13th Gen Ram OC Question

[TalonRahl]

Calling all Intel/ASUS experts!

Long story short.
I am experiencing micro stutter. I want to increase my 1% lows to resolve this.
I am having a hard time finding proper terminology for ASUS motherboard voltages. I only recently learned that 12th gen and 13th gen supposedly dont have a VCCIO voltage? So much new stuff to learn coming from AMD.
I also am having a hard time finding DDR4 voltage STARTING recommendations for overclocking RAM for a 13th Gen Z790 ASUS TUF PLUS WIFI D4 board....
I have been pulling my hair out for weeks trying to fix this stutter issue. I have narrowed it down to ram and upgraded ram, but now I don't know how to properly get a safe overclock for it. I know my stutter is related to my 1% lows because they are bad from my benchmarks. Now I am just looking for one or more lovely individuals to help me figure out what specific voltages I need to tinker on this ASUS Intel board to get it to be stable and not fry the memory controller.
Here is RAM code

F4-4000C18D-32GTRS

I downclocked to 3800mhz so I could tighten timings to 17 20 20 38. Though I feel there is potential for a better tRFC and tREF I just left them as auto in BIOS since I didnt want to mess with them before figuring out voltages. Intel XMP 4000mhz 18 22 22 42 was unstable and slow. It is SK HYNIX die so I know I wont be able to get something like 16 16 16 36 or anything, but I know that I should be able to achieve the clock and timings I set with proper voltage as its not too crazy of an OC. tRFC and tREF recommendations are also welcome.

More info in document. If any of the pictures in the document are hard to read I will reupload into this post. I appreciate you my potential Intel/Asus expert. You are my last hope.

https://docs.google.com/document/d/16oKsmjVkB98EDPY1qfksI8GKbmKs1EqRublkDScMOqQ/edit?usp=sharing

[RONOTHAN##]

I'll preface this by saying most of my 13th gen OC experience is on DDR5, not DDR4, though I have some experience on DDR4, some experience with ASUS boards, so I should be able to help enoguh. 

 

5 minutes ago, TalonRahl said:

I only recently learned that 12th gen and 13th gen supposedly dont have a VCCIO voltage?

It's more or less replaced by the VDDQ TX voltage, or what ASUS likes to call IVR Transmitter for whatever reason. On pretty much every board but ASUS, you set this to 1.35V and call it a day. On ASUS boards for whatever reason though (at least the DDR5 versions), it likes to be higher, so play with it higher to try and get high speeds working. I'd want to keep this voltage below 1.45V for long term stability, though some people say 1.55V is safe.

 

9 minutes ago, TalonRahl said:

tRFC and tREF recommendations are also welcome.

tRFC will depend on the specific Hynix bin (usually it will need to be 500-600 from what I know though, granted I primarily tune B die so don't know for sure), though tREFI can be set as high as is stable. Generally it's 33k if you have a hotbox of a case, 64k if your case has decent airflow, and 128k if your case has good airflow and a fan pointed directly at the memory. You stress test this with the GPU running a stress test of some sort since your GPU is likely to cook your RAM at the same time. 

 

If you want a better answer for this, either look directly on the label for the number above the barcode as that will tell exactly what Hynix IC it is, or use something like Thaiphoon Burner to read the SPD and report back with said report. 

 

14 minutes ago, TalonRahl said:

https://docs.google.com/document/d/16oKsmjVkB98EDPY1qfksI8GKbmKs1EqRublkDScMOqQ/edit?usp=sharing

Scrolling through here, the main thing I noticed is the MC PLL voltage. 1.26V is waaaay too high, you shouldn't really set this much above 1.1V, and even that is likely to be too high as this voltage sweet spots (my 13700K likes it at 0.99V). Lower this immediately, since from what I know that's high enough to damage the chip. 

 

You have the VCCSA voltage set to 1.3V, which is probably fine, though be aware that some CPUs stop scaling with it. Mine scaled to 1.45V for max frequency, so that's not a given, but it's not unheard of for settings above 1.2V to give issues. 

 

Other than that, you can only really play with VDDQ TX, VCCSA, and the VDIMM voltage to try and get higher frequency to work. There are some boards I'm aware of that have very specific VDIMM requirements for high frequency, and it's very possible you have one that needs something like 1.45V exactly to work at 4000MT/s. I'd experiment first at 3800MT/s, see what causes the system to become unstable, then work on trying to get 4000MT/s to work. 

[leclod]

47 minutes ago, TalonRahl said:

I have narrowed it down to ram

How ? I doubt that diagnose (And I own DDR4 3600cl14)

By the way did you run a Memtest86 ?

Edited February 14 by leclod

[TalonRahl]

4 hours ago, leclod said:

How ? I doubt that diagnose (And I own DDR4 3600cl14)

By the way did you run a Memtest86 ?

I narrowed it down to the ram because of the event viewer errors I was getting that after replacing every single piece of the build EXCEPT for the ram was still there, but also my overall latency was way too high for the 3200mhz cl16 ram that I had. Replaced the ram and no more error. I would go back to find the error and can, but at the moment now I am mainly just trying to learn how to properly OC my ram without frying the memory controller in my chip. Doing this because I know that tuned ram can really improve 1% lows specifically rather than average fps and in so doing if I bring my 1% lows up I am less likely to have microstutter. As I said my stutter problem isn't readily apparent because its just a microstutter. Only really able to tell when panning the camera in games and the lack of smoothness. The only way to get it smooth is to framerate cap my games with Rivatuner. Though whats the point of having the kind of build I do if I just have to cap the fps. 
So now I am learning how to specifically OC ddr4 ram with a 13th gen intel chip. I am more used to AMD boards and there terminology and also OCing ram for amd is different due to the fabric. Only reason I am even posting on a board is because nearly every single person that is OCing ram for 13th gen is using DDR5 so most of the guides out there have been for that. And for the most part that doesnt help me as the voltages slightly differ from what I have read. 

Also memtest86 passing 4 cycles isn't going to find every error that can occur with ram. My ram that I replaced ran through memtest fine. It wasn't until I tested it with OCCT that it even threw any errors. And that was at stock XMP. Which again is why I replaced the ram which I believe I said in my post. 

Anyways since from those timings(3600CL14) I can tell you most likely have B die ram I am not 100% that you will be able to help me with my OC since mine is hynix. 
But even though your original reply was to doubt the efficacy of my diagnose. I would still humbly listen to any other input you may have. 
I am here to learn and appreciate all input. 

[TalonRahl]

5 hours ago, RONOTHAN## said:

It's more or less replaced by the VDDQ TX voltage, or what ASUS likes to call IVR Transmitter for whatever reason. On pretty much every board but ASUS, you set this to 1.35V and call it a day. On ASUS boards for whatever reason though (at least the DDR5 versions), it likes to be higher, so play with it higher to try and get high speeds working. I'd want to keep this voltage below 1.45V for long term stability, though some people say 1.55V is safe.

 

That makes a lot more sense and I had adjusted that value in my original OC. I think I had it set to 1.36. I think I will leave that there while I mess with the VCCSA and DRAM voltage and set them while I test for stability most likely with OCCT. Though if you have any other stability testers that wont fry the ram in the process I would appreciate anymore input there too. 

 

 

5 hours ago, RONOTHAN## said:

tRFC will depend on the specific Hynix bin

After trying this thaphoon burner it actually was unable to specify the die for me. 
Though through some research and also looking at the sticker you referenced that has a C at the end of the code. I have now pretty much confirmed its CJR. And from what I can tell on the forums that is a good thing as its not a bad die to have. Not samsung b die, but still not terrible to overclock. It is 16 gb 1 rank. I am not sure what or how that would effect anything either as I have only ever had dual rank before. 

Based on this die and this rank do you know of any documentation that would help specify the best tRFC to run or "recommendation"? 

 

 

 

 

5 hours ago, RONOTHAN## said:

Generally it's 33k if you have a hotbox of a case, 64k if your case has decent airflow, and 128k if your case has good airflow and a fan pointed directly at the memory. You stress test this with the GPU running a stress test of some sort since your GPU is likely to cook your RAM at the same time. 

Is this something you can change without changing anything else and just testing base on how hot it makes your ram? I presume you don't want anything above 50 C? 

 

 

5 hours ago, RONOTHAN## said:

Scrolling through here, the main thing I noticed is the MC PLL voltage. 1.26V is waaaay too high, you shouldn't really set this much above 1.1V, and even that is likely to be too high as this voltage sweet spots (my 13700K likes it at 0.99V). Lower this immediately, since from what I know that's high enough to damage the chip. 

I plan to just put this back to auto as now that I know more so what it is from more research I don't think I will need to adjust it for the type of OC I am going for. Unless I see some sort of random instability that only needs a minor adjustment to fix. 

 

5 hours ago, RONOTHAN## said:

You have the VCCSA voltage set to 1.3V, which is probably fine, though be aware that some CPUs stop scaling with it. Mine scaled to 1.45V for max frequency, so that's not a given, but it's not unheard of for settings above 1.2V to give issues.

Isn't 1.2v stock auto? 
I will try to see if with the 3800mhz 17 20 20 38 timings if I am able to boot with 1.2v and just bump it up by increments of .01 or whatever its min is till I find a stable one. I initally had a really hard time figuring out what voltage in the asus board WAS the SA voltage due to finding that pll voltage in the submenu. Though now knowing that that is I will leave that at auto. I only had it at that voltage for a very short time before reseting everything to stock xmp all auto. So I "hope" my chip is fine. Yikes if I messed that up. Intel/asus is not my strong suit. 

 

 

6 hours ago, RONOTHAN## said:

Other than that, you can only really play with VDDQ TX, VCCSA, and the VDIMM voltage to try and get higher frequency to work. There are some boards I'm aware of that have very specific VDIMM requirements for high frequency, and it's very possible you have one that needs something like 1.45V exactly to work at 4000MT/s. I'd experiment first at 3800MT/s, see what causes the system to become unstable, then work on trying to get 4000MT/s to work. 

So do you think knowing now what kind of die I have that a settings below is a good "starting" point to find stability? And do you believe with knowing my die that maybe I could even bring in some of my primary timings further? Personally I know even less about secondary and tertiary timings to even want to mess with more than just the tRFC and tREFI. 

VCCSA 1.2v
VDDQ 1.36
VDIMM 1.43 

With clock and timings 
3800mhz 17 20 20 38 
tRFC 550 tREFI 64k

Thank you again for even taking the time to read through and respond to my post. 
Its MEGA appreciated. I have tried posting in TomsHardware, Reddit, ASUS and here. And so far you have given me the more helpful of information thus far. 
This is also my first time even really posting in forums like these as I have for the most part been able to figure out any sort of other OCing info on youtube. Though with RAM it seems like the rabbit hole gets quite deep. 

 

 

[TalonRahl]

Also not sure how useful this other info might be from Thaiphoon Burner. 

[leclod]

1 hour ago, TalonRahl said:

But even though your original reply was to doubt the efficacy of my diagnose. I would still humbly listen to any other input you may have. 
I am here to learn and appreciate all input. 

Do I understand correctly, you have 2 problems : micro stutter and errors ?

You solved the errors by replacing the Ram.

I understand the relation between micro stutter and 1% lows.

But are your 1% lows in line with what's expected for your hardware ?

 

"but also my overall latency was way too high for the 3200mhz cl16 ram that I had"

How do you measure it ? I remember not being pleased with what I got either. I concluded there's no pleasing latency value to be had.

We could compare if you tell me how. I don't suffer micro stutter as far as I know.

(I still humbly think you're chasing the wrong horse.)

 

Finally did you look into such forums :

 

Edited February 14 by leclod

[TalonRahl]

10 minutes ago, leclod said:

Do I understand correctly, you have 2 problems : micro stutter and errors ?

You solved the errors by replacing the Ram.

I understand the relation between micro stutter and 1% lows.

But are your 1% lows in line with what's expected for your hardware ?

 

"but also my overall latency was way too high for the 3200mhz cl16 ram that I had"

How do you measure it ? I remember not being pleased with what I got either. I concluded there's no pleasing latency value to be had.

We could compare if you tell me how. I don't suffer micro stutter as far as I know.

(I still think you're chasing the wrong horse.)

 

My Aida64 latency initially was upwards of 70ns. 

With just the few things I have tried with this new ram I have gotten down to 54ns. Though this still feels high. 
Though before doing any testing in games I wanted to make sure the ram was stable and to do that I need more OC knowledge. So I havent tested to see yet whether the microstutter was gone. 
In games specifically like say Cyberpunk for example i had USED to get 1% lows of around 60ish fps maxed on everything including path tracing with frame gen. 
Though now a days I am getting more so around 35 to 40 in my 1% lows during testing of that old ram I have since replaced. Pretty severe decrease. 

Again at this point I am mainly here for recommendations of overclocking rather than figuring out the stutter. 
ALTHOUGH.... I could be open to suggestions you might think of on what has caused the microstutter and I can tell you whether or not I have tried to fix that. 
I will preface this by saying my bios is upto date. My chipset is upto date. My drivers are all upto date. 
I have replaced the motherboard with same model to see if that was the problem. I have replaced the cpu with the same model to see if that was the problem. I have changed to a way better and more efficient psu than originally had. I have also gotten an AIO to reduce temps on my cpu thinking that this also could have been the issue. I don't overclock the gpu at all. I slightly undervolt the cpu but only -.04 on vcore adaptive. I have tried different ssds and also made sure the slots used were using the proper spec. I do not put my nvme on the cpu specific lanes of the board just to circumvent any issue with the graphics lanes. I have all of my specs in my signature now for ease of access. 

I am up for more ideas. If it resolves my microstutter I simply dont care what or how I have to do it. At this point. I will do whatever it takes. I appreciate the time you have taken to honestly reply at all. 
So far this forum has been the most helpful or responding that I have tried. 

[leclod]

8 minutes ago, TalonRahl said:

My Aida64 latency initially was upwards of 70ns.

I didn't buy Aida so I can't meet you there.

 

Did you look into such forums :

 

[TalonRahl]

5 minutes ago, leclod said:

I didn't buy Aida so I can't meet you there.

 

Did you look into such forums :

 

Okay I will take a look at that.

Also. 
Kind of wondering if this random find that I found of some samsung b die gskill 3600 cl14 ram is the ram you have? 
I might buy it. Looks like it would be exactly what I am looking for. And I wouldnt really have to try and do anything to it. 

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820374093

[leclod]

3 minutes ago, TalonRahl said:

Kind of wondering if this random find that I found of some samsung b die gskill 3600 cl14 ram is the ram you have?

Yeah, seems to be mine.

I don't recommend buying though. I didn't notice any improvement after buying it.

I'm not saying they are bad or less than they should be, I'm saying Ram isn't an issue.

I'm no expert though, just my opinion based on texts I read.

[TalonRahl]

@leclod Your potential fix unfortunately doesnt relate to what I am experiencing because in the game where I have the most micro stutter I already reduce the renderer count cores used to 7 which effectively does the same thing as what you are referencing. 
Not to mention also but the 13700k has a larger L3 cache and suffers less issues of oversaturation that the 12700k did. 

That is a good idea of a potential fix. Though unfortunately not one I think will work for me. 
But just for the sake of trying. I will try it anyway. Why not right?

[TalonRahl]

2 minutes ago, leclod said:

Yeah, seems to be mine.

I don't recommend buying though. I didn't notice any improvement after buying it.

I'm not saying they are bad or less than they should be, I'm saying Ram isn't an issue.

I'm no expert though, just my opinion based on texts I read.

Most people unless they are benching their 1% lows wont notice an improvement at all. 
But lower latency nearly always boosts them by a considerable margin depending on the difference in latency you get. 
I will buy you an Aida64 code if you would be willing to try it. Its like 2 dollars. Though I would need a way to PM you it. 

[leclod]

Just now, TalonRahl said:

I will buy you an Aida64 code if you would be willing to try it. Its like 2 dollars.

No no, 2 dollars I can do. Let's see...

[TalonRahl]

1 minute ago, leclod said:

No no, 2 dollars I can do. Let's see...

G2A is what I used. 
https://www.g2a.com/aida64-extreme-key-global-i10000049796001?uuid=b544b282-2444-4e04-a777-127dbf54c5bf

[leclod]

I got it working for 0 dollar with a random key, thank you anyway.

So now, what do I do ? Edit : I think I got it...

 

Edited February 14 by leclod

[RONOTHAN##]

1 hour ago, TalonRahl said:

Based on this die and this rank do you know of any documentation that would help specify the best tRFC to run or "recommendation"? 

Somewhere around 550 should work if I remember correctly for that IC. 

 

1 hour ago, TalonRahl said:

Though if you have any other stability testers that wont fry the ram in the process I would appreciate anymore input there too. 

Y-Cruncher with the VT3 stress test for high frequency, TestMem5 with the Anta777 Extreme1 preset for timings is what I use on DDR4. 

 

1 hour ago, TalonRahl said:

Is this something you can change without changing anything else and just testing base on how hot it makes your ram? I presume you don't want anything above 50 C? 

More or less, yeah. You still do need to test it, because there 100% are kits where that rule doesn't really apply, but they're quire rare. 

 

1 hour ago, TalonRahl said:

Isn't 1.2v stock auto? 

1V IIRC is stock for VCCSA. 

 

1 hour ago, TalonRahl said:

So do you think knowing now what kind of die I have that a settings below is a good "starting" point to find stability? And do you believe with knowing my die that maybe I could even bring in some of my primary timings further? Personally I know even less about secondary and tertiary timings to even want to mess with more than just the tRFC and tREFI. 

VCCSA 1.2v
VDDQ 1.36
VDIMM 1.43 

With clock and timings 
3800mhz 17 20 20 38 
tRFC 550 tREFI 64k

Yeah, those are a decent starting point. Odds are you'll end up with all of those voltages higher if you're going to go for 4000MT/s or higher (16Gb CJR does do very high clock speeds, it might be worth ditching Gear 1 and trying to max out the kit in Gear 2).

 

7 minutes ago, TalonRahl said:

Most people unless they are benching their 1% lows wont notice an improvement at all. 
But lower latency nearly always boosts them by a considerable margin depending on the difference in latency you get. 
I will buy you an Aida64 code if you would be willing to try it. Its like 2 dollars. Though I would need a way to PM you it. 

If you want to go the free route, Intel MLC is basically a free version of the Aida memory test that's a bit more consistent, albeit it only works on Intel. It's a little more janky to setup as you need to download the GUI off of github, but it does work really well. https://github.com/FarisR99/IMLCGui

[leclod]

10 minutes ago, TalonRahl said:

G2A is what I used.

That's what I got. I guess you'll tell me it's bad

I can't run 3600 Gear 1 somehow, I did read somewhere the memory controller on my cpu is no champ. So I run it at 3466

[TalonRahl]

10 minutes ago, leclod said:

I got it working for 0 dollar with a random key, thank you anyway.

So now, what do I do ? Edit : I think I got it...

 

The memory and cpu test 
I am looking to see what your number is here. 
If you hear a squeel I have confirmed in multiple forums that its normal and is sometimes just the vrm and will cause no damage to the board. 

[leclod]

7 minutes ago, RONOTHAN## said:

If you want to go the free route, Intel MLC is basically a free version of the Aida memory test that's a bit more consistent, albeit it only works on Intel. It's a little more janky to setup as you need to download the GUI off of github, but it does work really well. https://github.com/FarisR99/IMLCGui

I remember getting BSODs with Intel MLC + GUI

[RONOTHAN##]

Just now, leclod said:

I remember getting BSODs with Intel MLC + GUI

Yeah, pretty sure there was a version that had that bug. I got those same issues until I reverted to an earlier version. Haven't checked in a few months though if that bug is still in place, though the older version should still work. 

[TalonRahl]

2 minutes ago, leclod said:

That's what I got. I guess you'll tell me it's bad

I can't run 3600 Gear 1 somehow, I did read somewhere the memory controller on my cpu is no champ. So I run it at 3466

What kind of cpu do you run? 
That latency does seem high but is still lower than what I had. So you get the idea of why I was talking about my latency being high. I was upwards of 65 to 70ns. 
 
I would be running that ram at full 3600mhz with my cpu and board I believe which would increase the memory read and also reduce the latency. Plus your north bridge clock is like half mine. 
I presume since I have a newer cpu and different board? 

[leclod]

12 minutes ago, TalonRahl said:

What kind of cpu do you run?

It's written, Intel 12400

No idea about why the North Bridge Clock

Edited February 14 by leclod

[TalonRahl]

53 minutes ago, leclod said:

It's written, Intel 12400

No idea about why the North Bridge Clock

Ahh I see I see. 
Well that would be a huge difference in gains as well. 
Like your ram with that cpu is like premium fuel in a 1999 honda. 
Where as a 13700k would be more like a supra. The gains would show themselves more. 
We will see. I bought the ram. I will come back and post my stock xmp latency with it. Though I may adjust the tRFC and tREFI since its b die.

@RONOTHAN## What do you think of this samsung b die ram? And what did you say the tRFC could be for samsung bdie? 
I would imagine this would also be CR1 though I wont know until I try it. Its probably dual rank rather than single. And I see conflicting reports on how that affects performance. My understanding is single rank is like going around the track once where as dual rank is going around the track twice. From simple terms like that it makes single rank seem better and yet people say dual rank is better in some forums? 

 

 

If it runs stock as good as its looking I may just try and OC the tRFC and tREFI because the secondary and tertiary timings are like completely out of my knowledge base until I learn more. Though I would want to figure out more info on how to do that too. 

Do you think 3800mhz with same timings is achievable with staying at only the xmp 1.45v? 

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820374093 3600mhz 14 15 15 35 certified b die. 

[leclod]

1 hour ago, TalonRahl said:

Well that would be a huge difference in gains as well. 
Like your ram with that cpu is like premium fuel in a 1999 honda.

Ok, maybe but why ?

I've got p-cores close in performance to yours. At least I think I do.

In my mind there's not much difference between a 12400 and 13700K until one can use/needs more than 6 p-cores. Which in my case doesn't happen 95% of the time.

The Ram is dual rank. I've got tRFC at 630 and tREFI at 25000

 

 

 

Edited February 14 by leclod