bandwidth tends to pull ahead. On my own machine I tested both profiles on the same stick, and in CPU-heavy strategy games the CL30 profile gave me smoother frame times. In a batch photo export, the higher-bandwidth profile finished a couple of minutes sooner on a 40-minute job. Both matter. Neither dominates.
What does 40 40 40 77 mean in RAM?
That four-number string is the primary timing set, always listed in the same order: tCL – tRCD – tRP – tRAS. Each one controls a different stage of a memory access. You don't need to memorize all four to buy good RAM, but understanding them stops you from being fooled by marketing.
- tCL — the CAS Latency we've been discussing. Cycles before the memory answers a read request.
- tRCD — Row-to-Column Delay. Time between opening a row and reading a column inside it.
- tRP — Row Precharge. Time to close the current row before opening a new one.
- tRAS — Row Active Time. Minimum time a row must stay open. Roughly tCL + tRCD + a few cycles, and it's the least impactful of the four in day-to-day use.
So "40-40-40-77" describes a fairly loose kit. Probably a DDR5-5600 or DDR5-6000 JEDEC profile with conservative timings. Tighten tCL and tRCD and you drop latency significantly. Leave them loose and you get stability at the cost of responsiveness.
Why is my RAM running at 3000 MHz instead of 6000 MHz?
You bought DDR5-6000. Task Manager says 3000 MHz. Nothing is broken. Two things are happening.
First, DDR stands for Double Data Rate. It transfers data on both the rising and falling edge of the clock signal. So a stick running at 3000 MHz physically moves data at 6000 MT/s. Every tool that shows "MHz" is showing you half the transfer rate. This trips up almost everyone on their first DDR5 build.
The real culprit: EXPO/XMP not enabled
If Task Manager or CPU-Z is reporting something like 2400 MHz or 2800 MHz (half of 4800 or 5600), then your RAM is running at the JEDEC baseline default. Every DDR5 stick boots at 4800 MT/s until you tell the BIOS to do otherwise. The higher profile is stored but dormant.
- Reboot into BIOS/UEFI.
- Find the memory profile setting. AMD calls it EXPO. Intel calls it XMP. Some boards call it DOCP or just "Memory Profile."
- Enable the profile that matches your kit's rated spec.
- Save and exit. Confirm in CPU-Z that the "DRAM Frequency" reads half your rated MT/s.
Beyond that, a few scenarios cap you below your rated speed even with the profile on: four sticks instead of two, dual-rank modules, a weak CPU memory controller, or a motherboard that hasn't validated your specific kit. On an AM5 board I set up for a friend last year, four sticks of DDR5-6000 refused to run above 5200 MT/s no matter what. Two sticks ran clean at 6000. That's just how the platform behaves.
Should I just buy the fastest kit my motherboard supports?
No. Beyond a certain point you're paying a lot for bandwidth you won't feel, and risking instability if the CPU's memory controller can't handle the speed. For most builders, a DDR5-6000 CL30 kit hits the sweet spot. Spend the leftover budget on more capacity or a better GPU.
Putting it together
The next time you're staring at a product page with a wall of numbers, do one thing: calculate the true latency. CL × 2000 ÷ MT/s. Ten seconds of mental arithmetic will save you from buying the wrong kit and, more importantly, from wondering for months why your expensive RAM doesn't feel faster.
And if you ever see someone confidently claim that higher MHz always wins, you now know better. You don't have to correct them. Just smile, run your own numbers, and enjoy your smoother frame times.
