OC Corsair LPX CMK16GX4M2Z3600C18

[Jamie6870]

OC Corsair LPX CMK16GX4M2Z3600C18 question

After the OC of my CPU (i5 10600K) and video card, I get to the memory.
At least ten years ago I also did this
I have read the 101 but still have questions.
I have 2 x 8GB Corsair LPX CMK16GX4M2Z3600C18.
The settings are on XMP. And if I may believe it, Corsair's LPX has room for improvement.
If I understand it correctly with DDR4 RAM OC 101, XMP is already moving forward nicely.

My question is; does it make sense to apply an OC to this memory in combination with the i5 10600k?
Or is it really marginal?
But.. when there's more in it, I like to take it out as well.
(Sorry, English is not my native language...)
Greetings

Forgot... this is CPU-Z's screen dump

 

[EarthDog]

Or is it really marginal?

It's marginal at this point.

That said, it doesn't have to have a point to have fun.. if you want to tinker and see better benchmarks... you've come to the right place.

 

[Jamie6870]

Ok thanks.. Yes, fiddling around, getting better benchmarks, for me it's a kind of fun
So, if you,or someone else is willing to guide me.. what to read, or settings to look at, that would be great.

 

[Zerileous]

As ED said, it's a lot of trouble for a probably imperceptible gain, but if that's your cup of tea then it can be a great time. The idea is to make small incremental changes and test for stability. AIDA64 (has a free trial) memory test or Prime95 set to blend, then custom, then set to use 14GB of RAM will allow you to test for stability. A short test of 5-20 minutes is fine until you get to what you want to run, and then a longer test of a few hours to verify. You might have to back down a bit. Take a lot of notes (in this case, I prefer pen and paper because you can write while in the BIOS).

You can try to adjust some of the secondary timings as they are set by the board on automatic mode. You could try tRC=64, CR=1T, tRFC=460 just as ballparks. I would change one at a time in case things go haywire.

You could also try to lower the primary timings such as 18-20-20-38, again just a random ballpark. Or you could try to increase the clock speed. Or you could do a combination of all three (but work on one at a time). You can compare if tighter timings at 3600 is faster than looser timings at 3800 for example. There's also a memory latency tester in the AIDA64 program.

You can give the memory more voltage as well 1.4v is completely fine.

If you reach a problem and it seems like you can't get back to stable settings, first restore defaults, then enable XMP, and then start working from the last stable settings you had. You'll run into trouble if you try to just load a profile from an unstable setting due to things that go on in the background (i.e. memory training).

For some background theory, it's useful to look at a latency table. This helps you understand the relationship between clock speed and timings. The unit of a "timing" is clocks. So CL 16 is 16 clock cycles. We often refer to clock as a speed (because it influences directly how fast our PC does things), but more correctly it is a frequency. 3600MHz is really 1800MHz because the memory is DDR (Double Data Rate) so when you see the higher numbers, those are effective clocks, meanwhile the actual number of 1800Mhz or 1900MHz is the actual clock. The effective clock is double the actual clock. Depending on your program and BIOS you could see either one. You also may see MTs (mega transfers per second) used for the larger number. The higher the clock speed, the less time (measured in nanoseconds on the latency table) a single number in the timing is going to take. So naturally clock speed and timings have to increase somewhat together as physics and silicon quality are ultimately going to dictate how long it takes an electron to travel a certain distance. Well hopefully that's not clear as mud. Good luck and have fun!

 

[Jamie6870]

Thanks for your time and explanation!
I understand the marginal gain, but will look into it
Definitely fun to explore the possibilities.
Your explanation will definitely help and serve as a starting point to get started!
Takes some time, but I will keep an update.

 

[Jamie6870]

Immediately downloaded the AIDA64 trial version....
And saved the starting point. And I am posting the image below.
To be continued....
(latency, copy, write, read)

 

[Jamie6870]

When am I talking about speed gain...?
How many values difference can be made as a gain?

 

[Jamie6870]

*** deleted ***

 

[Woomack]

You won't make significant performance gains but you can play with memory timings. Try CL16 or CL17, higher DIMM voltages up to 1.45-1.50V (this is still safe), later maybe something with additional timings. Check tRFC at 100-200 lower values. The screenshot in earlier post shows 630 so try something like 520, maybe 480 ... maybe lower, I don't really know how this IC is acting.
If I'm right then Corsair is using the same memory chips in their DDR4-4000 kits so maybe try DDR4-4000/4133, probably at a higher DIMM voltage but XMP profile timings.

 

[Jamie6870]

Thank you.
I will set the values and let you know the results!

 

[schtm]

CMK16GX4M2Z3600C18

Here are my timings for the same kit. Please let me know if I can improve. For instance I can run the tRAS much lower, but I read it should be dependent on other timings. I can run in 4000 mhz but the timings get worse and the memtest is slower, especially the tFAW.

https://imgur.com/a/I5ilivv

 

[schtm]

Here are my timings for the same kit. Please let me know if I can improve. For instance I can run the tRAS much lower, but I read it should be dependent on other timings. I can run in 4000 mhz but the timings get worse and the memtest is slower, especially the tFAW.

https://imgur.com/a/I5ilivv

New results when maintaining MCLK=FCLK. Not sure if these are good or not, seems very similar and no big latency gains?