General RAM Overclocking

[TheJ187]

Hey guys, new to the forum and to overclocking in general. Have done mild CPU overclocking but that's about the extent of my knowledge. In the process of building a new rig and have been doing some research on different types of RAM. One thing I don't quite understand is the variations in the memory and the latency.

For example, the RAM I'm looking at is 3200Mhz with a CAL of 14. I am also looking at higher Mhz RAM that has a higher CAL. Given the equation to know the Ns response of each RAM, why would you go with the higher Mhz since technically they all (for the most part) are slower in regards to Ns? Am I missing something?

Equation i was given was CAL / Mhz *1000

Please excuse my ignorance if I'm just overlooking the obvious. Thanks

 

[Johan45]

Higher speed has higher bandwidth, depends what's important to you. Most Samsung based RAM will OC about the same regardless of being 3200 or 4000 MHz.
If you haven't decided on RAM yet just get something suitable for your system. In general use/gaming you're not going to notice a lot of difference in performance. For those who are heavily influenced by benchmarking competitively every little bit counts but we're also paying a premium most of the time to get the highest performance we can.

 

[TheJ187]

Ok I follow you. But doesn't the higher CAL that comes with the higher speeds cause it to technically be slower than say a slower speed with lower CAL?

3200 CAL 14

4000 CAL 18

 

[Johan45]

Higher latency but more throughput(bandwidth) different software responds differently some like bandwidth other prefer low latency

 

[TheJ187]

So it will take more data at a time but it just moves it slower? Am i gathering that correctly?

 

[Johan45]

More or less

 

[wingman99]

So it will take more data at a time but it just moves it slower? Am i gathering that correctly?

Memory speed is how fast the data transfers. CAS latency is the clock cycles, example 14 need before data transfer. Faster memory can transfer full speed when pipelining regardless of CAS latency.

Because memory modules have multiple internal banks, and data can be output from one during access latency for another, the output pins can be kept 100% busy regardless of the CAS latency through pipelining; the maximum attainable bandwidth is determined solely by the clock speed. Unfortunately, this maximum bandwidth can only be attained if the address of the data to be read is known long enough in advance; if the address of the data being accessed is not predictable, pipeline stalls can occur, resulting in a loss of bandwidth. For a completely unknown memory access (AKA Random access), the relevant latency is the time to close any open row, plus the time to open the desired row, followed by the CAS latency to read data from it. Due to spatial locality, however, it is common to access several words in the same row. In this case, the CAS latency alone determines the elapsed time. https://en.wikipedia.org/wiki/CAS_latency

 

[storm-chaser]

Yeah, generally speaking the faster the memory speed the higher it must be clocked.

For example a kit of PC100 (100Mhz memory) has a cas latency of 2 or 3 clocks (with timings like 2-2-2-5)
And a typical DDR3 kit running at 1600Mhz generally has a cas latency of 7 or 8 clocks (with timings like 8-8-8-24)
And a typical DDR4 kit running at 4000Mhz generally has a cas latency of 17 or 18 clocks ( with timings like 17-17-17-36)

Ideally, for overclocking and system responsiveness in general, you want to run the tightest possible timings at a prescribed speed which will give you the lowest possible latency result. So when I shop for memory I find a kit with the lowest possible cas latency *at the speed I want*

 

[Alaric]

Ideally, for overclocking and system responsiveness in general, you want to run the tightest possible timings at a prescribed speed which will give you the lowest possible latency result. So when I shop for memory I find a kit with the lowest possible cas latency *at the speed I want*

This^, although fine tuning depends on other factors. Pre Ryzen AMD chips, for example, respond better to tighter timings (comparing reasonably close speeds). On an FX or Phenom II I'd rather have 1600 MHz at 8-8-8 than 1866 MHz at 11-11-11. Intels like raw speed (generally, again) and timings are secondary to MHz, within reason. Ryzen changed that for AMD, as the Infinity Fabric runs at memory speed, with tighter timings a bonus but not as important as with previous AMD offerings.

 

[TheJ187]

Ok, I dont know if I worded my question right in the beginning or not so I'll ask again in a different way. All this information above is great and I've learned quite a bit. But, I'm building a new rig and at the same time starting to advance my basis on overclocking apart from the CPU side that I'm familiar with. So the below example is a better way of asking what I want to ask.

2 types of memory, different speeds, different latency. Which is technically a faster memory and moves more data?

G. SKILL TridentZ Series DDR4 3866 CAS 18

Corsair Vengeance RGB DDR4 3200 CAS 14

If you are supposed to use the equation to find out the "ns" speed of memory (CAS / Mhz *1000) then technically the 3200 is faster in those terms. So, that being said, its faster but with less bandwidth. So, is the slower ns with more bandwidth better than the lower bandwidth with a higher ns? That's what im not following

 

[Alaric]

In every scenario I can imagine, 3866 MHz CL18 is the faster of the two. And you can probably tighten up the timings on that set with a decent motherboard. My DDR4 3000 C14 is much faster at 3733 MHz C15.

 

[TheJ187]

does what you said not kinda contradict itself? You're saying in my case that the 3866 is faster but in yours the 3000 is when the comparison is pretty much the same?

 

[Alaric]

??? No, I said my RAM is faster at 3733 MHz than at 3000 MHz even though the CL is higher. Way faster. The memory OC was documented here https://www.overclockers.com/forums...866-MHz-Too-much-is-enough?highlight=3866+MHz But the Aida scores start in post # 56. 3866 MHz @18-21-21 is measurably faster (more bandwidth) than 3733 MHz @16-16-16 or 3600 MHz @14-15-15

 

[TheJ187]

Brb. Going back to school to learn how to read. Thanks for that

 

[Alaric]

LOL. No prob, happens to everyone on occasion.

 

[TheJ187]

lol agreed. But your example of what you have makes sense because it falls under the equation: (CAS / Mhz *1000) = Ns times

14/3000*1000= 4.67

15/3733*1000=4.02

Whereas my example that I'm stuck between doesnt, so do you still agree that the 3866 is faster?

14/3200*1000= 4.38
18/3866*1000= 4.66 even if i get it to 17 its still slower
17/3866*1000= 4.4

 

[Alaric]

I didn't use too many equations, other than the JEDEC standard and a couple others to determine the ratio between timing numbers. The main thrust of my OC was testing speeds/timings to get maximum bandwidth (outright speed). Latency spread between the three Aida64 tests in post#56 is 2.3 ns. 2.3 billionths of a second didn't have any negative effect on my bandwidth in my testing, and latency showed higher at 3600 CL14 (42.0 ns) than at 3733 CL16. My take on that was either margin of error or just the randomness of the benchmark. Background processes will have some effect, etc..

 

[Johan45]

lol agreed. But your example of what you have makes sense because it falls under the equation: (CAS / Mhz *1000) = Ns times

14/3000*1000= 4.67

15/3733*1000=4.02

Whereas my example that I'm stuck between doesnt, so do you still agree that the 3866 is faster?

14/3200*1000= 4.38
18/3866*1000= 4.66 even if i get it to 17 its still slower
17/3866*1000= 4.4

First I think you're getting speed and performance mixed up. Faster RAM (ie 4000 MHz Cl 18) will always be faster even though performance may be similar with some lower speed RAM that has tighter timings (3200 MHz Cl 14). DDR4 performance also depends on sub-timings so don't be too focused on CL.
For most users it's really not important which RAM they get as long as they choose a relatively moderate speed 3200-3600 with reasonable timings. Running high-speed RAM with tight timings is for benchmarkers like myself since at the end of the day any gains you see will be mostly in synthetic benchmarks. For general use/gaming RAM is pretty much RAM, you're going to get the most gains overclocking your CPU.

 

[TheJ187]

Ok, that makes sense. I appreciate the info