• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

DDR4 RAM overclocking 101 guide

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
Good guide batboy.
I am just wondering if the following process I am going to describe is good or not as a first step for memory overclocking:
Lets assume we have a memory module with 14-14-14-34 timings @ 3200MHz.
As far as I know the latencies are provided in terms of clock cycles. Which means in case of 14 the latency is 1/3200MHz*14Cycles = 8.75ns which is a pretty good timing. If we want to use the memory module @ 4000MHz than the latencies shall be set (as a starting point) to 1/3200MHz*14Cycles*4000MHz = 17.5 ~ 18 and 1/3200MHz*34Cycle*4000MHz = 42.5 ~ 43. So @4000MHz first I would set the timings to 18-18-18-43. This setting shall be stable, and - of course - the voltages shall be raised, as well as you have mentioned. Then I would do a performance test to check if it is really stable or not. And as a next step I would lower the values by 1 and do the performance test again. I would do this until I find the minimum timings in case of each values.

In case of your memories the process gives -lets say - right result.
You have 15-15-15-35@3600MHz which (with the given formula) is 18-18-18-40@4200MHz. And thanks to your effort we know that those modules work with 17-17-17-36@4200MHz. It means that by lowering timing value by 1 we can get the thoughest timing in 7 steps which is not too much, I guess.

What do you think, would this process work?
 
It's a good starting point but not guaranteed to work. I would start with 19-19-19 leaving everything else on auto and work down from there with 1.35- 1.4V.
 
I don't know exactly what's going on and some help would be appreciated.

I just build a new workstation. Actually my first AMD build.
A 1950X on a ASRock X399 Taichi with 4x 8GB HyperX Predator 3000MHz.
I have some issues trying to deal with both CPU and RAM OC.

For the RAM I was trying to keep everything on "auto", just changing its XMP. So the speed goes from [email protected] to [email protected] (its real speed). I guess it's fine but I think it can go beyond that speed with some fine tune adjustments.
I was geting BSoDs but I'm 99% sure that the problem was a weird combination of PSU (a 8 years old Corsair HX750 Gold) and somehow a corrupted benchmark app that was running the stress tests. I don't know if was Aida64 or EVGA Precision XOC, both crashed.

I replaced the PSU for an later version of that Corsair with the Platinum label and feel it's more stable to deal with OC.

I'm affraid to run more stress tests cause I believe if I get more crashes or BSoDs I can end up with some corrupted Windows file.
Here is the result: https://valid.x86.fr/2iprxl

For the concern the CPU is cooled by an Enermax Liqtech TR4 II 360.
 
Hello!


I Got this thread reccommended by a user from Overclock.net and came seeking for guidance and knowledge.

Problem: I recently get 2x8GB sticks of G-Skill Tridentz 4000MHz 17-17-17-37 1.35v, but I'm struglin to make the fully stable, as I'm having random app crashes and BSODs.

My system:

I9 9900K: 5.2GHz (Delided + Direct Die Contact) | Uncore: 47GHz Vcore: 1.34v (fully stable and never pases 84ºC)
Z390 Aorus Master | F9b Bios (latest)
VCCIO: 1.25v | VCCSA: 1.30v
DDR4 Speed: 4000MHz 17-18-18-38
RAM Voltage: 1.45v

Whit this settings i'm able to boot, get into Windows, install progrmas, run a few lopps on Cinebench R15... but when I open Chome/mozilla and 2 or 3 tabs it sudenly closes, trying to open any Adobe App or game, fails miserably giving BSOD.

I have also tried RAM Volt: 1.60v | VCCIO: 1.35v | VCCSA: 1.35v but it was even worse, causing it to fail Mem. Training / Boot, increasing main timmings (like 18-18-18-38 | 19-19-19-39 ) wouldn't help either...

Worth to mention that XMP is not working at all, and the only thing that stabilized the memory (so far) was using some Subtimings that I found on internet from a group of guys that where doing extrme OC to DDR4 with Samsung's B-Die kits, but since I don't undertand how do subtimmings relate with eachother every attempt to change any value caused boot fail, and I can't see the logic behing the subtimmings... It's so frustrating because it look so close to be fully stabl.... seems like it's just some stupid setting that I must be missing to take care of.

Any idea? perhaps some one can refer me to a guide about timmings On B-Die...


Many thanks in advace.
 
For the DRam voltage did you try anything between 1.35 and 1.6V overvolting Dram can make it unstable as well you need to find the "balance" point. I would try resetting BIOS to defaults and only set the XMP with SA at 1.35v and IO at 1.3 V and start working from there. Having that OC on the CPU could be part of the problem so you need to isolate the Memory, get it stable then OC the CPU
 
(...)
I have also tried RAM Volt: 1.60v | VCCIO: 1.35v | VCCSA: 1.35v but it was even worse, causing it to fail Mem. Training / Boot, increasing main timmings (like 18-18-18-38 | 19-19-19-39 ) wouldn't help either...

Worth to mention that XMP is not working at all, and the only thing that stabilized the memory (so far) was using some Subtimings that I found on internet from a group of guys that where doing extrme OC to DDR4 with Samsung's B-Die kits, but since I don't undertand how do subtimmings relate with eachother every attempt to change any value caused boot fail, and I can't see the logic behing the subtimmings... It's so frustrating because it look so close to be fully stabl.... seems like it's just some stupid setting that I must be missing to take care of.

Any idea? perhaps some one can refer me to a guide about timmings On B-Die...


Many thanks in advace.
Dude, I'm not an expert but if you want to make your system stable or even to last, never push RAM voltage beyond 1.45V
I guess 1.6V could burn your mem sticks.
 
Hi guys! thanks for being that fast :)

@Johan45:

Yeah I tryed almos everything... deafult settings, CMOS, only OC'n the RAM...I tried almost every voltage between 1.35 and 1.6v... no matter what I do XMP won't even boot, but i'm well aware that dut to the T-tipology 2 RAM stick are harder to OC than 4. but it should be Stable al some point, even with ridiculous Voltages, wich leads me to think that is not voltage related, looks to be mote timmings issue.

@Quiron:

B-Die is setted at 2v as maximun voltaje (Maximun as Extreme overclock on air, but not reccommended for 24/7) I used 1.6 just to give it a try, but i'm not planning to run it that high every day.
 
There is no clear info on how high is max safe voltage for any IC, also B-die. JEDEC specs are 1.20V +/-10%, but as we know since DDR4 premiere, these modules can work 100% stable without issues up to at least 1.5V, regardless of manufacturer. Somehow there are overclockers reporting random dead sticks at 1.7V+. B-die naming doesn't mean that all IC are the same. 2V+ is what "extreme" overclockers are setting and usually not under full load, with limited RAM capacity and for a short period in benchmarks.

If you are using 2x8GB higher binned B-die, then there is nothing easier to overclock. On four memory slot motherboards have to use 2/4 memory slot counting from CPU side.
If XMP is not working, then it's a BIOS/motherboard issue. Gigabyte is known to fail BIOS, and they eventually make something good or drop support at all. However, there are higher frequencies on the QVL so I would check if XMP works on earlier BIOS versions (still supporting your CPU). I bet they tested it on early BIOS release and later messed something up.

In a typical situation, you can set 4000 and all timings+additional voltages at auto except VDIMM which you have to set manually at 1.35V. Then motherboard should set timings within its timing table. Then you can also check if its stable at 4000. Once you stabilize it at any timings, then try to tighten the timings and check if it still works stable. Usually, tests like AIDA64 memory+cache (only these 2 picked from stability test) are showing memory and IMC stability. When something is clearly wrong, then it crashes once loads ~97% of RAM, so in 1-2 mins.

If you are using command rate 1N at 4000+ then it can be an issue as most motherboards, even these top OC series, won't work above ~3733 at 1N. Depends on for what you are using your PC, you may find 4266-4400 at more relaxed timings like CL18/19 or 3600 at tighter like CL15/16 better.

VCCIO and VCCSA highly depend on the CPU but in a typical situation you will need 1.35V SA and 1.25V IO for ~4266-4400. Max safe will be something like 1.40V SA and 1.35V IO. There is no guarantee on a safety but OC series motherboards are scaling up to these values and I just doubt that manufacturers would set higher voltages knowing it may burn something.
My 9900K was below the average and it could still make 4500 with 1.40V SA and 1.30V IO.

I have no idea if your Gigabyte has memory OC profiles. If there are any then you can check various combinations like lower settings with manual higher memory clock. It may save some time on testing.
 
Last edited:
@Woomack:

Thanks a lot for sharing your experience, actually it is not the first time I see users reporting stability gains going from 4000mhz to 4166 and avobe. And only time I setted 1.6v it didn't passed Mem.training memory wasn't under 100% load... I just wanted to check if there was some kind of scalation between Freq./volt/ and stability.

I've emailed RAM support team, perhaps they can give some advice or point me into the right direction, because as I said, I have little experience about setting timings and high freq. RAM OC'ing... if I try to improve the stability by changing the timings I'll be playing a very ugly lottery.

Just for those who might be curious, here are current settings:

CAS: 17
tRCD: 18
tRP: 18
tRAS: 38
tRC: 54
tWR:24
tCWL: 16
tRRD_S: 6
tRRD_L: 10
tWTR_S: 8
tWTR_L: 15
tCCD_S: 6
tCCD_L: 10
tRFC: 450
tRTP: 15
tFAW: 48
C. Rate: 2
tREFI: 15600

I've notice that when running MemTest64 errors come up only at "Random Tests" all other doesn't show any error.

As for the VCCIO and VCCSA I have maxed both to 1.35v and 1.36v respectively, but it didn't give any inprovement, I'd say that it was quite the the opposite, mi CPU's IMC likes to be arrund VCCIO: 1.18v - 1.25v and VCCSA: 1.28v - 1.35v
 
Hi again guys.

Yesterday I tried some faster steps just to see if there could be any stablity improvement, but I didn't, failed to boot 4266, 4200, 4166 and 4100, however I also tryed XMP again and it worked, but wait, it wasn't stable at all, system would crash a few secods after launching MemTest64 without even BSOD... VCCIO|SA RAMv & and timings were all on auto. HWinfo was reporting 1.320 for IO / 1.366 for SA and 1.35v for RAM.

I also tryed lowe speed steps like 3900 and 3866MHz but with little increments on Voltage for VCCIO|SA and RAMv, starting from 1.180 up to 1.35v and 1.45v for RAM... didn't work either. Best resould I achieved is the pofile that I already posteed above where at leat I can run MemTest64, is gives many errors but doesn't crash.

Any idea?
 
Have you tried running memtest at default to check that the memory is OK
 
Have you tried running memtest at default to check that the memory is OK

Indeed! memory runs just fine with all on Deafult and no XMP o modified Freq./Voltaje not a single error on Memtest64 nor Memtest86


I've been seraching, reading every post and study every information about each timing and wich ones are related to others. I'm now trying to create a few tables on Excel as "guides" to see if I can get some other good point based on logyc (instead of just trying someone else settings), or at least something stable avobe 3800MHz. I still don't understand how do work tCWL and tCCD (both _S and _L). could some one try to tell me?

Also I've seen that tWR can be:

- tWR: 2 clicks above tRRD (2+tRRD_S) or;
- tWR: tRRD plus tWTR (tRRD_S + tWTR_S)

But wich one would be better?
 
Last edited:
@Woomack;

Answering to your quetions and tips:

I would be more than happy to just set the XMP settings working fully stable, I'm not pretending to brake any world record at 4500+ MHZ :p
The thing is that any of my resoults looks to make any logic or sense, only reasonable explanations could be 1- faulty kit (quite rare, but possible) 2- one or more settings on the mem confinguration are just totally wrong.

Yeah, I'm pretty aware that Gigabyte BIOS's aren't the best (that's me being 'polite' here), as I got Dual Bios I'll be flashing one of the chips backwards on different versions until the dates of MoBo & Mem coexisted first time on market. that should be where Gygabite and G.Skill tested compatilibilty with eachother and therefore the same BIOS version, but I'm afraid to loose all the extra stability that pushed my CPU to 5.2GHz... I'll see...

Mi 9900K looks to be a pretty one, not only for the Cores OC but the thing that i'm able to activate XMP 4000MHZ 17-17-17-37 running VCCIO: 1.32~ and VCCSA 1.364~ (being that on Auto settings) and get into OS leads me to thing that with proper timings they could be even lower. BTW: VCCSA: 1.4v isn't little bit craizy? just above 1.35 MoBo colours it in Yellow lit if it where going to take off...

Thanks for your time!
 
@alv-oc reading your statement I realized that my first mem kit was "awarded". One of four HyperX Predator 3000MHz (Hynix CJR) dead after a few attempts of soft OC'ing. I didn't went beyond 1.35V or under CAS 15 and I ended up wit 24GB at 2933MHz.
I've just ordered another kit (now 3200MHz) and I'll try everything from the begining.
 
Yesterday, after conducting 4 hours of tests, different BIOS, many settings..... I realised that I was so deeply focused on making this RAM kit work that I forgot that I had almost same issue with my previous RAM kit (Corsair Dominator Platimun 3000MHZ CL15), I had a very hard time making it work at it's rated settings. Gigabyte support team gave some good assistance back then (3 months ago), but problem solved almost by it's self, one day XMP just worked, I didn't give it more attention since then...

So I'm pretty sure that there must be something wrong at hardware level, most probably with the board but as @Quiron said RAM can also be faulty. That Corsair kit should't have had any issues since the beginning in first place but as I was using them on my previous Pc there was the 'reasonable doubt'... it could be RAM issue?? well, it's 'almost' impossible to get 2 faulty RAM kits, specially the G.Skill wich are hihgly and carefully binned chips, however I'm going to RMA the RAM kit anyhow, the only way to find out the problem is checking in everything, easiest way to start is RAM, I can get 1 cheap module just to keep my system runing meanwhile. If G.Skill prooves that both sticks ar just fine then I would be sending the MoBo to RMA, if Gigabyte can also proove that MoBo is fine (wich I doubt based on a few experiences on Gigabyte's forum...) then I would have to sent the CPU to the retailer, but this would be worst scenario since I bouhgt a Delied version (delided by them, not me)

Ooh Lord... what a pain in the @®$€... this is going to take ages.... :( I hope that it doesn't take so long that Intel starts selling the HDET 10nm :cry:
 
I spent a solid five consecutive days dialing in my RAM overclock. I found out a long time ago that getting greedy too early in that process will bork your operating system. LOL

Yeah I know, I had to instal Windows 3 times during the last 7 days (BTW: I got original working key, it si not Cracked version). But for that matter I use MemTest86 that works from a bootale USB and does't need OS and avoids the "inestabilities" that it might cause to the memory while runing test. But it doesnt matter, MemTest86 starts giving hundreds of errors at as soon as starts test Nº6...
 
SSDs are so affordable right now, and OEM product keys are cheap, so it's worthwhile to have an OS on a 60-120 GB SSD laying around for projects like that. Don't need updates, just your monitoring software, and just plug it in and OC the snot out of everything. If the OS gets corrupted you can be back in business in minutes with an .iso from your main boot drive.
 
So this is it.
Yesterday I changed my HyperX Predator 3000MHz C16 with one faulty stick for the same brand and model but 3200MHz and I'm glad it's stable since the installation at 2400 as well as 3200MHz.

Here is the SAFE preset I used from Ryzen DRAM calculator:

23wfg7x.png

------------------------

I got good read/write values but lack of latency.
I ran MemTest for more than 3 hours with no errors.

mFnMO2Q.png

------------------------

I also tried the FAST preset but it didn't post (mobo beeps):

2pul3L1.png

Just to share with you guys what I did. Any advices will be appreciated.
 
Hi guys

I just got reply form Gigabyte support team, they toñs me that they could replicate the same issue just as I said, thay also said that they already scalated the issue to the BIOS team.

good new is that looks lile I wouldn't need to RMA the MoBo but bad new is that they wouldn't tell me how log can it take to release a fix, perhaps I got lucky and they include it on the next official release that is about be released at the website, oherwise I will be waiting months (spectations baswd on the past released BIOS updates)

Thats pretty frustrating
 
Back