Optimizing RAM for i5-10600K

The point

I want to determine if it is worth to buy a new RAM kit for my system or if it is better to OC my current sticks.

SPOILER: I will not buy a better kit, but instead try to overclock the RAM I own.

The new kit in question: Kingston FURY 32GB (2x16GB) 3600MHz CL16 Renegade Black
https://www.kingston.com/dataSheets/KF436C16RB1K2_32.pdf

The system

Intel i5-10600K
MSI Z490 Tomahawk
Crucial 4x8 GB 3200 MT CL16

Two kits with model no. BL8G32C16U4B.8FE and BL8G32C16U4B.M8FE1 respectively.

obraz

Why this thread?

A comment made on this forum about optimal RAM for 12th gen sparked a thought, and It wouldn’t let go of me. 3600 MT CL16 being the sweet spot for Alder Lake. But what about 10th gen?

I have found some reddit comments stating that if I am not going straight to 4000 MT CL16 it is not worth the hassle, and that 3200 MT CL16 performs reasonably well. Am I missing out on performance?

I run a Timespy benchmark only to find out that my system is 400 points below the average of 10600K+RTX2060 systems.

What now?

Should I overclock my current sticks or buy the “better” kit?
If attempting OC, what speed and timings should I target? Are there more settings apart from clock and timings? Gear 1/2 are for newer platforms, but what about Command Rate?

I will be thankful for any kind of response!

Invoking the experts
@GigaBusterEXE @MisteryAngel

After skimming through this article:

I just bumped the MT from 3200 to 3600 without changing anything else.
System booted normally.

Memory upgrade for the sake of clock+latencies is basically never worth it. It’s a nice to have thing. But more capacity = always good. DDR4 is cheap these days considering the high prices on most other stuff

2 Likes

New interest unlocked: Overclocking

I don’t have the original test run with 3200 MT CL16, but it was something around 7290.
Test pass with 3600 MT CL16:

3600 MT CL16 left, 3200 MT CL16 right

I will NOT buy the new kit. I will try to surpass the benchmark average by OC’ing.

Thanks for chiming in. The difference between 3200 and 3600 is really negligible.
As for capacity, I am totally comfortable with the 32 GB I have as I do not exceed 24 GB while using my computer.

what ever the max native speed is of the cpu’s mem controller is the sweet spot for ram.
if your mem controller tops out at 2133 then the best ram to use for that system will be the lowest latency 2133 you can find.
if its 3766 then again 3766 ram with the lowest latency…

if you choose to use ram that takes advantage of a motherboard oc… say 3200 ram oc while your cpu nativley maxes out at 2133, you will trade off capacity for speed.
in that if you have a 2 channel 4 slot motherboard that you would normally fully populate with 2133, you be will limited to 1 dim per channel rather than native 2.
meaning you can only use 2 sticks of ram where you would have normally been able to use 4.

your max on the cpu mem controller is 2666 so 2666 max if you want to use more than 2 sticks.
if you want to use 3200 then get 2 high capacity sticks rather than 4 lower capacity. (so 2x16 if you want 32gig rather than 4x8gig)

1 Like

I would run you ram at 1.39v for extra stability while overclocking
Lowing the command rate is faster

I believe the Uncore is the freq your imc and L3 cache run at, could be wrong haven’t messed with 10th gen since last year

You might see a difference in score simply by the software

If you have epic game store, Uplay, origin, GoG ect in your task bar to the right exit them and rerun

Even steam uses a surprising amount of cycles, you can try small mode and it gets rid of the web elements in steam and uses the classic launcher from the early 2000’s

1 Like

Thanks for the tips.

To clarify, I run the 10600K stock.
The PC is stable at 3600 MT CL16 (with all four sticks). For now I will not mess with Command Rate (2T by default), I will stick to frequency and timings.

My understanding is, that the IMC runs at the speed of RAM.
IIRC the cache can be overclocked seperately and is not tied to IMC/RAM frequency. I will check the GN video about it.

I’ve normalized for that, both test passes were made with fresh boot and only steam in the background running the Timespy Demo. But you are right I should do several runs to confirm a boost/regression. Didn’t spent much time on it today.

What programs do you recommend to validate the OC?
Prime95 comes to mind.

Edit: Oh boy, what have I embarked on.

sorry autocorrect changed uncore to incorrect I fixed it in my post and your quote

1 Like

With stock/auto voltages I was able to boot 4 sticks at maximum 3600 MT CL16. Then I adjusted tREFI to 65535 (max value).

The latency improvement is 19% compared to baseline (according to Aida64).
CPU Score on Time Spy improved by 4,2% compared to baseline.

I used the average of 3 passes for each test.

Maybe next weekend I’ll play with voltages. It is a fun learning experience, but nothing worth doing.

For sure it’s not worth to buy expensive RAM. Buy a better GPU, goddammit!

or more cores

If you mess with turbo settings you can overclock through undervolting
Cache ratio is something to definitely look into

1 Like

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.