Question What makes higher RAM speed better?


Apr 16, 2019
So as far as I can google, ram does work PER a set of cycles.

If you divide 3200 by 16 ( you get 200) isnt exactly the same speed as 3600 18 (you also get 200)? Didnt the same exact amount of work get done?

If not, whats the difference, why is the higher speed better?

im mostly asking upon looking at DDR5 ram and find they have the same first word latencies as ddr 4 ram does, but much higher frequencies. Im trying to learn why the higher speed is better despite having the same first word latency and higher cas timings. (and im assuming its better because why would they change it from ddr 4 if it wasnt)


Feb 27, 2023
What makes higher RAM speed better?

In general go for speed. The speed of the ram is more important than the latency, because the latency is not a precise indication of the latency during system use - it also depends on your CPU and your memory controller.

If you divide 3200 by 16 ...

To compare the RAM, you should divide the latency with the speed, not the other way around. So if you are considering two different types of RAM, and the result of the CL/speed is almost the same, go for the RAM with the highest speed. If you use applications where latency must be as low as possible, then you should go for the sticks with the lowest CL.
Last edited:
Whether or not latency or speed is better is a little more complicated question to answer than one might think. But to try and summarize it in points:
  • Faster RAM (i.e., a higher DDR number & more channels) is better when
    • You have a task that is CPU dependent and is simple enough that the CPU can provide a result really fast
    • Data access is more or less sequential
    • Instruction access is infrequent and the instructions can likely stay resident in cache most of the time
    • An example of this is file compression/decompression. The de/compression algorithms are stupid fast for a CPU to do now because most of the algorithms were developed 20+ years ago and likely have been optimized to hecc. And you can technically work on the file in chunks, giving each CPU core a piece to do.
    • Faster RAM is also better when you have more cores and are doing a highly multithreaded task. While say dual channel DDR4-3200 might be fine for an 8-core processor, that's going to starve a 64-core processor.
  • Quicker RAM (i.e., a lower latency value) is better when
    • You have a mix of instructions to run or data to process... which is most of the time for a CPU.
    • Another point is your task requires instructions that doesn't all fit in cache, or is routinely evicted from it.
  • Like
Reactions: klavs
That assumes each time that RAM is accessed, only one thing is grabbed.

Latency only affects the first access to data in RAM. Subsequent accesses do not see that latency penalty so benchmarks that rely heavily on streaming performance such as unzipping or media decoding are pretty insensitive to latency and primarily dependent on memory bandwidth. Those are also the benchmarks that don't seem to improve with faster cache for the same reason.

This is also why GPUs use memory with absolutely terrible latency numbers as they are tweaked for memory bandwidth above all else. Rendering in realtime is definitely a very streaming type of memory access so latency doesn't matter, only speed does.
How ram is used differs between Intel and AMD processors.
Intel has a very good ram controller that can anticipate the need for ram data before it is requested. Ryzen, on the other hand depends on good speed for performance.

If you had a car capable of 100mph, it would take more than an hour to traverse the distance. That is because it might take 10 minutes to get up to 100mph.
A 95mph car might do better if it could get to 95mph in 10 seconds.
(I did not do the actual differential calculus equations, but you get the idea.)

The memory usage pattern will make a difference also.

Your first task in ram selection is to get enough.
After that, the details are harder to quantify.