Help me overclocking my GSkill Ripjaws 3200MHz CL 16 DDR4 RAMs.

[dvxtawa5lrwb]

8 GB each. 4 sticks on 4 slots, 32 GB RAMs in total.

Bought them as 2x8GB and 2x8GB. As you can see, the chips are different. 2 of them are Hynix, the other 2 are Samsung.

I know that even the models and brands are the same, using RAM sticks with different branded chips in them may cause issues. However, I'm using them for several years now and never had any issues with my system.

I normally run them using the default XMP profile. If you check the image, you will see that they are not running on ideal numbers. That's because I tried a few options I saw online to overclock them, failed, so I had to reset BIOS. In short, I am aware they are not running with XMP right now and I already fixed it when you are reading this post.

I literally have no clue about any number or adjustment when it comes to RAM overclocking. I would like to have a stable RAM overclock but it fails every time. If someone could lead me step by step, I would like to try your adjustments. Even it seems like the chips are terrible, I would like to try to make the numbers better and faster. I mean, trying doesn't hurt, right?

I would also like yoiu to answer this question additionally: As I said before, 2 sticks are Samsung and 2 sticks are Hynix. In what order should I place them? Right now they are Samsung - Samsung - Hynix - Hynix. If you say Samsung - Hynix - Samsung - Hynix will be a better choice, I can switch them. Choosing 1-3 or 2-4 slots are important, but does it also matter if the chips are different?

CPU: 5700X - No overclock, running at the default boost clock which is 4.6 GHz.
Motherboard: MSI B550-A PRO

 

[FreedomEclipse]

8 GB each. 4 sticks on 4 slots, 32 GB RAMs in total.

First step. Take out two sticks. Having all slots populated will stress the hell out of your CPUs IMC. lead to instability and generally not the best for OC. 2x8GB or 2x16GB is the way to go if you are serious about it.

Also - the two are different. One is running Samsung Chips while the other is Hynix. Hynix tends not to OC very well. Where as Samsung can be pretty good. You arent going to be able to OC very well with two different sets installed. never mind the 4x8 issue.

 

[ym2209]

Hello, I have a similar RAM set with a combination of Samsung and Hynix, here are some times that I could apply

 

[Nathaaab]

As was said, the ICs won't overclock the same, so you can't really do much with mismatched ram sticks.

 

[dvxtawa5lrwb]

What if I only go with the Samsung ones, what settings should I go for at this case?

 

[Nathaaab]

You need to identify exactly what they are, as Taiphoon sometimes doesn't report things correctly at all.
They are G.Skill, so you can follow this guide which can also help you for the overclocking part after :

MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

If there isn't anything that corresponds there, you can probably find out if it's B die by trying low tRFC that B die can usually do unlike other ICs, but be careful as this timing can lead to corrupted data easily.

 

[VT1185MF_Rosaria]

Your 1st set of G.Skill ram is either single rank 8Gbit Samsung C-die or B-Die as the ram die P/N corresponds to both Samsung C and B die ram. If you want to find out which is which, find the lowest trfc setting that it will run. B-die runs at very low trfc even at their bad bins (think of 400-260 trfc, the lower the better). If it is really B-die, then you need to test out the VDIMM scaling behavior of the ram sticks in conjunction with your IMC and motherboard.

If it is C-die, don't bother and use your CJR set instead. Sure, it has mediocre trfc timings (sometimes it can be double compared to good b-die) but compared to C-die CJR scales clocks and latency better. If you're lucky you can bring down CJR's trfc to 280-260.

For example on my CJR sticks, the timing sets I use are very pedestrian and conservative (I used trfc 560) . But these sticks booted at 260ns trfc although loosening some secondary and tertiary timings are necessary for it to work with acceptable stability as I didn't really attempt to stabilize and work it as a daily driver.

With how little vdimm voltage I used in mine I'm ok with it as I don't have to mind thermal limitations on my ram oc.

 

[dvxtawa5lrwb]

You need to identify exactly what they are, as Taiphoon sometimes doesn't report things correctly at all.
They are G.Skill, so you can follow this guide which can also help you for the overclocking part after :

MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

If there isn't anything that corresponds there, you can probably find out if it's B die by trying low tRFC that B die can usually do unlike other ICs, but be careful as this timing can lead to corrupted data easily.

Mine's code is 04320X8810B, so I can verify it's Samsung.

 

[SyWillCry]

Mine's code is 04320X8810B, so I can verify it's Samsung.

Use the Samsung sticks in slots 2 and 4. Enable XMP, but manually set tRFC to 300. Run a stability test. If that's stable for at least 10 mins then you can try go for a full generic B-die tune, though likely with worse primaries than standard as 3200 16-18-18 isn't exactly the best bin.

 

[ShrimpBrime]

Mine's code is 04320X8810B, so I can verify it's Samsung.

Its a very early kit of b-die. The XMP profile is 12 years old. It designates C die timings.

You can try to see if it scales with voltage. 1.50 to 1.55v and see if they go. If not, you treat it like any other bottom kit and run wider timings to gain the frequency. Like C18 might get you to 3466mhz better than hammering the v-dimm.

The more popular B-Die kits are known for voltage scaling. You unfortunately do not have the correct b-die kit to call it good unfortunately.

So if all 4 sticks run whatever timings and its stable, might be better to just do that and have a stable system with larger capacity.

 

[dvxtawa5lrwb]

C18 might get you to 3466mhz

Would that be better than 3200MHz CL16? I don't know the math behind it.

 

[ShrimpBrime]

Would that be better than 3200MHz CL16? I don't know the math behind it.

It would have more bandwidth at similar latency. So yes, slightly better. Its not going to produce extra fps really. Would improve total system performance when read write and copy from ram.

 

[dvxtawa5lrwb]

It would have more bandwidth at similar latency. So yes, slightly better. Its not going to produce extra fps really. Would improve total system performance when read write and copy from ram.

The default XPM profile is 16-18-18-18-38 with 3200MHz. Should I changes those 18s or just only 16 to 18 and 3200 to 3466 on 1.55v?

 

[Waldorf]

@dvxtawa5lrwb
really depends on what you are doing with your pc.
i would prefer just one kit, and to get a little more on the clocks maybe even if you lower CL for that, as ryzen 5xxx series is usually best around 1800 MHz IF/ram,
so you can run it 1:1.
but outside gaming, you might be better off with more ram using all 4 sticks with lower clocks.

i would make sure to run stock memory, and test until you are certain its stable with things like Memtest64/TM5 or HCI (1600% or more),
or you will waste a lot of time trying to get the tweak stable, when you cant pinpoint where its coming from (clock/timing/voltage).

i wouldnt try for max you can ever get out of b-die, as i doubt it will manage it even on 1.5.
go with 1.35v (1.45 max), change to 1800 for IF/ram, cl 18 and see if it works, then go with lowering timings, secondary can sometimes be better than one step on CL,
e.g. my kit does 3600C14/1.45, but i can do 3600C16@1.35, with lower secondary's, that outperform C14 while using a lot less voltage.

@ym2209
you can have the same identical dies, and it would still not make it 100% to run identical clock/timings/voltages,
just from difference in bios/board/temps etc, and its one of the worst ideas to tell ppl numbers, just because you have "similar" ram.

@VT1185MF_Rosaria
just because its b-die, doesnt make all b-dies the same.
enough differences were some will only gain on clocks, while some will only go for lower timings (without adding voltage).
almost anything thats known b-die (as in guarantee) is usually binned stuff, and either 3600C14 or 4000CL20, no matter the brand, so b-die coming with only 3200C16,
wont usually do more, or they could have sold the dies as 3600 or C14, not even talking that they would need different ProcODT settings, most boards will mess up on auto.

@SyWillCry
might wanna get your info up to date before giviong recommendations.
XMP is intel, not even guaranteed in any form to be stable past stock on amd, you want AMP (or A-XMP)certified kits, ignoring that even if it was,
AM4 is much better off using manual settings for clocks/timings/voltages for ram/IF/SoC.


edit to incl some tuning info.

 

[dvxtawa5lrwb]

@Waldorf As I mentioned before, my GPU is RTX 4060 and it is the main factor that my performance, FPS in my ca,se is being limited. Would making RAM adjustments really make a noticeable difference?

As you can see on the images I shared in the post, I bought 2 different 2X8GB kits and one of them has Hynix, the other one has Samsung. I feel like that 32GB is relaxing my experience. People said that Hynix is not going to be good for overclocking, so recommended me to use 2X8 (16GB) Samsung kit instead. I just wondered which option would give me faster and better stuff. I can of course do some benchmarks to see which one is giving me better results, but would like to hear it from you before trying it myself.

Hypothetically:

32GB RAM, 3266MHz, CL16. - Tried it and it's stable. Since I'm not experienced on RAM OC, I couldn't go further than increasing the voltage, which made me stay at these values.

16GB RAM, 3466MHz, CL18. - Didn't try it but it will be the first thing I will try. Maybe 3600, maybe 3333MHz, who knows.

Keep in mind that I just play AAA games on 1080p.

 

[Waldorf]

@ShrimpBrime
usually enough to worth spending time on tuning, even on gaming pc.
and thats on the older 3700x.

so yes, ram tuning can get you fps.

@dvxtawa5lrwb
personally, if i would keep the cpu/ram/mb for longer, sell the kits and get 2x16gb 3600C18@1/35v.
otherwise from some exceptions (like flight simulator), most of the time you are better off with fast 16gb than slow 32gb.
the sweetspot is 3600 (1800) MHz, so try to push clock as close, then lower rfc/secondary timings, and if any "one step more" results it being unstable, then start increasing vdimm to 1.4/1.45.
i wouldnt go for more than 1.45, unless you have no problems with temps (like cpu and gpu are LC), as you will just start having trouble above that for almost no gains vs 1.45.

 

[dvxtawa5lrwb]

If I had some money to spend on my PC right now, I would definitely go for a better GPU instead of buying another stick of RAMs, which should be understandable. But I will try 16GB 3466MHz, benchmark and see how the results are. Thanks for it.

 

[Waldorf]

4060 is fine for FHD, maybe get a vrr/gsync screen, makes enough difference to "smoothen" fps to be worth it.

 

[ShrimpBrime]

@ShrimpBrime
usually enough to worth spending time on tuning, even on gaming pc.
and thats on the older 3700x.

View attachment 406134View attachment 406136

so yes, ram tuning can get you fps.

@dvxtawa5lrwb
personally, if i would keep the cpu/ram/mb for longer, sell the kits and get 2x16gb 3600C18@1/35v.
otherwise from some exceptions (like flight simulator), most of the time you are better off with fast 16gb than slow 32gb.
the sweetspot is 3600 (1800) MHz, so try to push clock as close, then lower rfc/secondary timings, and if any "one step more" results it being unstable, then start increasing vdimm to 1.4/1.45.
i wouldnt go for more than 1.45, unless you have no problems with temps (like cpu and gpu are LC), as you will just start having trouble above that for almost no gains vs 1.45.

These charts simply prove my point? Are any of these games important to OP? He plays them?

4 to 8 fps, so 5 average, with a margin of error, at least 3fps, not to mention all the other variables, that seems like a whopping actual 2-4fps gain 3200 to 3600mhz.

Keep in mind, he has to OC to get there. Its not xmp vs xmp kits here.

Honestly, OP should just set it, play it and concentrate on the platform upgrade with a gpu in the future, its already EOL a couple years ago now....

I had a 4060. Nice little cards. They seem to OC relatively well.

 

[dvxtawa5lrwb]

Which one should I go with between Samsung C-Die and Hynix C-Die?

 

[ShrimpBrime]

Which one should I go with between Samsung C-Die and Hynix C-Die?

It won't really matter.

 

[Waldorf]

@ShrimpBrime
sure, at least you didnt say "none"

my case for "tweaking/tuning" is because all machines i "custom" build in the past 20y (no matter if at/outside work), and optimized everything, i gained ~10% overall perf,
vs the same parts used in builds by others.
i always like to get what i can out of the hw, without going overboard on it (incl time spend), and so far it paid off.

@dvxtawa5lrwb
given the biggest difference between the sticks are the dies, i would go with the samsung, i would expect it to be able to get more than stock/xmp out of it,
at least vs the hynix.