Question Random Memory BSOD's with XMP Profile

[sze5003]

I'm using an Asus z370E board with corsair lpx
(CMK16GX4M2B3000C15W) memory. 4 sticks so 32Gb total.

I used to have 16gb and bought another 2 sticks 2 years ago.

With my 8700k enabling the XMP profile for 3000mhz has not given me any issues. In fact it has ran like this for 3-4 years now.

Recently I've swapped my 8700k for a 9900k (this past weekend actually). Enabling the XMP profile with the 9900k I thought everything was fine for a few hours, until I started getting random BSOD's like Memory_management and IRQL_Not_Less_or_Equal, etc.

So I disabled XMP and changed the frequency in bios to 2993 and that also would bsod after some time in windows. I have settled on 2666mhz and it ran fine for a day or two without any blue screens.

Is the cpu bad or just luck related to the memory controller?

 

[Furious_Styles]

Might need to bump the IMC voltage. Also are the sticks exactly the same or different in some way?

 

[deustroop]

Intel specifies default memory speed at 2666 .This should be the speed if there is no manual input after installation. Can you achieve this ?

 

[sze5003]

Might need to bump the IMC voltage. Also are the sticks exactly the same or different in some way?

They should be the same sticks. I made sure when I bought an additional 16gb that I got the same model ram.

I've looked into setting the voltage and timings manually but haven't found any guides for my board that explain what I should be changing.

Intel specifies default memory speed at 2666 .This should be the speed if there is no manual input after installation. Can you achieve this ?

Default memory speed for these modules are 2133 which I did not see any issues with when I first swapped cpu's.

I manually changed it to 2666 and no problems there either but I was only able to have the pc on for 2 days at this speed before leaving.

Currently I'm on vacation so I'll be testing it some more when I get back. Usually if it blue screens, it would happen within 10-30 minutes when I was attempting to use higher speeds or the XMP profile.

 

[Furious_Styles]

You just want to bump the IMC and perhaps ddr voltage +0.1 or 0.2

 

[sze5003]

You just want to bump the IMC and perhaps ddr voltage +0.1 or 0.2

I'll have to try that when I get back home. I have also read about buying the fastest ram available for my board and then selling mine. But I don't feel like going through that. Ideally 3200mhz or 3600 is ideal for the 9900k?

 

[Furious_Styles]

I'll have to try that when I get back home. I have also read about buying the fastest ram available for my board and then selling mine. But I don't feel like going through that. Ideally 3200mhz or 3600 is ideal for the 9900k?

You won't see much difference between those sets. Even C14 vs. C16 won't be noticeable/worth the hassle. However if you do choose to go that route get a 2x16gb set to take some pressure off of the IMC.

Hopefully those voltage adjustments take care of it though.

 

[BonzaiDuck]

Just a heads-up regarding "OC" RAM, XMP profiles and motherboard BIOS's.

Some folks go goo-gah over the DDR4-4000 RAM sticks, but I just chose to uses DDR4-3200 and stuck with it. I bought good RAM. Starting with a 2x8= 16GB kit of G.SKILL Trident-Z 14-14-14 modules, I decided I needed and wanted more, so I bought another kit. No problem there, either.

Then, this January, I had a motherboard USB-controller disaster from a static charge through a vaping pen -- so stooooo-pid. The horror! The . . . Horror! . . . . I went from high to low in a hurry.

Replaced the board. Decided to buy 2x16GB Trident-Z 14-14-14 modules. The previous four modules will go into another system I'm building. Before even booting to Windows, I ran HCI-Memtest-64 and they were perfect after 400% coverage. The RAM is still OK.

But I'd made a mistake with the BIOS configuration. I chose the XMP profile for 3200Mhz, leaving all the settings to "Auto" for the new motherboard. [New motherboard is also just fine.] I found that my IMC (VCCIO) voltage was given 1.33V by the motherboard. 1.35V is considered the "Edge of Danger". (And no wonder my CPU temperatures were a bit hot.)

Then, I discovered that I'd forgotten to set up the RAM "properly". "Properly" means that you set the RAM to the Intel spec -- in the OP's case, 2666. You then take note of the IMC/VCCIO and VCSSA voltages offered by "Auto" settings.

After everything else, you can now pick the XMP profile and RAM speed under the RAM-maker's spec. But you need to set the VCCIO and VCCSA voltages to the values you observed under the Intel RAM spec, and then edge them up a bit here or there to assure stability. (So . . . . . you have to test . . . , so download the OCCT program and get to work, or set up HCI Memtest-64 to run multiple instances -- if you don't want to wait 3 days for 300 or 400% coverage. )

I can't be sure about the OP's problem, but you will have instability if the VCCIO and VCCSA voltages are not set properly relative to each other. Worse, if the motherboard volted them too high, I could only speculate that the CPU's IMC has -- or is -- taking a dump. I say I speculate. Who knows?! Some folks have reported that it takes about a month to destroy the IMC with over-volting, even if the motherboard does it. While MY motherboard only pushed the IMC to the Edge of Danger, I wouldn't know about the particulars for anyone else.

BY THE WAY. SZE5003 should get a crash-dump for his BSODs. Download BlueScreenView and run it. Also -- check the Windows Event Viewer -- look for Critical Error 41 and find the Bug-Check code. The Bug Check code -- usually the last three hexadecimal digits -- will be an indicator to the cause.

 

[deustroop]

I'll have to try that when I get back home. I have also read about buying the fastest ram available for my board and then selling mine. But I don't feel like going through that. Ideally 3200mhz or 3600 is ideal for the 9900k?

I run the 9900k @ 3.2GHz but have set that speed manually ignoring XMP.

 

[BonzaiDuck]

I run the 9900k @ 3.2GHz but have set that speed manually ignoring XMP.

Can't be sure if you mean the CPU or the RAM. If it's the CPU, it would only make sense to allow VCORE to run at "Auto", giving the Intel processor spec speed and turbo.

If you mean the RAM, you can choose to set the speed manually with the four basic timings. But there's the exception that you won't get the best extended timings/latency settings -- those beyond the commonly-known four. At this point, and only from my personal observations over the last several years, I believe the XMP settings should be fine except for the two voltages (VCCIO for the IMC, and VCCSA, which "follows" or needs adjustment along with VCCIO). The RAM voltage, as far as I know but possibly with some exceptions for certain brands and models, should be 1.35V for the XMP profile. Again, that may differ, although I couldn't imagine it differing between motherboards for the same types of DDR4 RAM.

It has been my most recent experience that the default voltages for VCCIO and VCCSA, with speed set at the Intel memory spec (2,133, 2,666, etc.), yield a good baseline and may only need to be adjusted upward no more than 0.10 to 0.20V, assuming that the default you find for the Intel speed spec is around 0.9V+. Apparently when they spec "OC" RAM, they choose to make it run at those spec speeds, with the RAM itself volted to 1.35V and the VCCIO and VCCSA volted pretty close to the default for the Intel spec. In fact, if the CPU itself is set to "Auto" VCORE and speed, you may not need to change the VCCIO and VCCSA from the default values yielded by "Auto" at the Intel spec RAM speeds.

Anyone here could correct me about these particulars, and I would be happy for that, nevertheless.

We can talk about these things to great length, but the central difficulty here is the OP's instability and the likelihood that he either has defective RAM or improperly set voltages, or that the latter have damaged his processor IMC. At the moment, I obviously can't say which it is. But there's a path to finding out, and the first steps would be getting the Bug-check code from his crash-dumps and Event Viewer log, and then moving forward possibly to test one RAM stick at a time. The last thing he would want to face is damage to the processor -- if any.

 

[sze5003]

Can't be sure if you mean the CPU or the RAM. If it's the CPU, it would only make sense to allow VCORE to run at "Auto", giving the Intel processor spec speed and turbo.

If you mean the RAM, you can choose to set the speed manually with the four basic timings. But there's the exception that you won't get the best extended timings/latency settings -- those beyond the commonly-known four. At this point, and only from my personal observations over the last several years, I believe the XMP settings should be fine except for the two voltages (VCCIO for the IMC, and VCCSA, which "follows" or needs adjustment along with VCCIO). The RAM voltage, as far as I know but possibly with some exceptions for certain brands and models, should be 1.35V for the XMP profile. Again, that may differ, although I couldn't imagine it differing between motherboards for the same types of DDR4 RAM.

It has been my most recent experience that the default voltages for VCCIO and VCCSA, with speed set at the Intel memory spec (2,133, 2,666, etc.), yield a good baseline and may only need to be adjusted upward no more than 0.10 to 0.20V, assuming that the default you find for the Intel speed spec is around 0.9V+. Apparently when they spec "OC" RAM, they choose to make it run at those spec speeds, with the RAM itself volted to 1.35V and the VCCIO and VCCSA volted pretty close to the default for the Intel spec. In fact, if the CPU itself is set to "Auto" VCORE and speed, you may not need to change the VCCIO and VCCSA from the default values yielded by "Auto" at the Intel spec RAM speeds.

Anyone here could correct me about these particulars, and I would be happy for that, nevertheless.

We can talk about these things to great length, but the central difficulty here is the OP's instability and the likelihood that he either has defective RAM or improperly set voltages, or that the latter have damaged his processor IMC. At the moment, I obviously can't say which it is. But there's a path to finding out, and the first steps would be getting the Bug-check code from his crash-dumps and Event Viewer log, and then moving forward possibly to test one RAM stick at a time. The last thing he would want to face is damage to the processor -- if any.

I probably should have waited to come back from vacation before making this post so I could test everything thoroughly but it was bothering me so I had to.

The default voltage is 1.35 for the XMP profile. I've only had the 9900k running for 5-10 minutes at that xmp profile before I started getting blues screens.

This same ram was running for 3 years using the XMP profile on the 8700k.

I believe it's a matter of upping the voltages rather than having destroyed the memory controller so quickly.

There's a plethora of dram config options on my asus z370E board so i will need to figure out exactly which ones I'll have to change.

Before I left my house I turned off XMP and manually set the frequency to 2666 and it was fine after that.

I'll try and get the event viewed bug code and see what may reveal.

 

[BonzaiDuck]

I probably should have waited to come back from vacation before making this post so I could test everything thoroughly but it was bothering me so I had to.

The default voltage is 1.35 for the XMP profile. I've only had the 9900k running for 5-10 minutes at that xmp profile before I started getting blues screens.

This same ram was running for 3 years using the XMP profile on the 8700k.

I believe it's a matter of upping the voltages rather than having destroyed the memory controller so quickly.

There's a plethora of dram config options on my asus z370E board so i will need to figure out exactly which ones I'll have to change.

Before I left my house I turned off XMP and manually set the frequency to 2666 and it was fine after that.

I'll try and get the event viewed bug code and see what may reveal.

Let's review the important background facts.

VDIMM, VRAM or whatever it's called is the voltage to RAM, and for OC DDR4 RAM my experience tells me it should be 1.35V at the RAM-maker's spec. At the Intel spec and lower speed (i.e., 2666), it might be 1.20V.

VCCIO is the usual designation for voltage to the CPU's IMC or integrated memory controller. Between my Skylake or Kaby quad-cores and your two generations of processors with the Z370E board, nothing essential about the architecture should have changed: they're all Skylake cores, so the IMC is still part of the CPU.

VCCSA or system agent voltage for the "System Agent": SA handles all the IO between CPU and other components like PCIE and also affects RAM overclocking. Again, I can't see how much anything would have changed between my old chipset and your newer chipset. The VCCSA seems to "follow" VCCIO voltage requirements. I'm inclined to say that it should be just a little bit more than VCCIO voltage, and the spread between VCCIO and VCCSA has a significant impact on stability.

As long as the XMP profile sets RAM voltage to 1.35V monitored value, you could either leave it at "Auto" or just fix it at 1.35V.

The other two voltages for the Intel "default" speed (2666 for yours and 2133 for mine) should default on "auto" to between 0.9+V and 1.1+V. That's the range where you'd start making minor tweaks to those voltages for "OC" RAM -- IF NECESSARY.

And my best understanding of it -- I could be mistaken or slightly inaccurate -- If the processor is set to run at "Auto" speed and VCORE, you wouldn't need to change the VCCIO and VCCSA for "OC" RAM. But they shouldn't be pushing up toward 1.35V.

I just wanted to make sure you understood the difference between the RAM voltage spec'd at 1.35V for OC, and the VCCIO/VCCSA voltages which should never exceed 1.35V and are still likely overvolted if they exceed 1.30V.

You're doing the right things in the right sequence to obtain more information for resolving this issue.

 

[deustroop]

We are looking at 2 pair of sticks that apparently work perfectly at their default speed 2666 but cannot, at this time, be overclocked.
If one stick fails that will disable all ?
Whats the RMA period for the ram ?
As I said , I run the ASUS Z390-A with cpu 9900k and 2x16GB memory @ 3.2GHz (3200) on standard memory rated for that speed. I set that speed manually ignoring XMP.

 

[BonzaiDuck]

We are looking at 2 pair of sticks that apparently work perfectly at their default speed 2666 but cannot, at this time, be overclocked.
If one stick fails that will disable all ?
Whats the RMA period for the ram ?
As I said , I run the ASUS Z390-A with cpu 9900k and 2x16GB memory @ 3.2GHz (3200) on standard memory rated for that speed. I set that speed manually ignoring XMP.

What about the full set of latency timings? Have you verified that they're the same?

I've had new RAM arrive defective, but maybe only once in my experience. I've had other RAM just get old and die -- Crucial Ballistix.

And, giving my nod to your change of direction on the OP's problem, it might just be a bad stick and an RMA exercise. Pain in the butt, but it happens . . .

On the warranty period, I said I used G.SKILL. I think they have a lifetime warranty. Sound incredible, and I could check again, but I think it's the case.

 

[sze5003]

We are looking at 2 pair of sticks that apparently work perfectly at their default speed 2666 but cannot, at this time, be overclocked.
If one stick fails that will disable all ?
Whats the RMA period for the ram ?
As I said , I run the ASUS Z390-A with cpu 9900k and 2x16GB memory @ 3.2GHz (3200) on standard memory rated for that speed. I set that speed manually ignoring XMP.

In my case it's 4 sticks of 8gb of the same corsair ram. Which ran fine at their advertised speed of 3000 with the 8700k just by enabling the XMP profile and not changing anything else for 3 + years.

I wouldn't be able to return it now. Just need to figure out what settings to change on my board to attempt to run them at 3000 or I may just leave them at 2666. Default was 2133 when I initially installed them.

 

[BonzaiDuck]

Find out the bug-check code if you can.

I don't know what you did to them during the 3 years with the older processor. I DO know that two pairs of RAM need a slight VCCIO/SA boost over using one pair. With four slots filled, there's a bit more stress on the IMC. That's why I let my spend-thrifty impulse drive me to replace 4 DDR4-3200's with two. I wasn't having any problem with four, though. But I had to replace the motherboard -- another story -- and just chose to do it at that time.

 

[sze5003]

Find out the bug-check code if you can.

I don't know what you did to them during the 3 years with the older processor. I DO know that two pairs of RAM need a slight VCCIO/SA boost over using one pair. With four slots filled, there's a bit more stress on the IMC. That's why I let my spend-thrifty impulse drive me to replace 4 DDR4-3200's with two. I wasn't having any problem with four, though. But I had to replace the motherboard -- another story -- and just chose to do it at that time.

Didn't do much with these ram sticks other than run them at 3000mhz using the XMP profile with the 8700k. Some coding and gaming but mainly browsing and lots of gaming. The only difference was the 9900k which I bought used (1 year of use according to the user).

I'm getting home tomorrow so I'll look through event viewer and blue screen viewer to look for that bug code.

I read about your vape pen incident, sounded horrible and glad you were able to recover. I used to smoke, then vaped for a really long time, but completely quit a few years ago after news went out about people getting sick.

I never had any issues and I think it turned out I just had low vitamin d, got tired after working out every day because of that, made me feel like it was due to the vape. On a long flight I didn't have nicorette left so I just manually quit like that. I do miss the vape flavors though

I would buy new ram but my goal is to do a full upgrade next year with intel 12th gen and ddr5 since 11th gen was pointless..so I just scratched the upgrade itch with a new case and 9900k instead.

 

[bononos]

I had a quick peek at the BIOS settings just now and didn't find VCCIO/VCCSA voltage settings, maybe they are in the more advanced settings deeper in the submenus. AFAIK the DRAM voltage is not the same VCCIO/VCCSA is it? The default XMP setting is 1.3V on my sticks.

 

[sze5003]

These are the two bug check codes I found from by BSOD on 5/1. Attached the pics for reference.
1. 0x0000000a
2. 0x0000001a

Both memory related. Right now all 4 corsair sticks are running at 2666 frequency with no other changes in Bios other than the manual frequency change I set after I was testing different speeds.

My board also does not have VCCIA/VCCSA voltage options labeled easily as such. There is a DRAM config menu that hasa bout 70 items in there.

 

[deustroop]

Can you discover whether there is only one bad stick ? Pull it and the run XMP on the three remaining.If that is stable, at least you have an option.

 

[BonzaiDuck]

I had a quick peek at the BIOS settings just now and didn't find VCCIO/VCCSA voltage settings, maybe they are in the more advanced settings deeper in the submenus. AFAIK the DRAM voltage is not the same VCCIO/VCCSA is it? The default XMP setting is 1.3V on my sticks.

What brand motherboard do you have? On ASUS boards, there is an "Advanced" page for setting the CPU core speeds, choosing XMP, the RAM speed with submenu for timings, then the voltages. Wherever you find the item for setting CPU VCORE, you should find the VCCIO and VCCSA voltages, along with the RAM voltage.

But you're right -- the RAM voltage is a separate item, and if the computer defaults the VCCIO to >= 1.3V for Auto settings under XMP, you could confuse it with the RAM voltage. As I said, find out what voltage is given at the Intel memory spec setting under Auto. That should inform you of the voltage's magnitude for the OC spec speed setting. Otherwise, the IMC and SA are closer to being overvolted with some of these motherboards.

 

[BonzaiDuck]

Can you discover whether there is only one bad stick ? Pull it and the run XMP on the three remaining.If that is stable, at least you have an option.

That's the next tedious step. To save time and trouble, it should have a systematic approach. I wouldn't know if removing one at a time and testing three at a time is more efficient than just testing one at a time. You could also split the kits of two -- the OP said he bought two different kits of the same Corsair make/model?

Over years, I"ve interacted with customer support for at least a couple mem-makers via e-mail or "support ticket". When asked about doubling RAM by adding an identical kit, the support-techs sort of back away from any guarantee that the two kits combined will run at spec. They point to the fact that they sell kits of four sticks which are binned and/or tested. With G.SKILL regarding my TridentZ "GTZ" 3200 14-14-14 kits, they told me "maybe -- probably -- try it -- otherwise, consider running them at 3000". Of course, they worked fine and I put them through about 5 cycles or 500%-coverage with HCI-Memtest64.

Of course, the OP had used the two kits for some three years without problem. If they were handled properly, anyone's guess is as good as mine (or better) for what might have happened to them. I can't remember what was said here -- did he test them again before removing them from the 8700K system? Did he disconnect and drain the PSU before removing the sticks from the old board?

Main thing -- find out which modules are defective, remove them from the equation, complete an RMA request and move forward. There shouldn't be a need to buy new RAM, provided the company offers RMA support and backs up their warranty. The only caveat arises as to whether they currently continue to produce that RAM, or whether they have an inventory. You don't want to break the match of two kits, RMA one of them, and then find they sent you "new equivalent" RAM which isn't spec'd precisely the same as the one you kept.

Anyone else? Not much more I can add, given the OP's assessment of his bug-check codes.

 

[sze5003]

What I'll do is split the 4 sticks in 2 kits. The first 2 kits were black and the second kit I bought is white.

I swapped cases at the beginning of the month and I did drain the PSU and made sure everything was off and had no power before I moved everything to the new case and did a much nicer job with cable management.

I'll run the first 2 kits at xmp Memory only profile and boot and see what happens. While I'm gonna open the case I'll take out the cmos batter too just in case.

Is there anything I can run once I'm booted to windows to see if the xmp profile causes a bsod or crashes?

When this happened last time I wasn't doing much with the pc it was pretty quick after I had booted.

 

[BonzaiDuck]

What I'll do is split the 4 sticks in 2 kits. The first 2 kits were black and the second kit I bought is white.

I swapped cases at the beginning of the month and I did drain the PSU and made sure everything was off and had no power before I moved everything to the new case and did a much nicer job with cable management.

I'll run the first 2 kits at xmp Memory only profile and boot and see what happens. While I'm gonna open the case I'll take out the cmos batter too just in case.

Is there anything I can run once I'm booted to windows to see if the xmp profile causes a bsod or crashes?

When this happened last time I wasn't doing much with the pc it was pretty quick after I had booted.

Different color? Or different model? You apparently know what you're doing. Mixing two different models, however, is a no-no.

If it were me, I would still tweak the IMC and SA voltages, but if they show as below 1.35V as monitored by software or in BIOS, they won't fail unless they're defective.

I'm still mystified as to why others can't find these voltages in their BIOS. Someone could enlighten me as to how that is . . . I personally always choose an ASUS motherboard model that is between mid-level and high-end, so I get the features I want but I don't pay $400 for a motherboard. But I've had budget boards that still feature those voltages in BIOS, Gigabyte as well as ASUS, I think.

 

[sze5003]

Different color? Or different model? You apparently know what you're doing. Mixing two different models, however, is a no-no.

If it were me, I would still tweak the IMC and SA voltages, but if they show as below 1.35V as monitored by software or in BIOS, they won't fail unless they're defective.

I'm still mystified as to why others can't find these voltages in their BIOS. Someone could enlighten me as to how that is . . . I personally always choose an ASUS motherboard model that is between mid-level and high-end, so I get the features I want but I don't pay $400 for a motherboard. But I've had budget boards that still feature those voltages in BIOS, Gigabyte as well as ASUS, I think.

They are the same model kit just a different color. The first kit I got was a black color and later when I added the additional 2 sticks for 32gb I couldn't find them in the same color.

I'm pretty sure each board has ram voltage that can be modified. They are all under different named categories depending on the board.

I've also always had Asus boards. So I know for mine it's under advanced mode, AI tweaker.

Here is what my voltage is for the Ram with everything on auto and frequency set to 2666.