Comprehensive Memory Overclocking Guide

[MageTank]

12 hours ago, Digitrax said:

From you article: "I'll be frank here. I have no idea what SG or DG means..."

"SG" is "Same Group" (meaning the same memory page)

"DG" is "Different Group" (the same way that "DR" is "Different Rank" and "DD" is "Different DIMM")

 

Thoroughly enjoyed your guide, though.

Digitrax, thanks for the clarification on that. I'll update the guide to reflect that info, and credit you for your contribution. This certainly helps understand what those timings are, and when they are applicable depending on memory configuration.

[Anomander]

Hi all, I'm pretty new to sub-timings (primary ones are a lot easier to understand). Can someone have look at what I've done and let me know if something look very off / incorrect. I mostly tested which values will work:

 

Also any chance someone can tell me what values to fill for the RTL and IOL timings? I wasn't able to even understand how these work in the bios since the values that I can insert are not what I see in the Timing Configurator. Thank you in advance!

[MageTank]

15 hours ago, Anomander said:

Hi all, I'm pretty new to sub-timings (primary ones are a lot easier to understand). Can someone have look at what I've done and let me know if something look very off / incorrect. I mostly tested which values will work:

 

 

Also any chance someone can tell me what values to fill for the RTL and IOL timings? I wasn't able to even understand how these work in the bios since the values that I can insert are not what I see in the Timing Configurator. Thank you in advance!

Your timings are actually looking very good. tFAW at 16 is a little concerning with tRRD_S at 6. If you can lower tRRD_S to 4, then a tFAW of 16 is fine. Otherwise, you may need to raise tFAW to 24 to prevent FAW from closing too soon. 

 

I would set your tWTR_L and tWTR_S timings to auto, then slowly lower tWRRD_SG and tWRRD_DG. Lowering these two timings will automatically lower tWTR_L and tWTR_S for you. Do keep in mind that your tWRRD_DG is already looking pretty tight, so I can't imagine you'll be able to get it any lower. 

 

Lastly, to adjust RTL/IO-L values, use the IO-L offsets. Change both IO-L offset channels to 22, then 23, then 24, etc until you reach instability. Increasing the IO-L offsets will automatically lower the RTL/IO-L values. Once you are finished with the IO-L offset, use the RTL Init formula value (listed in the description of the timing on the right) to prevent your RTL values from training beyond the formulas specification then manually dial the trained timings into each RTL channel.

 

All things considered, very impressive work with your RAM. 4000 C17 is very respectable.

[Anomander]

Thank you very much for the guidance! I did some more changes, but for some reason I can't find a value for the RTL channels to set. Can you please have another look and let me know what value I should try with? RTL value currently is 73 or so I think - 25 + 4 + (2 * 17) * 10 = 73

Should they be between 51 and 65 (so between D0 and D1)? I tried setting RTL Init Value to 80 (just to test) and both RLT Channel values to 80 as well, but the PC didn't boot up. Also same with 72 and 73.

 

[MageTank]

21 hours ago, Anomander said:

Thank you very much for the guidance! I did some more changes, but for some reason I can't find a value for the RTL channels to set. Can you please have another look and let me know what value I should try with? RTL value currently is 73 or so I think - 25 + 4 + (2 * 17) * 10 = 73

Should they be between 51 and 65 (so between D0 and D1)? I tried setting RTL Init Value to 80 (just to test) and both RLT Channel values to 80 as well, but the PC didn't boot up. Also same with 72 and 73.

 

 

 

I believe you should be using the D1 values, so RTL init would be either 71 or 72. Since you already tried 72 and it failed, give 71 a try. You can also try manually setting your RTL/IO-L values then input the RTL Init to prevent it from training outside of those manual values.

 

Here are my current values:

Note: I am using all 4 DIMM slots, so all of my RTL/IO-L values must be trained within 2T of each other.

 

You can also see that my secondary timings are looser than yours, but I've made up for it with tighter primary and tertiary timings. My configuration is designed to achieve the lowest possible latency that I can, which results in this kind of performance:

 

 

You'll have to feel it out to find that balance, but so far you are definitely on the right track. 

[Anomander]

Thanks for the tips! But, your clocks are crazy (including CPU). I believe my PC won't boot up even with all secondary/tertiary timings on auto and 16-17-17-39 (not to mention something like 15 and wtf RAS 30?!?) so I went with 17-17-17-39 and started adjusting secondary timings.

Can you tell me what voltages are you running and up to what point you'd consider safe for both CPU and RAM?

[MageTank]

On 1/11/2020 at 2:41 AM, Anomander said:

Thanks for the tips! But, your clocks are crazy (including CPU). I believe my PC won't boot up even with all secondary/tertiary timings on auto and 16-17-17-39 (not to mention something like 15 and wtf RAS 30?!?) so I went with 17-17-17-39 and started adjusting secondary timings.

Can you tell me what voltages are you running and up to what point you'd consider safe for both CPU and RAM?

I run my VDIMM at 1.5v however I have a fan pointing directly at my RAM. I run VCCIO and VCCSA at 1.15v each. CPU vCore is at 1.39v with LLC set at level 2. 

 

Desk is a little messy, but here is a picture of my current setup:

 

CPU is on a custom 1260mm loop with 18 140mm Corsair ML140's in push-pull. My CPU stays pretty cool even under AVX2 load, it's delidded and using a custom IHS from Rockitcool that I lapped even further to be completely flat. 

 

That pink fan is just a generic desk fan that I bought from some random dollar store, and was the only way I could get a fan mounted to my system that made any sense. Was considering zip ties with a 200mm fan but this method is easier to remove in a hurry and it runs pretty quiet on the low setting.

 

As for safe voltages, it really depends on your cooling solution and overall current. Voltage in and of itself is harmless, but current can be brutal. For RAM, it doesn't really matter from a longevity standpoint in regards to VDIMM, but keep in mind that RAM only scales so far with voltage before cooling becomes an issue. 1.5v is the highest I can go on air, and I don't recommend going beyond that unless you have more exotic cooling solutions. VCCIO and VCCSA can be pretty taxing on the memory controller. I personally stop at 1.15v because it's all I need for stability. I do not recommend exceeding 1.2v, most people will push up to 1.3 or 1.35v but that's just crazy in my opinion. Properly tweaked tertiary timings allow you to get away with much lower IMC voltages. As for vCore, it depends. Higher volts and low current is very safe. If you keep your current capped, volts can be relatively harmless, but understand that a capped current might limit your performance or straight up crash your overclock. It's about finding that balance and making sure your thermals are not crazy. I wouldn't push more than 1.37v on air, and would not push more than 1.44v on a custom loop like mine.

[Guest]

!!!!This thread should be stickied!!!! 

 

@MageTank 

Here's a quicky setup. Nothing special. No sub timings changes. Even Trfc is stock auto.

Manually set 16-16-38, adjusted ram voltage to 1.610v and off we go. I tweak for benchmarking only, so no I don't test for stability as a daily driver, however..... it's pretty close even just like this. 

Vccio gets pumped mad voltage on auto with this board. I generally don't seem to need more than 1.35v for anything, doesn't change anything seemingly.

Cpu is set 50x v-core auto. Cache at 4800mhz. Memory 4300mhz. 

 

This is an excellent guide. Thanks for the work in helping peeps get performance!! 

 

 

 

 

[Nikos4Life]

Hi all,
 

this is my first post here even I have been watching LTT and reading this thread for a long time. 

I will explain my issue: 

I am trying to squeeze a little bit more of my DDR4 kits paired with an ASUS MAXIMUS XI (z390 + 9900K). 

Corsair Domintator Torque 3200CL14 

I managed to achieve 4000CL17 24/7 stable but I would like to play with sub-timmings a little bit. 

My issue comes everytime I turn on Maximus tweak mode 2. 
Automatically cuts mem write speeds by ywo (AIDA 64).  

 

Does anyone know why this happens?


RAM VOLTAGE: 1.55 

VCCSA: 1.28

VCCIO: 1.29

Please if anyone needs more information I will write it asap.

Best,
Nikos

24/7 config stable


NEW CONFIG (tweak mode 2 - tFAW 16)

[MageTank]

2 hours ago, Nikos4Life said:

Hi all,
 

this is my first post here even I have been watching LTT and reading this thread for a long time. 

I will explain my issue: 

I am trying to squeeze a little bit more of my DDR4 kits paired with an ASUS MAXIMUS XI (z390 + 9900K). 

Corsair Domintator Torque 3200CL14 

I managed to achieve 4000CL17 24/7 stable but I would like to play with sub-timmings a little bit. 

My issue comes everytime I turn on Maximus tweak mode 2. 
Automatically cuts mem write speeds by ywo (AIDA 64).  

 

Does anyone know why this happens?


RAM VOLTAGE: 1.55 

VCCSA: 1.28

VCCIO: 1.29

Please if anyone needs more information I will write it asap.

Best,
Nikos

24/7 config stable


NEW CONFIG (tweak mode 2 - tFAW 16)

 

Greetings Nikos, 

 

Looking at the timings in the pictures you provided, I can see a couple areas that would negatively impact write bandwidth. Pay close attention to tWRRD_SG, tWRRD_DG, tWRWR_SG and tWRWR_DG. These will have the largest impact on your write bandwidth, and it would appear that ASUS's Mode 2 setting is loosening these timings by a fair bit. The biggest concern would be wirth tWRWR_DG as it's going from 4 to 8, which is a pretty significant change in bandwidth. I would focus on getting that one lowered first, then work on lowering the rest. As always, prioritize stability and lower latency over bandwidth, and make sure this does not negatively impact either of the two. So far, your 24/7 configuration looks fantastic. Keep up the great work.

[Nikos4Life]

On 1/21/2020 at 10:48 PM, MageTank said:

Greetings Nikos, 

 

Looking at the timings in the pictures you provided, I can see a couple areas that would negatively impact write bandwidth. Pay close attention to tWRRD_SG, tWRRD_DG, tWRWR_SG and tWRWR_DG. These will have the largest impact on your write bandwidth, and it would appear that ASUS's Mode 2 setting is loosening these timings by a fair bit. The biggest concern would be wirth tWRWR_DG as it's going from 4 to 8, which is a pretty significant change in bandwidth. I would focus on getting that one lowered first, then work on lowering the rest. As always, prioritize stability and lower latency over bandwidth, and make sure this does not negatively impact either of the two. So far, your 24/7 configuration looks fantastic. Keep up the great work.

Hi @MageTank thanks for the welcome message,

 

Thanks for your advice I was able to correct the write speed. 

I have been reading this thread as well as other sites to get more information but I am not quite sure which way to go to improve performance. 

This is the new pofile I am playing with (seems quite stable so far, I know it is not 1000% tested but to keep working seems ok):

 

And here you have the voltages:

 

 

Which timmings should I try to improve? I think I can lower tRFC maybe a little bit more. 

Also I have a problem tightening RTL & IOL values, may you help me with those? 
Not quite sure how I am supposed to adjust or which one is the one I need to change since there are a lot of values about the same RTL. 
It is like CHA DIMM0 Rank0, CHA DIMM0 Rank1... and I am not quite suer which one is the right one to make it work.

Also, I would love to pursuit the 1T CR but again not quite sure if I need to set up the timmings perfeclty before attemping to go for it. 
I have read about Trace Centering helping with it but not sure if it is voltage related or what...

 

Thanks in advance.

Best,
Nikos

[MageTank]

19 hours ago, Nikos4Life said:

Hi @MageTank thanks for the welcome message,

 

Thanks for your advice I was able to correct the write speed. 

I have been reading this thread as well as other sites to get more information but I am not quite sure which way to go to improve performance. 

This is the new pofile I am playing with (seems quite stable so far, I know it is not 1000% tested but to keep working seems ok):

 

And here you have the voltages:

 

 

 

Which timmings should I try to improve? I think I can lower tRFC maybe a little bit more. 

Also I have a problem tightening RTL & IOL values, may you help me with those? 
Not quite sure how I am supposed to adjust or which one is the one I need to change since there are a lot of values about the same RTL. 
It is like CHA DIMM0 Rank0, CHA DIMM0 Rank1... and I am not quite suer which one is the right one to make it work.

Also, I would love to pursuit the 1T CR but again not quite sure if I need to set up the timmings perfeclty before attemping to go for it. 
I have read about Trace Centering helping with it but not sure if it is voltage related or what...

 

Thanks in advance.

Best,
Nikos

Greetings Nikos,

 

If you're going to be adjusting CR, it's best to dial that in before touching the RTL/IO-L values as command rate directly impacts those values. As for the RTL/IO-L values themselves, you need to only pay attention to the D1 values as you are using Channel A DIMM 2 and Channel B DIMM 2. D0 = DIMM 1 in each channel, D1 = DIMM 2 in each channel. As for how to adjust them, start by manually increasing the IO-L offset from 21 to 22. This will automatically lower your IO-L value. You will eventually top out around 24 or 25, which is normal. Afterwards, use the RTL Init formula (IO-L + IO-L Offset + CL (x2) + 10) and use that value to prevent your RTL's from training outside of that specific range. From there, you can manually lower them, but be mindful that their values cannot exceed more than 2T of each other or you will run into performance/stability issues. For example: Your current RTL values are CHA D1: 69 and CHB D1:71. This is exactly 2T apart, and is fine. If it were 69 and 72, or 68 and 71, this would be a problem. As long as you keep that in mind when adjusting the RTL values, you will be fine.

 

As for what you should be adjusting next, your RAM looks quite tight as it is. Tightening TRFC might help a little, but be mindful once you hit the 270-280 mark. Things can get unstable quickly with your memory configuration. Honestly, you should be very proud of the results. It's one of the most solid manual tuning that I have seen in a very long time.

[TrantaLocked]

It would be cool to get some advice on my situation.

 

I had an OC on my CPU and memory that initially tested fine and was running well for over a year. But recently, after I encountered a Chrome bug that made me suspicious, I wanted to double check, and after some trial and error with memtes86 v8.2 I found that my tREFI value was the culprit for new memory errors that show up within the first few minutes of the hammer test (#13). I'm already so close to stable that I only get errors in the memtest86 hammer test (#13), but not the others. I'm even able to pass older versions of memtest86 that don't have that specific test, so I'm thankful I was able to figure that out a while ago.

 

I used to have tREFI at max (65,535) for over a year but now suddenly I can only pass a clean hammer test if I set to 15,000 or lower. That seems like a giant leap, no? I suddenly have to power refresh four times as often for the same level of stability as before?

 

The one culprit I am currently thinking is that it could be the humidity in my room when I hang up clothes that didn't dry completely in the dryer. My dryer is really old and I'd rather not waste power re-running cycles for hours, so I have to hang some of the heavier pieces in my room.

 

I've read that higher levels of ambient temperature may require lower tREFI for stability. I also consider the possibility that every time it's too hot or humid in my room I'm unknowingly running an unstable system but because it's a rare corruption, I don't blue screen but might have occasional file corruption or other bugs in Windows, like the one I just posted a thread about with Chrome causing what seemed to be a major CPU hang. Obviously I don't test memtes86 every week so I wouldn't know this is going on. 

 

 

[MageTank]

3 hours ago, TrantaLocked said:

It would be cool to get some advice on my situation.

 

I had an OC on my CPU and memory that initially tested fine and was running well for over a year. But recently, after I encountered a Chrome bug that made me suspicious, I wanted to double check, and after some trial and error with memtes86 v8.2 I found that my tREFI value was the culprit for new memory errors that show up within the first few minutes of the hammer test (#13). I'm already so close to stable that I only get errors in the memtest86 hammer test (#13), but not the others. I'm even able to pass older versions of memtest86 that don't have that specific test, so I'm thankful I was able to figure that out a while ago.

 

I used to have tREFI at max (65,535) for over a year but now suddenly I can only pass a clean hammer test if I set to 15,000 or lower. That seems like a giant leap, no? I suddenly have to power refresh four times as often for the same level of stability as before?

 

The one culprit I am currently thinking is that it could be the humidity in my room when I hang up clothes that didn't dry completely in the dryer. My dryer is really old and I'd rather not waste power re-running cycles for hours, so I have to hang some of the heavier pieces in my room.

 

I've read that higher levels of ambient temperature may require lower tREFI for stability. I also consider the possibility that every time it's too hot or humid in my room I'm unknowingly running an unstable system but because it's a rare corruption, I don't blue screen but might have occasional file corruption or other bugs in Windows, like the one I just posted a thread about with Chrome causing what seemed to be a major CPU hang. Obviously I don't test memtes86 every week so I wouldn't know this is going on. 

 

 

I think you hit the nail on the head with the ambient temperatures. TREFI is extremely sensitive to change in temperatures, so much so that I keep a fan mounted directly above my RAM when running 65535 24/7. 

 

Another culprit to look at is tRFC. Both work hand in hand with one another and it's possible that a low tRFC value is providing too small of a window to refresh properly. Raising it slightly might resolve the issue. I would start by addressing the potential cooling issue, then move on to tRFC if the errors persist.

[TrantaLocked]

I could maybe turn the main case fan around so it blows inward. What do you think about that? I'll also try out the tRFC values, but I'd need to wait for it to be humid enough again because I've already aired out my room enough that it probably won't give me worse case scenario tests at the moment. 

[MageTank]

2 hours ago, TrantaLocked said:

I could maybe turn the main case fan around so it blows inward. What do you think about that? I'll also try out the tRFC values, but I'd need to wait for it to be humid enough again because I've already aired out my room enough that it probably won't give me worse case scenario tests at the moment. 

It won't take much airflow, just enough to make sure there are no static hotspots in between the DIMMs. Normally a top mounted fan exhausting air out of the chassis is enough to pull cooler air through it. If you are only running 2 DIMM's, this shouldn't be a big issue at all. Running 4 DIMM's with thicker heatspreaders can be a bit of a nuisance when pushing high  tREFI values.

[LukaH]

I've been lurking in this thread for a while now but haven't gotten around to reading the guide yet or trying tuning the memory on my own (lack of time). But I've recently bought a Ryzen 3600 and a Strix X570-F and decided to try out the Ryzen DRAM calculator which I've had no luck with on my old Crosshair 6 Hero with a 1800X with the same G.Skill 3200MHz CL14 kit 16gb (2x8). So I let it calculate the fast settings and it stuck like a charm, I was even able to OC the ram to 3400MHz. So what I want to know is how good are the timings it calculated (posted in the screenshot below). Also what benchmark should i run when I get around to fiddling around some more?

 

Thanks for the help. Also your system looks crazy AF

 

[MageTank]

2 hours ago, LukaH said:

I've been lurking in this thread for a while now but haven't gotten around to reading the guide yet or trying tuning the memory on my own (lack of time). But I've recently bought a Ryzen 3600 and a Strix X570-F and decided to try out the Ryzen DRAM calculator which I've had no luck with on my old Crosshair 6 Hero with a 1800X with the same G.Skill 3200MHz CL14 kit 16gb (2x8). So I let it calculate the fast settings and it stuck like a charm, I was even able to OC the ram to 3400MHz. So what I want to know is how good are the timings it calculated (posted in the screenshot below). Also what benchmark should i run when I get around to fiddling around some more?

 

Thanks for the help. Also your system looks crazy AF

 

 

Those timings look good to me, very tight. I am curious about the tRFC values. Are they really that low at 3400mhz? On Intel, going below 270 can be extremely difficult without perfect VCCIO/VCCSA, but I imagine Ryzen handles this differently. My experience with Ryzen is still quite limited, despite having a Ryzen platform for over a year (and more recently the Threadripper 3990X to overclock). You can probably push a little more voltage (1.42-1.44 VDIMM) and get 3600mhz working without having to make any drastic changes to your timings. As long as you keep your SOC voltage in check, 1.45v VDIMM for 24/7 use is harmless with decent airflow.

 

Also, thanks. My system right now is designed for benching and easy swapping of components. Once I settle down with a platform I like, I'll go a hardline custom loop with some copper pipes and dual D5 pumps. Just need to find a platform I'll be happy with for years to come.

[LukaH]

15 hours ago, MageTank said:

Those timings look good to me, very tight. I am curious about the tRFC values. Are they really that low at 3400mhz? On Intel, going below 270 can be extremely difficult without perfect VCCIO/VCCSA, but I imagine Ryzen handles this differently. My experience with Ryzen is still quite limited, despite having a Ryzen platform for over a year (and more recently the Threadripper 3990X to overclock). You can probably push a little more voltage (1.42-1.44 VDIMM) and get 3600mhz working without having to make any drastic changes to your timings. As long as you keep your SOC voltage in check, 1.45v VDIMM for 24/7 use is harmless with decent airflow.

 

Also, thanks. My system right now is designed for benching and easy swapping of components. Once I settle down with a platform I like, I'll go a hardline custom loop with some copper pipes and dual D5 pumps. Just need to find a platform I'll be happy with for years to come.

Yeh that's what i put in for the tRFC and had zero problems so far. The values in blue is what is reported by the BIOS. I'll try and push for 3600MHz and see what happens, I'm guessing that two intake fans (noctuas) in the top is decent enough to push for 1.45V?

 

I wasn't very impressed with 1st gen ryzen and x370 tweaking wise but x570 and zen 2 is like night and day for me.

[jvmunhoz]

Hi @MageTank, first of all I'd like to congratulate you on a great guide. Really great depth for such a complicated matter, especially for a beginner like me in overclocking. Since I'm a beginner, I'd like to ask a few things before tinkering with my configs here.

 

I have a MSI MPG Z390 GAMING PRO CARBON mobo, an i7-9700k and a Patriot Viper Steel Series DDR4 16GB 4400MHz (PVS416G440C9K). You can check my full setup here. I'd say it's a good mobo and should be able to give me some maneuverability with some settings (such as tREFI). Without the overclock the RAM is just really meh, so that's why I'm here.

 

I originally intended to use the XMP profiles that came with it, but I had no such luck. XMP profile 1 wouldn't even POST at 4400MHz 19-19-19-38 2T, with 1.45V of VDIMM and

VCCIO and VCCSA both jumping to 1.40V. I couldn't complain, since my motherboard specifications only gave support for XMP profile 2 which gave me 4266MHz 19-19-19-39 2T with 1.35V of VDIMM and VCCIO and VCCSA again at 1.40V. But to my dismay, although the system would post, it was as unstable as it gets. It crashed on the first minute of the first benchmark. So here I am having to make overclock it myself . The timings of the XMP Profile 2 are as follows:

 

 

I should probably begin by trying to raise the current VDIMM of 1.35V. At most 1.45V? (I've seen lots of controversies as to what's the max safe limits). Any other values that may seem a bit jarring and could cause the instability?

 

I intend to make a decent overclock, focusing on frequency, primary timings and tRFC and tREFI. I may fiddle with the other values, but won't tighten them that much since it's my first time doing it. Any other value I should focus on? Any other tips?

 

Thanks in advance!

 

[MageTank]

28 minutes ago, jvmunhoz said:

Hi @MageTank, first of all I'd like to congratulate you on a great guide. Really great depth for such a complicated matter, especially for a beginner like me in overclocking. Since I'm a beginner, I'd like to ask a few things before tinkering with my configs here.

 

I have a MSI MPG Z390 GAMING PRO CARBON mobo, an i7-9700k and a Patriot Viper Steel Series DDR4 16GB 4400MHz (PVS416G440C9K). You can check my full setup here. I'd say it's a good mobo and should be able to give me some maneuverability with some settings (such as tREFI). Without the overclock the RAM is just really meh, so that's why I'm here.

 

I originally intended to use the XMP profiles that came with it, but I had no such luck. XMP profile 1 wouldn't even POST at 4400MHz 19-19-19-38 2T, with 1.45V of VDIMM and

VCCIO and VCCSA both jumping to 1.40V. I couldn't complain, since my motherboard specifications only gave support for XMP profile 2 which gave me 4266MHz 19-19-19-39 2T with 1.35V of VDIMM and VCCIO and VCCSA again at 1.40V. But to my dismay, although the system would post, it was as unstable as it gets. It crashed on the first minute of the first benchmark. So here I am having to make overclock it myself . The timings of the XMP Profile 2 are as follows:

 

 

I should probably begin by trying to raise the current VDIMM of 1.35V. At most 1.45V? (I've seen lots of controversies as to what's the max safe limits). Any other values that may seem a bit jarring and could cause the instability?

 

I intend to make a decent overclock, focusing on frequency, primary timings and tRFC and tREFI. I may fiddle with the other values, but won't tighten them that much since it's my first time doing it. Any other value I should focus on? Any other tips?

 

Thanks in advance!

 

Thank you for the kind words. Luckily I have the exact same kit of memory you are using and I am quite familiar with it. If you look at my post below, you can see what I am currently running for my clock speeds and timings using this kit.

On 1/10/2020 at 11:50 AM, MageTank said:

I believe you should be using the D1 values, so RTL init would be either 71 or 72. Since you already tried 72 and it failed, give 71 a try. You can also try manually setting your RTL/IO-L values then input the RTL Init to prevent it from training outside of those manual values.

 

Here are my current values:

Note: I am using all 4 DIMM slots, so all of my RTL/IO-L values must be trained within 2T of each other.

 

You can also see that my secondary timings are looser than yours, but I've made up for it with tighter primary and tertiary timings. My configuration is designed to achieve the lowest possible latency that I can, which results in this kind of performance:

 

 

You'll have to feel it out to find that balance, but so far you are definitely on the right track. 

NOTE: I am using 4 DIMM's which activate 2DPC timings, so it won't be as simple as copy/pasting my values, but it should give you an idea on what things look like when tweaked as tight as they can possibly go. Also bare in mind that my voltages are different, as is the quality of my memory controller and the trace topology of my motherboard. I am running 1.5v VDIMM with active cooling on my DIMM's because I run tREFI at 65535. My VCCIO and VCCSA volts are 1.14v and 1.15v respectively. 

 

Much like you, I couldn't get the XMP to work at all for 4400, and 4266 simply isn't stable. I can stabilize it with some loose tertiary timings, but it performed far worse than simply running a very tight 4000 C15 configuration. That would honestly be my advice. Dial it down to 4000 C15/C16 or 4200 C16/C17 then tighten it up as much as physically possible.

 

As for your specific concerns with your timings and what may be causing instability, I would definitely tweak TFAW. Your tFAW value should be 4x tRRD_S. Since your tRRD_S has a value of 9, TFAW should be a minimum of 36, not 26. RTL/IO-L offsets look fine, still within 1T of each other. tRDWR timings might be a little tight compared to your tCAS, so loosening those up to 18 might help solve some of your stability issues. Still, I would work on dropping clock speeds down and tightening primary timings anyways. 

 

You have a solid board, and you have some of the best single rank B-Die kits around, so it's only a matter of finding out what that memory controller prefers. Hopefully my settings can take out some of the guesswork. If not, I can always try to help after each training session to see what looks off.

[jvmunhoz]

57 minutes ago, MageTank said:

Thank you for the kind words. Luckily I have the exact same kit of memory you are using and I am quite familiar with it. If you look at my post below, you can see what I am currently running for my clock speeds and timings using this kit.

NOTE: I am using 4 DIMM's which activate 2DPC timings, so it won't be as simple as copy/pasting my values, but it should give you an idea on what things look like when tweaked as tight as they can possibly go. Also bare in mind that my voltages are different, as is the quality of my memory controller and the trace topology of my motherboard. I am running 1.5v VDIMM with active cooling on my DIMM's because I run tREFI at 65535. My VCCIO and VCCSA volts are 1.14v and 1.15v respectively. 

 

Much like you, I couldn't get the XMP to work at all for 4400, and 4266 simply isn't stable. I can stabilize it with some loose tertiary timings, but it performed far worse than simply running a very tight 4000 C15 configuration. That would honestly be my advice. Dial it down to 4000 C15/C16 or 4200 C16/C17 then tighten it up as much as physically possible.

 

As for your specific concerns with your timings and what may be causing instability, I would definitely tweak TFAW. Your tFAW value should be 4x tRRD_S. Since your tRRD_S has a value of 9, TFAW should be a minimum of 36, not 26. RTL/IO-L offsets look fine, still within 1T of each other. tRDWR timings might be a little tight compared to your tCAS, so loosening those up to 18 might help solve some of your stability issues. Still, I would work on dropping clock speeds down and tightening primary timings anyways. 

 

You have a solid board, and you have some of the best single rank B-Die kits around, so it's only a matter of finding out what that memory controller prefers. Hopefully my settings can take out some of the guesswork. If not, I can always try to help after each training session to see what looks off.

Thanks for the fast response!

 

And what were the odds we had the same memory kit? Hahaha. It certainly helps a lot with some of the guesswork. I'll try dialing it down to 4000/C15 to see how it responds.

 

But since you are using 4 DIMM's, what are the values that I must tweak and what should I do with them since I just use 2 DIMM's? I'll take your config as a base, so I'd like to know what to look for.

 

Really appreciate the help!

[MageTank]

56 minutes ago, jvmunhoz said:

Thanks for the fast response!

 

And what were the odds we had the same memory kit? Hahaha. It certainly helps a lot with some of the guesswork. I'll try dialing it down to 4000/C15 to see how it responds.

 

But since you are using 4 DIMM's, what are the values that I must tweak and what should I do with them since I just use 2 DIMM's? I'll take your config as a base, so I'd like to know what to look for.

 

Really appreciate the help!

Since you are using single rank DIMM's in 1 DIMM Per Channel, you can ignore any timing suffix labeled "_DR" and "_DD". These correspond to Different DIMM (2DPC) & Different Rank respectively. You can set their values as low as 1, they will have zero impact on performance or stability. Focus entirely on the SG (Same Group) & DG (Different Group) timings. These are what will determine the most performance for your configuration. Start with one group at a time. For example: Tweak RDRD as a group, RDWR as a group, and WRWR as a group. Stress test in between each grouping to make sure your changes are valid and are not compromising the stability of your system. Also perform some quick performance benchmarks in between your tweaking to make sure you are not losing performance, or that you are not increasing the stress on your memory controller for no relative performance benefits. When it comes to timings, sometimes tighter isn't always better.

 

You will absolutely want to tinker with VCCIO/VCCSA voltages. The quality of your memory controller will ultimately dictate what is needed here, and it's difficult to really give you an answer as to exactly how much you will need. We can only really guess, and provide guidelines such as "Don't exceed 1.3v for each value". Honestly when you find the perfect timings, you'll know because you'll be able to reduce these two voltages quite a bit while remaining stable. It reaches the point where it just clicks.

 

Also keep in mind that my configuration is as tight as I can physically get it. I am at the limits of either my board, DIMM's or IMC and cannot go any further in any direction (timings, clock speeds or voltages). It may not be as simple as dialing in my settings, though I do hope it is that easy. Good luck!

[jvmunhoz]

Hey, @MageTank! It's been about a week I've been (slowly) tinkering with the RAM, and it's been a complete disaster lol.

 

My mobo doesn't like frequencies above 4200 C19, so I've taken 4200 as my limit. Some configurations between 4100Mhz and 4200Mhz (my IMC outright hates 4000Mhz, MemTest86 passed 4100 C19 but rejected 4000 C19) are totally passable at MemTest86, being 4133 C17 the best one performance-wise      . . All voltages are basically maxed for now, with VCCIO/VCCSA both at 1.25v and vDIMM at 1,45v. All other timings besides the primary ones are on Auto for now. But I can't, for the life of me, make any of those configs run Prime95 longer than 5 minutes.

 

So I gather I should now try to loosen some secondary and tertiary timings? (Anything seems off with any of those timings?) Or should I try to lower the frequency even more? (for a mobo that should theoretically be able to handle the XMP, that would be a major disappointment). Anything I'm doing wrong?

 

Thanks for the patience in helping me.

[MageTank]

16 hours ago, jvmunhoz said:

Hey, @MageTank! It's been about a week I've been (slowly) tinkering with the RAM, and it's been a complete disaster lol.

 

My mobo doesn't like frequencies above 4200 C19, so I've taken 4200 as my limit. Some configurations between 4100Mhz and 4200Mhz (my IMC outright hates 4000Mhz, MemTest86 passed 4100 C19 but rejected 4000 C19) are totally passable at MemTest86, being 4133 C17 the best one performance-wise      . . All voltages are basically maxed for now, with VCCIO/VCCSA both at 1.25v and vDIMM at 1,45v. All other timings besides the primary ones are on Auto for now. But I can't, for the life of me, make any of those configs run Prime95 longer than 5 minutes.

 

So I gather I should now try to loosen some secondary and tertiary timings? (Anything seems off with any of those timings?) Or should I try to lower the frequency even more? (for a mobo that should theoretically be able to handle the XMP, that would be a major disappointment). Anything I'm doing wrong?

 

Thanks for the patience in helping me.

Sounds like you found a "bad strap". Back with DDR3, I would fail ANY 2666mhz configuration, but could do 2933 with the exact same setup. With early DDR4 samples, I was able to pass 3200 and 3466, but fail 3333. Glad you had the foresight to try another strap instead of giving up at the first sign of failure, I may have forgotten to mention that in my original guide.

 

As for your timings, I agree that the secondary and tertiary timings need work. Your RTL/IO-L offsets are definitely out of whack a little and need to be straightened out. On your 4100 C19 profile, there is a 3T gap between each RTL channel. This needs to be within 0-1T. The same goes for your IO values, which are 2T apart. They need to be 0-1T. Use the RTL Init formula to help narrow the window and tighten those values, or simply loosen the tRDWR timings a little and see how they train afterwards.

 

The same goes for the 4000 C19 profile, but I would work on the 4100 first. In both, you need to adjust tFAW to be 4x tRRD_S. In your 4100 profile, tFAW's value would be 36 (assuming you did not adjust tRRD_L at all) and in the 4000 profile it would be 28.

 

I also didn't mention in this guide, but you shouldn't adjust tWTR_L or tWTR_S manually. Adjusting the tWRRD timings will automatically change those, so leave those two secondary timings on auto and influence them using the tWRRD tertiary timings. 

 

I really do need to update this guide, there is a lot that I've learned over the years that I have not included that would probably help people understand this a bit better. I hope that helps in your tweaking, if you have any questions, let me know.