How far can I push this DDR4 memory?

[r00tb33r]

Up until now I was running an unmatched set of memory, because well, I take care of myself last.  The unmatched set produced very mild memory overclocking results which are not relevant to this thread.

 

So I had the idea, I can shuffle around sticks between machines to come up with a matched set.  I noted that I can get a set of 4 of Kingston ACR24D4U7S8MB-8 (DDR4-2400), Micron chips labeled 6TB75 D9TBH, late 2016, early 2017.

 

How far could can these be pushed?  What numbers should I be aiming for?

 

I'm starting with my current baseline of DDR4-3200 (133x24) 16-19-19-52-1 at 1.35v, gear 1.  How far should these go if they're matched?

 

The board is a Z790 with a 13700F.

[RONOTHAN##]

This IC is Micron 8Gb. Rev. B, which shares no resemblance to the 16Gb. Rev. B that you'll find much more info on, at least recently. From what I know this IC does similar timings to Micron 8Gb. Rev. E but with no where near the frequency headroom you can find. Most people I've seen runnings these are right around the 3200MT/s mark, so you're probably not gonna get too much more frequency out of them (3600 might be the absolute most you can hope for, though then again at 3600 you'd start dealing with IMC issues to stay in Gear 1). Feel free to try however, worst that will happen is you need to clear CMOS. 

 

As for timings, you can probably push tRCD, tRP, and tRAS down slightly, so your primaries at 3200 are something like 16-18-18-36 or 16-17-17-34 depending on how lucky you are (given these are unbinned kits, probably the looser of the two). I don't know how they'd scale with voltage, again the info on this IC is rather scarce since not many people bothered to overclock it, so I'd stick with the 1.35V you're using as that should be safe. As for the sub-timings if you were to bother tuning them (realistically, that's going to be the only area where you might notice some improvement from where you currently at), find some Rev. E sub-timings off the internet, try them one at a time and see if they work.

[r00tb33r]

21 minutes ago, RONOTHAN## said:

This IC is Micron 8Gb. Rev. B

That's what it seemed like, but how exactly did you identify it?  I couldn't figure out which markings correspond to that.

 

Huh, sounds like almost not worth the effort.

[RONOTHAN##]

4 minutes ago, r00tb33r said:

That's what it seemed like, but how exactly did you identify it?  I couldn't figure out which markings correspond to that.

 

Huh, sounds like almost not worth the effort.

https://www.micron.com/sales-support/design-tools/fbga-parts-decoder

You plug the printed number (D9TBH) into the FBGA code, and it outputs the data sheet for that exact IC. The last number in the full part number is the die revision. 

 

I would agree though, it's probably not gonna be worth the effort. I'd probably see if a higher frequency would work since why not, but I wouldn't put any effort into getting it stable if it doesn't immediately work. If you want to do anything else, set tREFI to 65000 as that should bump memory performance a bit and should be stable. 

[r00tb33r]

2 hours ago, RONOTHAN## said:

https://www.micron.com/sales-support/design-tools/fbga-parts-decoder

You plug the printed number (D9TBH) into the FBGA code, and it outputs the data sheet for that exact IC. The last number in the full part number is the die revision. 

Thanks.

 

So I ran 2 sticks of D9TBH ICs in dual channel (one Kingston, one Micron OEM), the numbers I came up with are DDR4-3600 (133x27) 18-21-21-38-1 at 1.35v, gear 1.  From what I found, this is reasonably in line with DDR4-3600 CL18 kits, which seem to commonly be 18-22-22-42, so I guess this is fine?

 

It was too unstable at 3733 and there was no more timing or voltage that could help it.

 

AIDA64 benchmark passed, so that's a start.  What do you recommend to burn this in with?

 

Hopefully it will work with those numbers with 4 matched sticks tomorrow.

 

2 hours ago, RONOTHAN## said:

If you want to do anything else, set tREFI to 65000 as that should bump memory performance a bit and should be stable. 

Did that, thanks.  I was also peeking into the DDR4 overclocking guide:

https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md

 

It mentioned that older Micron chips (presumably like mine) have negative scaling beyond 1.35v, which has also been my observation.  At 1.40v was no go.

Likewise it recommended tCCDL of 8 for Intel, I did that early, it seemed to help.

 

About 19.5% bandwidth gain in AIDA64.  Is 60ns latency a good result?

[RONOTHAN##]

5 minutes ago, r00tb33r said:

It mentioned that older Micron chips (presumably like mine) have negative scaling beyond 1.35v, which has also been my observation.

To be fair, I've only ever heard of that referring to the 4Gb ICs. That said, I've also seen Samsung B die (an IC that can scale to over 1.7V) not scale about 1.35V, so those rules of thumb aren't always accurate. Every kit is individual, after all. 

 

If you got 3600MT/s CL18-21-21 to work, that's great. You might need to drop to 3533 with 4 sticks depending on your IMC quality, but that's pretty great. 

 

10 minutes ago, r00tb33r said:

What do you recommend to burn this in with?

The two tests I've had the best luck with memory stress testing on 13th gen are Y-Cruncher VT3 (this errors with frequency) and TestMem5 (Anta777 Extreme1 preset I've had better luck with DDR4, though some people swear by Absolut or 1usmus_v3). I'd do an hour of VT3 and 4 hours of TestMem5 (this should be 8 loops of Extreme1, though it's been a while since I've done 13th gen DDR4 tuning so I could be wrong)

 

14 minutes ago, r00tb33r said:

From what I found, this is reasonably in line with DDR4-3600 CL18 kits, which seem to commonly be 18-22-22-42, so I guess this is fine?

If you're curious why, there's an IC, Samsung 8Gb C die, that will very reliably do 3600 CL18-22-22-42. No looser no tighter, very little variance, and the 3600 CL18 speed bin was built almost exclusively around that. Micron 8Gb Rev. B is probably similar, though it likely could do CL16 if you had better sticks. 

[r00tb33r]

Okay, so today I have 4 sticks of Kingston ACR24D4U7S8MB-8 (DDR4-2400), 2 of datecode week 45 of 2016, 2 of datecode week 15 of 2017.

 

3 of the sticks work at DDR4-3600 (133x27) 18-21-21-38-1 at 1.35v, gear 1 setup from yesterday

 

One stick refuses to get with the program.  The absolute best it could do stable is 3500, datecode week 15 of 2017.  Even less together with other sticks.

 

22 hours ago, RONOTHAN## said:

To be fair, I've only ever heard of that referring to the 4Gb ICs. That said, I've also seen Samsung B die (an IC that can scale to over 1.7V) not scale about 1.35V, so those rules of thumb aren't always accurate. Every kit is individual, after all. 

So while chasing that one uncooperative stick here are my observations regarding voltage...

3200 baseline at 1.40v ok.

3400 at 1.40v ok.

3500 at 1.37v ok, 1.40v fail.

3600 partial boot at 1.35v, 1.37v fail.  1.33v was also fail.

So I think at least for this case it confirms the negative correlation mentioned.  I'm sure the B-die Samsung chips do perform with a positive correlation as it is said, but I had only 1 of those sticks, so I swapped it out.

22 hours ago, RONOTHAN## said:

If you got 3600MT/s CL18-21-21 to work, that's great. You might need to drop to 3533 with 4 sticks depending on your IMC quality, but that's pretty great. 

Well, I'm running the 3 cooperative sticks together at DDR4-3600 (133x27) 18-21-21-38-1 at 1.37v, gear 1.  I crashed earlier at 1.35v.  We'll see how it goes.  I'm not yet sure it's 100% stable, but we'll just have to continue testing.

 

I think I'll swap the one uncooperative stick for that Micron OEM stick, since that ran 3600 fine the night before, and the chip FBGA code was the same.  I'll call it close enough.  I guess I'll have to do this again tomorrow.

[RONOTHAN##]

43 minutes ago, r00tb33r said:

I think I'll swap the one uncooperative stick for that Micron OEM stick, since that ran 3600 fine the night before, and the chip FBGA code was the same.  I'll call it close enough

Yeah, swap it to this and see what happens. Hopefully it works. Good luck!

[r00tb33r]

21 hours ago, RONOTHAN## said:

Yeah, swap it to this and see what happens. Hopefully it works. Good luck!

With that one I have 4 sticks running DDR4-3600 (133x27) 18-21-21-38-1 at 1.37v.  I bumped that voltage from 1.35v when I crashed yesterday, if going by that negative stability correlation it would seem counter-intuitive, but I figured I'd try that with 4 sticks thinking that maybe it would help with the increased capacitance of the second stick per channel.  I admit I could be completely wrong about it.

For now AIDA64 memory benchmarks passed.  I did not run any burn-it yet.

 

Overnight the system did not crash (as in Windows didn't crash), but I don't think it's 100% stable either.  For example my AX210 WiFi device got knocked out to the point I had to reinstall the driver to get the device to start again.  I've seen a Chrome tab crash, which while isn't unheard of for a Google product, doesn't happen very often.

 

What can I tweak to get it more stable?

[RONOTHAN##]

7 minutes ago, r00tb33r said:

What can I tweak to get it more stable?

It might be tRCD that's not quite stable, that can give errors rather late into stress tests. I'd bump it and tRP up to 22 and see if that helps.

[r00tb33r]

1 hour ago, RONOTHAN## said:

It might be tRCD that's not quite stable, that can give errors rather late into stress tests. I'd bump it and tRP up to 22 and see if that helps.

Did that just now.  We'll see how it goes.

 

Do you know by chance which BIOS option locks the base clock to exact 100MHz as opposed to something like 99.8?  Some kind of PLL control?  I've seen SkatterBencher show it in one of the videos but I don't remember which.

 

[EDIT]

Huh, I couldn't find any BCLK options at all, it's like this board doesn't support it, or doesn't show it for a non-K processor?

MAG Z790 Tomahawk Wifi DDR4...

[RONOTHAN##]

15 minutes ago, r00tb33r said:

Did that just now.  We'll see how it goes.

 

Do you know by chance which BIOS option locks the base clock to exact 100MHz as opposed to something like 99.8?  Some kind of PLL control?  I've seen SkatterBencher show it in one of the videos but I don't remember which.

It should be disabling spread spectrum, though in my experience unless the board has an external clock gen, it doesn't always bring it up from 99.8MHz. 

[r00tb33r]

On 7/20/2024 at 1:39 AM, RONOTHAN## said:

If you want to do anything else, set tREFI to 65000 as that should bump memory performance a bit and should be stable. 

Through a long process of elimination I found that this value isn't stable on my memory, board, processor combo.  Do you know a formula or how I can find a more appropriate value for tREFI?  For now I'm running the default the board sets so I can pass some long memory tests.

On 7/20/2024 at 3:50 AM, RONOTHAN## said:

TestMem5 (Anta777 Extreme1 preset I've had better luck with DDR4

Thanks, seems like a pretty solid test.  I'm having difficulty getting the DDR4-3600 config to pass, can only do a few minutes before the first error.  So far I'm only passing at DDR4-3466 18-22-22-56-1 at 1.37v.  (I haven't worked on the tRAS number yet.)  It also does CL17 at 3400 but not beyond.

 

3500 is the least stable, I don't know why, hard to even get it to boot into Windows.  3600 is more stable, it will run for a day (I didn't give it time beyond that), will download a 100Gbyte torrent with only 7 damaged pieces...  You get the idea.  But, not passing the test.

 

Here are the things that I found.  Beyond 3200 1.37v is the most stable voltage for 4 sticks.  1.38v is unworkable, 1.35v and 1.36v are less stable than 1.37v.  This memory seems to strongly favor command rate 1T, and CL18 (and no higher).  Likewise for tRCD and tRP numbers beyond 22 tend to be less stable until about 30.  Is there a formula that explains why longer timing delays are less stable?

 

Any ideas what else I could do to get 3600 more stable?

 

On 7/22/2024 at 1:15 AM, RONOTHAN## said:

It should be disabling spread spectrum, though in my experience unless the board has an external clock gen, it doesn't always bring it up from 99.8MHz. 

Yeah, I understand that success getting a perfect number out of a PLL isn't guaranteed.  Seems my BIOS is missing all BCLK options and spread spectrum.  Folks on MSI forums suggested I update the BIOS...  Seems like a last ditch thing.  I'll do that when I get around, as I've got a nice undervolt config and I checked that configs saved to disk have the BIOS version numbers in them, meaning the new BIOS could reject the config from an older version, and I'm not really loving the prospect of re-doing everything it took to get a good undervolt on a non-K processor where CEP can't be disabled.

[RONOTHAN##]

2 hours ago, r00tb33r said:

Do you know a formula or how I can find a more appropriate value for tREFI?

There really isn't. tREFI is the amount of time between memory refreshes (a RAM chip is just a fancy capacitor, and they will need to be recharged after a certain amount of time), and it tends to error later into stress tests as this is usually pretty temperature sensitive. Because of that, I don't tune it very granularly, I basiaclly check ~262k (the register limit) if I have some memory cooling, ~131k if I don't, then keep halving this value until it's stable. I have yet to find a memory kit where 65k doesn't work, but it seems like you have. I'd drop it down to ~32k and see if it works, otherwise just leave it on auto (auto should be ~12k IIRC). 

 

2 hours ago, r00tb33r said:

Any ideas what else I could do to get 3600 more stable?

Messing with the CPU VDDQ setting can sometimes help, though the fact you don't have VCCSA control is really the limitation here. As for what VDDQ setting you should try, this varies a lot from one board to the next and even one BIOS revision to the next, so I've seen anywhere from 1.1V be ideal to 1.45V be ideal on just MSI motherboards. 

 

2 hours ago, r00tb33r said:

Is there a formula that explains why longer timing delays are less stable?

Not really. This is just a quirk I've seen on the 13th gen memory controller, where certain memory voltages work best at certain timings. No idea why it does that, it just sometimes does. As for 3500 being less stable, Intel is known for having the 1.33 multliplier (or strap and it's called by the XOC folks) be more stable at high frequencies than the 1.00 strap. 9900Ks, for instance, would have chips do 4133MT/s with relative ease, but 4100 would be very unstable. 

[r00tb33r]

2 hours ago, RONOTHAN## said:

There really isn't. tREFI is the amount of time between memory refreshes (a RAM chip is just a fancy capacitor, and they will need to be recharged after a certain amount of time), and it tends to error later into stress tests as this is usually pretty temperature sensitive. Because of that, I don't tune it very granularly, I basiaclly check ~262k (the register limit) if I have some memory cooling, ~131k if I don't, then keep halving this value until it's stable. I have yet to find a memory kit where 65k doesn't work, but it seems like you have. I'd drop it down to ~32k and see if it works, otherwise just leave it on auto (auto should be ~12k IIRC). 

Yeah, the board is doing like 13.5k right now.  Cool, will try those numbers once I find the maximum stable data rate.

2 hours ago, RONOTHAN## said:

Messing with the CPU VDDQ setting can sometimes help, though the fact you don't have VCCSA control is really the limitation here. As for what VDDQ setting you should try, this varies a lot from one board to the next and even one BIOS revision to the next, so I've seen anywhere from 1.1V be ideal to 1.45V be ideal on just MSI motherboards. 

System agent voltage?  There is "CPU SA Voltage Mode" and "CPU SA Voltage".  I just now I started with VDDQ of 1.35v and an offset of +0.1v on SA (0.934+0.1).  That seemed to make it unstable to the point it didn't boot into Windows.  I tried 1.37v for both, for no rational reason.  In the TestMem5 test it lasted maybe just a little shorter than before.  I'll try 1.45v for both, since it's been said that's about the maximum safe for both.

2 hours ago, RONOTHAN## said:

Not really. This is just a quirk I've seen on the 13th gen memory controller, where certain memory voltages work best at certain timings. No idea why it does that, it just sometimes does. As for 3500 being less stable, Intel is known for having the 1.33 multliplier (or strap and it's called by the XOC folks) be more stable at high frequencies than the 1.00 strap. 9900Ks, for instance, would have chips do 4133MT/s with relative ease, but 4100 would be very unstable. 

Yeah, that also seemed to be the case on the 11th gen system you were helping me with last year I think.

[RONOTHAN##]

2 minutes ago, r00tb33r said:

System agent voltage?  There is "CPU SA Voltage Mode" and "CPU SA Voltage".  I just now I started with VDDQ of 1.35v and an offset of +0.1v on SA (0.934+0.1).  That seemed to make it unstable to the point it didn't boot into Windows.  I tried 1.37v for both, for no rational reason.  In the TestMem5 test it lasted maybe just a little shorter than before.  I'll try 1.45v for both, since it's been said that's about the maximum safe for both.

yeah, system agent. The non-K LGA 1700 CPUs have the SA voltage locked to 1V or lower, so even though some motherboards let you set it higher, it just rounds down to 1V. 

[r00tb33r]

1 minute ago, RONOTHAN## said:

yeah, system agent. The non-K LGA 1700 CPUs have the SA voltage locked to 1V or lower, so even though some motherboards let you set it higher, it just rounds down to 1V. 

Yeah, I think you're right.  I'm looking in HWMonitor and it shows SA 0.934v, meaning it just ignored whatever I set.  I guess I can try setting 1.0v to see if it will take that.

 

VDDQ of 1.45v was a BSOD on boot, though, then.

[RONOTHAN##]

4 minutes ago, r00tb33r said:

Yeah, I think you're right.  I'm looking in HWMonitor and it shows SA 0.934v, meaning it just ignored whatever I set.  I guess I can try setting 1.0v to see if it will take that.

 

VDDQ of 1.45v was a BSOD on boot, though, then.

Do know that SA readings seem to be bugged on some CPUs and/or motherboards, where it changing the voltage won't report it being different unless you did a clear CMOS before hand (it does still change, but it doesn't look like it changed). 

 

The newer BIOS revisions tend to prefer lower VDDQ from what I've seen, so take that for what you will. I don't know what revision you're on though. 1.35V was usually optimal on most of the earlier BIOSes though (it was only one I remember where 1.45V was necessary)

[r00tb33r]

15 minutes ago, RONOTHAN## said:

Do know that SA readings seem to be bugged on some CPUs and/or motherboards, where it changing the voltage won't report it being different unless you did a clear CMOS before hand (it does still change, but it doesn't look like it changed). 

 

The newer BIOS revisions tend to prefer lower VDDQ from what I've seen, so take that for what you will. I don't know what revision you're on though. 1.35V was usually optimal on most of the earlier BIOSes though (it was only one I remember where 1.45V was necessary)

So I did a CMOS reset, and immediately went into the BIOS, loaded the profile, set SA offest to +0.05v (+50mV), still the default 0.934v in HWMonitor sadly, should have been 0.984v instead.

 

The default VDDQ the board sets is 1.2v.  1.4v was also a BSOD on boot, so was 1.1v.

[RONOTHAN##]

2 minutes ago, r00tb33r said:

So I did a CMOS reset, and immediately went into the BIOS, loaded the profile, set SA offest to +0.05v (+50mV), still the default 0.934v in HWMonitor sadly, should have been 0.984v instead.

 

The default VDDQ the board sets is 1.2v.  1.4v was also a BSOD on boot, so was 1.1v.

It might not be a Clear CMOS that does it, I forget how I've gotten my boards reporting correctly after changing it. There is a way to do it, but I'm too tired to spend the hour troubleshooting on my Unify-X to double check. It should be running at 1V though, which is the highest it works at. 

 

Try 1.3V VDDQ and see if that helps much. That said, VDDQ is probably not the limiting voltage, and therefore won't help you get high memory speed to work. 

 

 

It's also worth noting that there is another you could do, tuning termination settings manually, but generally that takes hours, if not days to dial them in by pure trial and error, and there's no guarantee that it would even help. If you value your time at all, I wouldn't do this. 

[r00tb33r]

7 hours ago, RONOTHAN## said:

There really isn't. tREFI is the amount of time between memory refreshes (a RAM chip is just a fancy capacitor, and they will need to be recharged after a certain amount of time), and it tends to error later into stress tests as this is usually pretty temperature sensitive. Because of that, I don't tune it very granularly, I basiaclly check ~262k (the register limit) if I have some memory cooling, ~131k if I don't, then keep halving this value until it's stable. I have yet to find a memory kit where 65k doesn't work, but it seems like you have. I'd drop it down to ~32k and see if it works, otherwise just leave it on auto (auto should be ~12k IIRC).

Yeah, I have a general idea of how DRAM works.  I could not find a better tREFI number that works that is better than what the board defaults to.  Everything else fails the test or crashes even earlier.

4 hours ago, RONOTHAN## said:

Try 1.3V VDDQ and see if that helps much. That said, VDDQ is probably not the limiting voltage, and therefore won't help you get high memory speed to work. 

Yeah, I tried values in 1.1v to 1.45v range, including 1.25v, 1.30v, 1.35v, etc.  I think stability only decreased.  (Foreshadowing?)

4 hours ago, RONOTHAN## said:

It's also worth noting that there is another you could do, tuning termination settings manually, but generally that takes hours, if not days to dial them in by pure trial and error, and there's no guarantee that it would even help. If you value your time at all, I wouldn't do this. 

You mean VTT?  I looked around and saw a mention of DDR4 Micron B-die on Reddit (probably the more recent one, which isn't the same), and a daily VTT of 0.75v to 0.8v, also mentioned a negative correlation.  The BIOS default appears to be 0.6v, at least that's what it says in the option hint.  So I tried values in the range from 0.5v to 0.8v.  Stability was only decreasing.

 

So I reverted to the saved preset before I messed with VDDQ, SA, or VTT.  It basically wouldn't do DDR4-3600 anymore, either BSOD during boot, or fails the TestMem5 test within the first 3 seconds.  I did a CMOS reset, loaded the profile again.  Still no go at 3600.  Something has degraded.  Before I could do 4 or more minutes of the test before the first error.

 

I loaded the preset for DDR4-3466 that passed tests in MemTest5 Extreme @ anta777 preset.  Fails in seconds.  I reset CMOS again and loaded my DDR4-3200 baseline that I ran for about a year.  That seemed to be fine.  I loaded the DDR4-3466 preset again.  It seemed fine running the test.  I tried bumping it to 3600 again, instant fail.  Ended up resetting CMOS again.

 

Testing to see if it will even do DDR4-3466 anymore.  Of course this all comes with a risk but it's not like I've gone into extreme voltages with this yet, at least I don't think I did.

[RONOTHAN##]

3 minutes ago, r00tb33r said:

You mean VTT?

No, I mean the actual termination settings of the memory system. There's something like 20 different options for it IIRC, called things like RSQ TX, and most of them are resistance values (they'll be towards the bottom in the memory timing menu, below all the subtimings). You tune these to minimize signal reflections in the memory system and therefore higher memory frequencies are easier for the memory controller to understand and control. Unfortunately I can't give you any tips on how to do that if you really want to, I've only tried to do it twice, once on AMD and once on Intel, and only on the AMD setup did I have any amount of luck (went from 3800MT/s 1T GDM off blue screening instantly to being able to run benchmarks, though not fully stable). All I can say is that it takes forever on AMD, and it takes even longer on Intel since there's a couple more options as well. 

 

VTT is completely different, and while manually setting it can help, I've only seen it help when memory voltages exceed 1.5V. 

 

11 minutes ago, r00tb33r said:

Testing to see if it will even do DDR4-3466 anymore.  Of course this all comes with a risk but it's not like I've gone into extreme voltages with this yet, at least I don't think I did.

I doubt anything degraded since you haven't used any voltage remotely dangerous. I see two possibilities: 3466 is right on the edge of stability for this CPU and it needs to have the frequency walked up to in order to work (older Intel had this happen in some instances), or you're running into retrain instability. I don't personally know which is more likely, since I've yet to hear of 13th gen exhibiting either behavior especially at only 3466, but at this point I'd probably just drop the frequency back down to 3200 and leave it there as either behavior is not something you want to fight to fix on a daily system. 

[Levent]

I believe I have got the same ICs in my shitty old 2133 kit. I have taken those up to 4000mhz with CL22 on a Zen2 CPU. Voltage needed past 3600mhz is not worth it, it also does not tighten timings that nicely. So far, 3 years later of OCing later, they are still working fine for me at 1.38v at 3600C18. I dont remember what I had rest of the timings set at though.

 

[r00tb33r]

22 minutes ago, Levent said:

I believe I have got the same ICs in my shitty old 2133 kit. I have taken those up to 4000mhz with CL22 on a Zen2 CPU. Voltage needed past 3600mhz is not worth it, it also does not tighten timings that nicely. So far, 3 years later of OCing later, they are still working fine for me at 1.38v at 3600C18. I dont remember what I had rest of the timings set at though.

 

I highly doubt that because this memory does not work at all at voltages above 1.4v, with negative stability correlation with voltage, consistent with information posted elsewhere.

 

You have something else.

[r00tb33r]

1 hour ago, RONOTHAN## said:

I doubt anything degraded since you haven't used any voltage remotely dangerous. I see two possibilities: 3466 is right on the edge of stability for this CPU and it needs to have the frequency walked up to in order to work (older Intel had this happen in some instances), or you're running into retrain instability. I don't personally know which is more likely, since I've yet to hear of 13th gen exhibiting either behavior especially at only 3466, but at this point I'd probably just drop the frequency back down to 3200 and leave it there as either behavior is not something you want to fight to fix on a daily system. 

I dunno, it used to pass laps in TestMem5...  Now it failed 20 seconds in, and failed again.  It just doesn't go anymore.  It just isn't the same as it's been earlier in the week.

 

I walked up from 3200, passing TestMem5 at 3333, 3400, and 3466 saving presets along the way.  3500 was the one that didn't work at all.  3600 ran for two days (1 at a time), and while not passing the test, not crashing with a BSOD.  Now I get BSOD after running the test which fails 3 seconds in instead of minutes like it did before, if it boots at all.

 

Now I'm back to the saved 3400 preset.  Hopefully it will be a pass (again).