Return to Level1Techs.com

Single- vs Dual Rank Ram and CL

#1

No clue if this belongs here.
Im looking at RAM for my upcoming R5 3600 build. Ideally i’d like 32GB. Prices Vary drastically though.

I’m looking at 3200Mhz Ram, as that seems like the bang-for-the-buck sweatspot. Can someone explain to me the difference between single and dualrank ram? Does it impact performance in any way?

Also, Latency… CL16 Dimm’s are around 75 bucks per 16GB. CL14 Dimms are double that. Is this really worth it?

0 Likes

#2

Single rank means the ram modules are on only one side of the stick. Dual means they are on both. Yes it matters for performance but varies from board to board. The manual for your board should have info about which ones to use where and at what speeds.

I’ll let an AMD employee give you the rest

3 Likes

#3

Disclaimer #2: The difference between DDR4-2667 and DDR4-3600C16 is about 8% on 3rd Gen Ryzen. If you have DDR4-3200C16 or something, be chill. You’re good. You might be 1% off from “best.”

This was all i needed really. Everything before that is more or less stating the obvious. Of course 3600CL15 is faster than 3200CL16.
Paying double for 1% in Performance isn’t worth it to me. And deciding between 1% plus Performance or half the RAM, i’ll take 32GB any day. Especially since i play Cities Skylines with Mods :wink:

1 Like

#4

Yeah, but it’s not obvious what that can mean for throughput. That’s kind of a big deal when you are using something memory dependant. Those workloads aren’t that common though

0 Likes

#5

As mentioned, single-rank sticks put all of their memory chips on one side of the stick and leave the other side empty. Dual-rank sticks have memory chips on both sides.

Twice as much physical real estate allows for the installation of twice as much memory, but requires that the memory controller do extra work in order to determine which side of the stick the desired data is on (or needs to go on). The paths then need to get switched (much like rail tracks get switched) before the data transfer can take place. This takes time… and ultimately means that dual-rank memory needs looser timings in order to function reliably.

A very small amount of latency is sacrificed for a very large increase in capacity.

If you’re playing “Benchmarking 2019 (ePeen Edition)” then paying double is mandatory, and you will be mocked mercilessly by members of the Samsung B-die cult if you don’t. It’s not worth it for “Cities: Skylines” (or virtually anything else for that matter). Your performance bottleneck is nearly always somewhere else in your system.

The best exception I can think of: Ryzen APUs. Because Ryzen APUs use main memory, latency translates directly to FPS. With 2200G/2400G on a 3xx/4xx motherboard, the Infinity Fabric was also tied to memory frequency. Folks gaming on an APU would probably stand to benefit most from (low-latency, high-frequency) single-rank memory. In 3rd Gen Ryzen processors, however, IF and MemFreq can be decoupled — so high performance memory isn’t quite as important as it was before.

1 Like

#6

RAM is quite the re-occurring nightmare to explain.

Ranking - Single/Dual:

  • This is mainly for compatibility reasons. Historically it was a huge problem, now not so much.

Latency and frequency:

  • RAM is manufactured the same way, what you see is silicon lottery and markup based on factory OC. (Stability is impacted as well…)
    Since DDR3 you do not have to worry too much about compatibility and you do not need to buy the exact module you need. This is a again a historical problem, where boards and CPUs could not handle profiles on the module and RAM would not work. This is the true benefit of DDR3 and XMP.

Performance and setup:
Frequency and timings are pretty much exchangeable. Faster speed is an improvement on both.
Ryzen favors clock speed over latency, while Intel goes the other way.
You have to remember that this is only true once you exceed native clock speed and for both Intel and Ryzen. So far the manufactures OC is focused more on Intel, but that will slowly change and you will see modules designed for 3200 a 3600 with better price distribution.

tl;dr once you reach native speed without 20+ CL you do not need to go further. Any gains will fall under the questionable OC category as they start to be less reproducible and harder to notice in complex solutions like games or workstation software.

1 Like