Glitchy BSOD, Game crashes, System restarts after getting new RAM and testing.

[Rrraaayyy]

Hya!
Last year I upgraded my setup, by building a new pc with parts, and since then I had all sorts of non-stop problems (system freezes, total system crash, bsods, application crashes) with this system.
After I've tried numerous fixes and changes, gpu driver versions, windows installations, cmd commands, stress tests for cpu gpu or ram, bios settings, even a different Motherboard, i've concluded that my RX 6700 (non-XT) that I had at the time was the culprit (I have loads of details about this "saga" as well), thus I've RMA'd it to the shop I bought it from and got another GPU, a RX 6750 XT, and after that the system was much more stable and the problems seemed to have gone away.
Until now.
Last week I found the same RAM I had, at a good price, so I went ahead and bought another kit, since I wanted to get more RAM down the line anyway (I only had one 2x8GB kit in my PC until this point).

RAM gets to me, I install it next to my current sticks, and theeen... disasters.

On the first day I've installed the RAM I kept getting Windows Crashes and System freezes on a 2 hour cooldown.

"Well shit, i guess i have bad RAM, time to RMA".

 

So I started testing the RAM with memtest86 (since I already had the stick ready, from the "bad gpu" troubleshooting era before this), and sure enough, with all sticks installed, after letting it run for the whole night, I had the test FAIL with like 40 errors, what's curious though is that when looking at the adresses and GB sector of each error, I had errors on ALL slots, but mostly the last slot.

Good. Ok let me just remove the good ram, that I had before, and run the test again only with this "bad" ram.

For the purpose of easier clarity: Before this, the initial sticks were placed in slots 2 and 4, in Dual channel mode of course, and i've read it's better to have the last slot occupied to avoid some sort of "echo" effect or something related to the motherboards and the way they are designed.

We will call the above configuration of occupied slots - "default".

 

• Tested the good sticks in default configuration - Passed, 0 Errors - to provide a baseline.
• Tested the good sticks on the other slots - Passed, 0 Errors - to check faulty slots.
• Tested the "bad" sticks in default configuration - Passed, 0 Errors. - huh?
• Tested the "bad" sticks on the other slots - Passed, 0 Errors - nah this isn't real
• And finally, I put all the sticks in the same configuration as I had them on the first day, when I got the errors and the system crashes, and let it run over night. - 10  Hours, 7 Passes, 0 Errors.

Well, shit.

I was so much in disbelief that I thought maybe the XMP profile somehow got disabled, so since it's not on OC speeds, the RAM is much more stable => no errors, but nope, XMP profile was enabled, exactly as I had it before.

There's also something else, once when running memtest86, I had the PC just reboot itself without my input, interrupting the test completely...

 

After that day and all those tests, I couldn't draw any conclusion so I just kept using the system as-is, with all RAM sticks installed.

After 3 days of the system running just fine, with no problems, now they started to popup again:

• CS2 Crashes mid game, straight to desktop.
• Ready or Not crashes while in mission.
• Hell Let Loose crashes while playing a match.
• System freezes accompanied by forced reboot
• Windows blue screen.

The Blue screen i've been getting is not quite "normal", it's only on half the screen, the rest being black, and it's very glitchy, the same section is visible like 3-4 times, on the same screen.

 

As I'm writing this article, I kept trying to play Ready Or Not but It crashed 3 times in like 50 minutes, CS2 also crashed again today....so I capitulated.

Is the RAM good for my system/cpu/mobo? - From what I've seen there shouldn't be a problem.
Maybe it's a defective Mobo?

Are AMD systems this bad when it comes to stability? - I've heard they have somewhat of a bad reputation, before the purchase, I didn't think so at all, but after all these months of pain, I don't know what to believe anymore...

Is this maybe related to the "bad" GPU I've had? The symptoms are very very similar... maybe that poor RX 6700 had no fault...

Could I have gotten "better" parts? - Yes, sure, these are not the "best" parts or brands known for the best quality or "premium-ness" of their products, but I'm not a guy who needs to play the latest and greatest on max settings.

I'm a simple PC gamer, I don't mind doing optimizations getting my hands dirty and lowering settings.... Besides, it's more efficient to go for a mid-range build money-wise, the jump to high end is waay more expensive and you don't get much more "bang" for the buck. - So that's why I've chosen these parts/brands, and none of the chosen brands scream "problems" to me. They are not the best but also not the worst. Mid-range!

 

PC Specs:

• System age - Hardware: 10 months, 18 days
• System age - OS: 5 months (I went through multiple Windows reinstalls recently...)
•                 OS^*: Windows 10 22H2 x64 bit, Desktop, Retail
•                CPU:  AMD Ryzen 5 5600 3.5GHz
• CPU Cooler:  ID-Cooling SE-224-XT V2 Black
•               RAM:  Corsair Vengeance LPX Black 16GB DDR4 3600MHz CL18 Dual Channel Kit - CMK16GX4M2D3600C18 x2 (meaning 2 of these 2x8GB kits)
•               GPU:  Sapphire Radeon RX 6750 XT
•           MOBO:  ASRock B550 Phantom Gaming 4 ATX
•                PSU:  Seasonic B12 BC Series, 80 PLUS Bronze, 750W
•        Storage:  1x SSD WD Blue SN570 1TB(System) + 1x SSD Samsung PM981 1TB(Storage)
•      Monitors:  1x Asus PG279Q@144Hz + 1x LG 24GM79G@144Hz, both connected to GPU, thru new Displayport cables

 

I swear this is some Jayz2Cents troubleshooting video where he finds the weirdest most insane issues.... like that GPU Bios version mismatch.,,,

I've decided to finally seek help, because I am tired, I am fed up and I don't know what to do anymore. I have no idea what is wrong with my system, it could be connected to the stability issues i've had with the "bad" GPU before, maybe the GPU wasn't even bad and it's the same problem but with different symptoms... I just want it to be over

 

 

In Event Viewer on the fateful day when I first installed the new RAM, there were LOADS of DistributedCOM, Bits-Client, and also a few Kernel-Power errors:

 

 

 

 

 

 

 

[Rrraaayyy]

Below I've uploaded the output from SysnativeBSODCollectionApp and Perfmon Reports as instructed in the "BSOD Posting Instructions" thread.
I've also added some more screenshots, 2 full memory dumps, 2 mini dumps and Event Viewer Logs Exports.

When running Perfmon /report in cmd, the program would hang indefinetly at the "running for 60 seconds" part, but I found the files for each report inside "C:\PerfLogs\System\Diagnostics", and I just copied those from there, hope it's ok.
If you need anything else let me know.

https://drive.google.com/file/d/1EI-ti3pJekEALUw6CxDVe_hd-ewfPVCN/view?usp=sharing

 

I know this is a long post but I'm a sucker for details and I wouldn't have provided the extra context if I didn't think it would be useful... sorry in advance.

Thank you, everyone.

[DoctorNick]

56 minutes ago, Rrraaayyy said:

Below I've uploaded the output from SysnativeBSODCollectionApp and Perfmon Reports as instructed in the "BSOD Posting Instructions" thread.
I've also added some more screenshots, 2 full memory dumps, 2 mini dumps and Event Viewer Logs Exports.

When running Perfmon /report in cmd, the program would hang indefinetly at the "running for 60 seconds" part, but I found the files for each report inside "C:\PerfLogs\System\Diagnostics", and I just copied those from there, hope it's ok.
If you need anything else let me know.

https://drive.google.com/file/d/1EI-ti3pJekEALUw6CxDVe_hd-ewfPVCN/view?usp=sharing

 

I know this is a long post but I'm a sucker for details and I wouldn't have provided the extra context if I didn't think it would be useful... sorry in advance.

Thank you, everyone.

Ello,

Edit: Long story short, the kit is only made for running with two sticks, not 4. Therefore the timings burned into the DOCP profiles is made for 2 sticks not 4. Can you run the kit without DOCP(xmp) enabled? aka auto speeds. If yes. Then set frequency, dram voltage and primary timings manually instead of using DOCP. 

[Rrraaayyy]

1 hour ago, DoctorNick said:

Ello,

Edit: Long story short, the kit is only made for running with two sticks, not 4. Therefore the timings burned into the DOCP profiles is made for 2 sticks not 4. Can you run the kit without DOCP(xmp) enabled? aka auto speeds. If yes. Then set frequency, dram voltage and primary timings manually instead of using DOCP. 

Damn, really?
I saw a table with RAM configurations in the manual, but after I've googled a bit it seemed ok, didn't see why I couldn't run 4 sticks at 3600... like, it's not some ultra high frequency for some ultra high end rig... I've also been on sites like Corsair/Kingston, searched the motherboard model on their site, and they listed 4x Ram Kits no problem....

This is the table from the motherboad manual, but looking at it, seeing how it refferenced a frequency of only 3200 Mhz and only for Matisse and Renoir, which are older CPUs, mine being Vermeer, I just thought it was out of date and not very useful, because searching the web if the RAM in question is supported for my CPU/mobo, I've had positive answers...


 

[Rrraaayyy]

Wait, I think I've read that wrong, are we talking about the RAM KIT here? The kit is the problem?
How in the hell if the RAM is compatible with the CPU and MOBO, you can't add more if you have the slots for it? What sort of anti-consumer anti-upgrade mechanism is that, lol !

I will test the ram as you said, and return with more results, but seeing as the crashes are very random (I've played like 4 hours without any issue, but then in 50 mins it can crash like 3-4 times), I really don't know how to draw conclusions...
Maybe I will also run another memtest overnight, but seeing as I ran it with XMP enabled for 10 hours and still got a pass... idk what to say.

[Rrraaayyy]

Also, Timings voltage specs and speed, dual channel mode and compatibility is EXACTLY why i've purchased an IDENTICAL kit....
How can the Timings and XMP profiles not run with 4 sticks... if they are all the same for each stick???

[Bjoolz]

2 hours ago, Rrraaayyy said:

Also, Timings voltage specs and speed, dual channel mode and compatibility is EXACTLY why i've purchased an IDENTICAL kit....
How can the Timings and XMP profiles not run with 4 sticks... if they are all the same for each stick???

 

2 hours ago, Rrraaayyy said:

Wait, I think I've read that wrong, are we talking about the RAM KIT here? The kit is the problem?
How in the hell if the RAM is compatible with the CPU and MOBO, you can't add more if you have the slots for it? What sort of anti-consumer anti-upgrade mechanism is that, lol !

I will test the ram as you said, and return with more results, but seeing as the crashes are very random (I've played like 4 hours without any issue, but then in 50 mins it can crash like 3-4 times), I really don't know how to draw conclusions...
Maybe I will also run another memtest overnight, but seeing as I ran it with XMP enabled for 10 hours and still got a pass... idk what to say.

 

A few things here. Running four sticks, as shown in the manual with 4 sticks having a lower officially supported speed, can be one possible reason for the issue. Running four sticks is simply much harder on the memory controller (Part of the CPU).

 

I don't see your kit on the motherboard's memory QVL. I haven't really cared about the QVL since second gen Ryzen (First and second gen was a bit of nightmare with RAM), but with four sticks you really should still use the QVL if possible. The QVL will tell you if it passed the control with 2 or 4 sticks. They test all kits with 4 even if they are two stick kits or even single stick. With it not being on the QVL list, it's a dice roll.

 

Buying the same model number RAM as you already have doesn't mean that you get the same RAM. Before every batch the RAM vendor will get the cheapest parts they can get which will do the speeds and timings they need. On a later batch, the RAM could be completely different from the original batch. This makes the QVL even more difficult to follow because no one tells you which parts were used when you buy the RAM, you just have to check after you received it. The QVL will tell you, for most kits, which manufacturer the components were from with the tested kit.

 

With Corsair RAM they print which parts were used on the sticker of each stick. It's encoded because they don't really want people to know about this. If you look at the sticker on each kit, it will have a version number. This is not really the version, it's who the parts are from and which version chipset (What is usually called a die with RAM) was used with that stick. You can see in the screenshot below that all kits from SK Hynix start with a 5. If they don't have the same version number they are not any more similar than any other two random kits. 

 

Example image of the memory QVL for Corsair sticks on your board with your CPU.

 

You can see the component manufacturer in the Chip field (And they include the version number behind the model number) and how many sticks it was stable with in the "DIMM Socket Support" field.

[Rrraaayyy]

53 minutes ago, Bjoolz said:

 

 

A few things here. Running four sticks, as shown in the manual with 4 sticks having a lower officially supported speed, can be one possible reason for the issue. Running four sticks is simply much harder on the memory controller (Part of the CPU).

 

I don't see your kit on the motherboard's memory QVL. I haven't really cared about the QVL since second gen Ryzen (First and second gen was a bit of nightmare with RAM), but with four sticks you really should still use the QVL if possible. The QVL will tell you if it passed the control with 2 or 4 sticks. They test all kits with 4 even if they are two stick kits or even single stick. With it not being on the QVL list, it's a dice roll.

 

Buying the same model number RAM as you already have doesn't mean that you get the same RAM. Before every batch the RAM vendor will get the cheapest parts they can get which will do the speeds and timings they need. On a later batch, the RAM could be completely different from the original batch. This makes the QVL even more difficult to follow because no one tells you which parts were used when you buy the RAM, you just have to check after you received it. The QVL will tell you, for most kits, which manufacturer the components were from with the tested kit.

 

With Corsair RAM they print which parts were used on the sticker of each stick. It's encoded because they don't really want people to know about this. If you look at the sticker on each kit, it will have a version number. This is not really the version, it's who the parts are from and which version chipset (What is usually called a die with RAM) was used with that stick. You can see in the screenshot below that all kits from SK Hynix start with a 5. If they don't have the same version number they are not any more similar than any other two random kits. 

 

Example image of the memory QVL for Corsair sticks on your board with your CPU.

 

You can see the component manufacturer in the Chip field (And they include the version number behind the model number) and how many sticks it was stable with in the "DIMM Socket Support" field.

Jesus Christ this is mental, I had no idea RAM was so finicky and so confusing.
I tought RAM was the easiest and most friendly upgrade you could do to a PC...but this stuff is less intuitive than any other part lol...
 

Watching how to build a PC videos from reputable channels, like LTT or Jayz, and also on the store and everywhere else, why is this point not pushed more often?
Pretty much every guide says "make sure you get compatible RAM, and if you want to add more ram, it's highly reccomended to choose the exact same RAM", and that's that, but I had no idea I was supposed to watch out for this obscure version thing or else the system goes haywire...
 

Checking the sticks themselves I can confirm that they are indeed different "versions". The "old" ones have ver 8.31 on the sticks while the "new" ones have ver 3.34
Furthermore, when going into CPU-Z I can also inspect the sticks slot by slot, and there are again differences in the "DRAM manufacturer":
2 are Micron (nice), but the other 2 are "Nanya" (nani?!).
Altough in CPU-Z they are listed "conescutively".... which slot does cpu-z decide is slot 1,2,3... and so on? It treats them like they are next to each other, however on the board they are intertwined.

Man I'm stumped.... why do tech gurus and online stores don't take care to present this aspect to consumers and newbies in their guides and videos??
I had no idea the version is such a sensitive matter.... So you just can't upgrade your RAM like this as time passes? Because how in the hell can you know which version you get, before ordering?
Am I supposed to just order RAM and return it until I get the right version :))) ?

[Rrraaayyy]

For example even if I get THIS specific RAM (the first one, Samsung C), that IS on the QVL list, if I don't get the 4 sticks kit and instead I opt to get the 2 sticks kit, then when I have more money I get another 2 sticks kit, even if it's "verified", because I order at a later date, thus from a different RAM batch, I could instead get a Hypnix and I run in the exact same problem?

[Bjoolz]

25 minutes ago, Rrraaayyy said:

For example even if I get THIS specific RAM (the first one, Samsung C), that IS on the QVL list, if I don't get the 4 sticks kit and instead I opt to get the 2 sticks kit, then when I have more money I get another 2 sticks kit, even if it's "verified", because I order at a later date, thus from a different RAM batch, I could instead get a Hypnix and I run in the exact same problem?

Correct. Mixing can work fine, it just fairly significantly reduces the chance of it working. I just mostly deal with it not working so my feelings about how often it fails is probably skewed a fair bit into the negative. 

 

EDIT: 

28 minutes ago, Rrraaayyy said:

Checking the sticks themselves I can confirm that they are indeed different "versions". The "old" ones have ver 8.31 on the sticks while the "new" ones have ver 3.34

This means that the old kit has parts from Nanya and the new ones have parts from Micron/Spectec (Micron bought spectec and I can't quite remember if they use the same code now). The numbers after the period is how many ranks (Maybe not ranks, don't quite remember, something with how the physical layout of the memory chips is) it has and the chipset version, the C-Die, M-Die, etc.

[Rrraaayyy]

2 minutes ago, Bjoolz said:

Correct. 

But if it happens to be the exact same version with the ones you already had, then it should be all good, yes?
Or are 2x8GB Kits just a "tourist trap", and you should always have 1 single kit in your pc (2x16GB or 4x8GB, in my case), and if you don't, you will have crashes like these (smh ram technology T.T)?

[Bjoolz]

9 minutes ago, Rrraaayyy said:

But if it happens to be the exact same version with the ones you already had, then it should be all good, yes?
Or are 2x8GB Kits just a "tourist trap", and you should always have 1 single kit in your pc (2x16GB or 4x8GB, in my case), and if you don't, you will have crashes like these (smh ram technology T.T)?

I updated my reply with some more information, please refresh. And yes, if the two kits use the same parts then they have a much greater chance of working. If it's on the QVL and validated with four sticks then it should work. There is some silicon lottery with the CPU involved as well, depending on the speed and timings you want. 

 

My experience is probably skewed like I said, but I personally prefer getting a new kit. And usually opting for two sticks over four when possible. 

 

Not related, but the new CPUs REALLY struggle with four sticks and DDR5. The officially supported max speed on 7000 series AMD with four sticks is 3600MT/s, basically DDR4 speeds. Four sticks is just much harder on the memory controller. 

 

EDIT: From AMD

[Whatisthis]

It’s always best to do a 2x kit. Memory controllers have a much harder time with 4 sticks.

 

And as pointed out, things can get weird if you buy the “same” kit at different times.

[Rrraaayyy]

I see, damn, guess I'll have to buy a bigger 2 piece kit, a 2x16GB, if I want to upgrade my RAM, return the "new" ones..... and sell the "old" ones at a loss on the used market O.o, seems really weird but I guess I have no choice.

I really thought buying 2 sticks initially and then another 2 was an easy and simple way to upgrade....guess not.

Thanks for the help, guys!

[Bjoolz]

5 minutes ago, Rrraaayyy said:

I see, damn, guess I'll have to buy a bigger 2 piece kit, a 2x16GB, if I want to upgrade my RAM, return the "new" ones..... and sell the "old" ones at a loss on the used market O.o, seems really weird but I guess I have no choice.

I really thought buying 2 sticks initially and then another 2 was an easy and simple way to upgrade....guess not.

Thanks for the help, guys!

You are far from the first to have this kind of issue and you there will be many more after you. It's one of the things I hate the most about manufacturers, that they can't just change the model number if they do changes to the RAM. 

[Rrraaayyy]

2 minutes ago, Bjoolz said:

You are far from the first to have this kind of issue and you there will be many more after you. It's one of the things I hate the most about manufacturers, that they can't just change the model number if they do changes to the RAM. 

Yes, given the actual consequences of having different versions of RAM, it's mind bendingly bad that you can't see this information more easily, before buying...

[Bjoolz]

4 minutes ago, Rrraaayyy said:

Yes, given the actual consequences of having different versions of RAM, it's mind bendingly bad that you can't see this information more easily, before buying...

And before coming to the conclusion that this actually is the issue, I would test the new RAM alone first. If it crashes when used alone it could be faulty. I should probably have stated that earlier, but I more considered my post informational and supplementary.  Memory testers miss issues with RAM quite often so just using it normally instead. 

[Rrraaayyy]

50 minutes ago, Bjoolz said:

And before coming to the conclusion that this actually is the issue, I would test the new RAM alone first. If it crashes when used alone it could be faulty. I should probably have stated that earlier, but I more considered my post informational and supplementary.  Memory testers miss issues with RAM quite often so just using it normally instead. 

Well I did do some tests with memtest86+ on a USB, and I had tested both the new and old sticks separately and I had no errors.
I also put the ram in the same configuration I had when I got the new RAM and the system crashes, ran overnight for 10 hours, 7 Passes and also had 0 Errors.

I would think that if any of the RAM sticks were faulty it should have shown in the individual tests, if not at least in the 10Hr test, right?

[Bjoolz]

51 minutes ago, Rrraaayyy said:

I would think that if any of the RAM sticks were faulty it should have shown in the individual tests, if not at least in the 10Hr test, right?

I've basically given up on RAM testers because they have come back clean on RAM I know is faulty so many times. 

[Rrraaayyy]

3 hours ago, Bjoolz said:

I've basically given up on RAM testers because they have come back clean on RAM I know is faulty so many times. 

Aah ok now I understand what you said in the post above..
Still, I'd say the stabillity issues are happening because of the different ram versions in combination with the extra strain on the memory controller..
If the RAM was good then.. what, that means bad CPU? Bad memory controller?
I would think that to be less likely so that's why I assume the issues are because of the RAM version difference...

[DoctorNick]

6 hours ago, Rrraaayyy said:

Aah ok now I understand what you said in the post above..
Still, I'd say the stabillity issues are happening because of the different ram versions in combination with the extra strain on the memory controller..
If the RAM was good then.. what, that means bad CPU? Bad memory controller?
I would think that to be less likely so that's why I assume the issues are because of the RAM version difference...

I see that I've created confusion.. The CPU should be able to handle 4 sticks at 3600 just fine. Memory run at very loose 18-22-22-42 timings. Therefore you can set this manually in bios. Go to BIOS and disable XMP. (set to auto, if disable isn't there). Set DRAM frequency to 3600, set dram voltage to: 1.35v. Set primary timings to 18-22-22-22-42 (yes 3 x 22). The screen looks like this under OC tweaker tab:

[Rrraaayyy]

  

9 hours ago, DoctorNick said:

I see that I've created confusion.. The CPU should be able to handle 4 sticks at 3600 just fine. Memory run at very loose 18-22-22-42 timings. Therefore you can set this manually in bios. Go to BIOS and disable XMP. (set to auto, if disable isn't there). Set DRAM frequency to 3600, set dram voltage to: 1.35v. Set primary timings to 18-22-22-22-42 (yes 3 x 22). The screen looks like this under OC tweaker tab:

 

Alright well, after running on non-XMP since your first post, I have now manually applied the settings you mentioned, and we'll see what happens I guess..

I haven't had any crashes since running on non-XMP settings, but it's not enough time to draw a conclusion, haven't had time to do ram tests, and also since it's non XMP it's probably much more stable anyway.

But regarding the different ram "versions", after I've concluded that I have 2 ram kits from two different batches, with different parts, effectively different ram, shouldn't that be problematic?:

22 hours ago, Rrraaayyy said:

Checking the sticks themselves I can confirm that they are indeed different "versions". The "old" ones have ver 8.31 on the sticks while the "new" ones have ver 3.34

Furthermore, when going into CPU-Z I can also inspect the sticks slot by slot, and there are again differences in the "DRAM manufacturer":
2 are Micron (nice), but the other 2 are "Nanya" (nani?!).

[--SID--]

nvm

[Rrraaayyy]

59 minutes ago, Rrraaayyy said:

 Alright well, after running on non-XMP since your first post, I have now manually applied the settings you mentioned, and we'll see what happens I guess..

Yeah so everything fine until I booted up a game and started playing.
2 Game Crashes and 1 glitchy BSOD with restart all in the span of 5 minutes :))))

 

That should indicate that the different RAM kits really don't like each other or maybe I've won the reverse silicon lottery, and my CPU/MOBO has difficulties with this 4 piece RAM combo.... I will continue to RMA the new ones, try to resell the old ones, and I will return to these forums to ask for suggestions about which 2x16 GB Dual channel kit to get, until then I think I'm going back to non XMP settings :))
 

[DoctorNick]

2 hours ago, Rrraaayyy said:

Yeah so everything fine until I booted up a game and started playing.
2 Game Crashes and 1 glitchy BSOD with restart all in the span of 5 minutes :))))

 

That should indicate that the different RAM kits really don't like each other or maybe I've won the reverse silicon lottery, and my CPU/MOBO has difficulties with this 4 piece RAM combo.... I will continue to RMA the new ones, try to resell the old ones, and I will return to these forums to ask for suggestions about which 2x16 GB Dual channel kit to get, until then I think I'm going back to non XMP settings :))
 

Did you test the new kit by themselves aka only the two sticks.? The kit is made to run by itself, also 3600 isn't officially supported by AMD 5600x, only 3200 with two sticks. If the two kits isn't rated for the same timings, you'll have to choose the ones from the kit with the loosest timings (highest numbers) and go with that. It's very important that you only set primary timings as the other timings is specific to 2 sticks. Instead use auto for all secondary timings and lower that after you get it stable. Keep in mind if you undervolt the CPU using curve optimizer, you'll have to disable that while tweaking as it will affect stability.