Question Intermittent lock up with DRAM LED lit up ?

[David_676]

Computer Type: Desktop
GPU: EVGA RTX 3080
CPU: Ryzen 9 5950x 16 core 32 Thread
Motherboard: Asus x570 Crosshair VIII Hero (Wi-Fi)
BIOS Version: 4501 04/19/2023
RAM: G.Skill Trident Z Royal 3600 CL14 F4-3600C14D-32GTRG
PSU: Be quiet! Straight Power 11 Platinum 1000W BN644 Fully Modular
Case: Lian Li 011 Dynamic XL
Operating System & Version: WINDOWS 11 Pro
GPU Drivers: GEFORCE GAME READY DRIVER - WHQL Driver Version: 536.67

Description of Original Problem: I built the computer almost 3 years ago and it ran fine for 1.5 years and I have been slowly troubleshooting it ever since. It happens once every 2 or 3 days, sometimes sooner. The computer locks up, screen freezes but stays on, doesn't respond to power button and the PSU switch is the only way to reset it. The debug code sometimes stays AA and sometimes it changes to 00. The yellow DRAM led sometimes lights and sometimes doesn't, when it does it lights slowly from off to a weak light, almost like it has limited power, like a candle in the wind, flickering.

Troubleshooting: I have updated the BIOS, reset BIOS, reinstalled OS and all drivers, RMA'd everything but the CPU. Downclocking the RAM helps a bit but everything I researched says that it should work fine at the 3600 I paid for. Wondering if anyone has seen this before.

As said, I replaced the MB and the PSU so the flickering DRAM light is very confusing, could this be a CPU problem? CPU was my last guess, but unless there is an AMD specific thing I don't know about, it seems this is the last option.

EDIT: I posted the wrong MB, the correct board is the Crosshair VIII not the VII.

 

[David_676]

2 different voltages. One doesn't apply to the other. Ram voltage is only used by the ram, MC voltage by the MC. It's like you going for a walk, you use a certain amount of energy to move your arms, pump your legs, if you run, you require more energy use, but that has zero affect on the amount of energy required by the monitors in your cellphone to keep track of heartbeats, gps, speed etc. The ram requires a certain amount of voltage to open doors, sort, shunt and whatever else it does, when you apply a higher speed, it often requires more voltage to do so, and remain stable.

That's what I thought but after looking into it a bit, it sounds like the BIOS will auto adjust VCCIO and such based on your RAM settings, I really don't know anything about it so maybe I read wrong, I may post in Overclocking to see if I can get more info on it. But my IMC did fail, maybe it was just a bad bin but I feel like there is something more to it.

 

[Misgar]

I was under the impression the stock JEDEC voltage was 1.20V and the "normal" voltage automatically applied by XMP was 1.35V.

Anything above 1.35V has to be applied manually and the higher you go, the greater the chance of permanent damage. Mind you, I could be wrong.

I'll have to check if increased RAM voltage can negatively affect the IMC.

 

[David_676]

I was under the impression the stock JEDEC voltage was 1.20V and the "normal" voltage automatically applied by XMP was 1.35V.

Anything above 1.35V has to be applied manually and the higher you go, the greater the chance of permanent damage. Mind you, I could be wrong.

I'll have to check if increased RAM voltage can negatively affect the IMC.

The voltage applied by the xmp on my ram was 1.45. I believe that is overkill tho, I think they just go that high because it's high end ram and it can handle it easily, but yes the current question is if that can affect the IMC, which some say it can and others say no, so I don't really know what to go with. The ones that say it can hurt the ram say it does so through the VCCIO voltage which is separate from the RAM but i think the BIOS auto sets that voltage higher if the ram voltage is higher, but I don't know for sure. I made a post about it in Overclocking but haven't got a response yet, hopefully someone will know for sure what the situation is. I will probably just lower the RAM voltage and set it just above the instability point, just to be safe.

 

[Misgar]

I found some old forum postings about DDR3, saying you shouldn't allow the XMP voltage to exceed 0.5V above the CPU Integrated Memory Controller VCCIO voltage. Hence, if VCCIO was set to 1.2V (for a DDR3 system) you shouldn't set your RAM voltage higher than 1.7V (1.2 + 0.5 = 1.7V). Presumably the same holds true for DDR4 and DDR5, albeit at their own respective (reduced) voltage levels.

There's an interesting anticle here:
https://www.pcworld.com/article/394...g-are-ok-even-if-they-void-your-warranty.html

If you buy a Corsair Vengeance of G Skill DDR4/4800 kit for instance, you can achieve those memory modules’ high data rates only by enabling the system’s XMP or eXtreme Memory Profile, which loads settings stored on the memory.

Those settings typically require increasing the voltage beyond the spec’d 1.2 volts. In the Corsair kit, you’d be pushing it to 1.35 volts. For the even higher clock rates of the G.Skill, it’s 1.5 volts. Because the memory controller lives inside the CPU today, that means increasing voltage to the CPU as well, which is where the companies get touchy about the warranty.

That last sentence is important.

This seems to confirm what I've read elsewhere anout VCCIO being linked to XMP voltages. When the BIOS raises the RAM supply voltage, to 1.35V (or 1.4, 1.45, 1,5V) for DDR4, it also increases the VCCIO in an attempt to maintain a "safe" margin, to protect the CPU's internal IMC. I'm wondering if too great a voltage differential forward-biasses the protection diodes in the I/O ports (this is going back to my 4000 series CMOS design days).

Since neither Intel nor AMD are 100% happy about replacing heavily overclocked (abused) CPUs, their warranties are null and void when you perform any overclocking, e.g. XMP or PBO. Of course they design their CPUs with overclocking in mind to get the best benchmark numbers in reviews and they know large numbers of people will boost their systems from day one, but it gives them grounds for refusing to replace a frazzled CPU.

When you look at most server hardware, they run CPU and memory at more conservative (stock) speeds, for improved reliability. Overclocking is rare in servers, but not unknown. Professionals don't need to overclock, they just ask the finance department for (a lot) more money and buy more expensive faster systems.

Whilst I enjoy overclocking many of my systems, I don't go past a certain point because I'd like my systems to still work in 5 years time.

in my 7950X system, I'm using the DDR5/4800 (non XMP) setting in my Corsair Vengeance RAM (2 x 32GB = 64GB) with no PBO. The CPU runs at roughly 80C under an NH-D15 for up to 10 hours during video rendering sessions. The GPU sits at 90C with the power level reduced to 95% in MSI Afterburner.

The Puget Systems benchmark below shows there is little to be gained on a AMD system (blue bars) in Adobe Premiere Pro by overclocking DDR5 RAM. Intel systems fare a little better. If system stability reduces as you push XMP overclocks higher and higher, a point is reached where a system crash 9 hours into a 10 hour video render is just not acceptable, especially if it means you have to start all over again.

If you are overclocking your RAM on a low to mid-range gaming system, to increase playable FPS above some minimum acceptable rate, fair enough. You can accept some CPU/RAM degradation and potential early failure until you can afford a faster system. But, if all you are trying to do is increase your FPS rate over 200, is it worth killing your system for an invisible improvement? Only you can say.

 

[David_676]

I found some old forum postings about DDR3, saying you shouldn't allow the XMP voltage to exceed 0.5V above the CPU Integrated Memory Controller VCCIO voltage. Hence, if VCCIO was set to 1.2V (for a DDR3 system) you shouldn't set your RAM voltage higher than 1.7V (1.2 + 0.5 = 1.7V). Presumably the same holds true for DDR4 and DDR5, albeit at their own respective (reduced) voltage levels.

There's an interesting anticle here:
https://www.pcworld.com/article/394...g-are-ok-even-if-they-void-your-warranty.html

If you buy a Corsair Vengeance of G Skill DDR4/4800 kit for instance, you can achieve those memory modules’ high data rates only by enabling the system’s XMP or eXtreme Memory Profile, which loads settings stored on the memory.

Those settings typically require increasing the voltage beyond the spec’d 1.2 volts. In the Corsair kit, you’d be pushing it to 1.35 volts. For the even higher clock rates of the G.Skill, it’s 1.5 volts. Because the memory controller lives inside the CPU today, that means increasing voltage to the CPU as well, which is where the companies get touchy about the warranty.

That last sentence is important.

This seems to confirm what I've read elsewhere anout VCCIO being linked to XMP voltages. When the BIOS raises the RAM supply voltage, to 1.35V (or 1.4, 1.45, 1,5V) for DDR4, it also increases the VCCIO in an attempt to maintain a "safe" margin, to protect the CPU's internal IMC. I'm wondering if too great a voltage differential forward-biasses the protection diodes in the I/O ports (this is going back to my 4000 series CMOS design days).

Since neither Intel nor AMD are 100% happy about replacing heavily overclocked (abused) CPUs, their warranties are null and void when you perform any overclocking, e.g. XMP or PBO. Of course they design their CPUs with overclocking in mind to get the best benchmark numbers in reviews and they know large numbers of people will boost their systems from day one, but it gives them grounds for refusing to replace a frazzled CPU.

When you look at most server hardware, they run CPU and memory at more conservative (stock) speeds, for improved reliability. Overclocking is rare in servers, but not unknown. Professionals don't need to overclock, they just ask the finance department for (a lot) more money and buy more expensive faster systems.

Whilst I enjoy overclocking many of my systems, I don't go past a certain point because I'd like my systems to still work in 5 years time.

in my 7950X system, I'm using the DDR5/4800 (non XMP) setting in my Corsair Vengeance RAM (2 x 32GB = 64GB) with no PBO. The CPU runs at roughly 80C under an NH-D15 for up to 10 hours during video rendering sessions. The GPU sits at 90C with the power level reduced to 95% in MSI Afterburner.

The Puget Systems benchmark below shows there is little to be gained on a AMD system (blue bars) in Adobe Premiere Pro by overclocking DDR5 RAM. Intel systems fare a little better. If system stability reduces as you push XMP overclocks higher and higher, a point is reached where a system crash 9 hours into a 10 hour video render is just not acceptable, especially if it means you have to start all over again.

If you are overclocking your RAM on a low to mid-range gaming system, to increase playable FPS above some minimum acceptable rate, fair enough. You can accept some CPU/RAM degradation and potential early failure until you can afford a faster system. But, if all you are trying to do is increase your FPS rate over 200, is it worth killing your system for an invisible improvement? Only you can say.

Yeah, I think you are right, I found a video of an ASUS board that sets VCCIO to half DRAM voltage on Ryzen and DDR4 so at very least that is happening, Although it seems that 0.725v for VCCIO shouldn't be high enough to kill the IMC.

View: https://youtu.be/b4ss2eeJcTc

VTTDDR I believe is VCCIO on this board from what I have read.

I may just drop it back to 3200 but 3600 is what everyone says is the best speed to run at for this chip as far as fclk and reliability and I would assume lifespan, I just think G.skill went overkill on the xmp voltage, so if I can get 3600 running at 1.38 or maybe even lower then I would think that would be within the lifespan range.

I appreciate you looking into this for me, I know I'm kinda being a pain trying to get higher speeds, and in the end I may just lower it, I just hoping I can find a solution so I don't feel like I waisted more money than I should have given that I already went overkill on everything and could have done this much cheaper.

 

[David_676]

After a ton of searching and very unclear answers, I think I have found the answer, I stumbled across this OC Guide that says this:

• One common limiting factor for the maximum safe voltage you can operate is your CPU's architecture. According to JEDEC, VDDQ, the voltage of data output, is tied to VDD, colloquially referred to as VDIMM or DRAM Voltage. This voltage interacts with the PHY or Physical Layer present on the CPU and may lead to long-term degradation of the IMC if set too high. As a result, daily use of VDIMM voltages above 1.60 V on Ryzen 3000 and 5000 and 1.65 V on Intel Consumer Lake-series Processors is not advisable as CPU degradation of the PHY is difficult to measure or notice until the issue becomes serious.

So it seems that the IMC is affected by the DRAM voltage, but the recommended daily driver maximum was 1.65v and I was running at the XMP of 1.45v. So if it was the OC that caused my problems, it had to have been a bad bin for it to degrade that fast. I think I am going to just run XMP and hope that this replacement CPU is a better bin and doesn't degrade early again. I may also set a small undervolt in CO just for the sake of temps and avoid PBO and Auto OC.

 

[Misgar]

Just remember the longer you leave RAM running at a really high voltage (1.50V and above) the shorter the life expectancy of your system. I instantly killed an old AMD CPU by typing in a 0.2V increase on the core voltage by mistake, instead of 0.02V. You may already have caused slight damage if you've been too enthusiastic.

 

[David_676]

Just remember the longer you leave RAM running at a really high voltage (1.50V and above) the shorter the life expectancy of your system. I instantly killed an old AMD CPU by typing in a 0.2V increase on the core voltage by mistake, instead of 0.02V. You may already have caused slight damage if you've been too enthusiastic.

I'm not gonna go over 1.45v DRAM, I'm not touching any other voltages. But even that I am going to try to keep it as low as possible, 1.375v wouldn't post, 1.38 did but just to help stability, I have it set to 1.39. The guide quote I posted said anything under 1.65v was safe for daily usage so even 1.45v should be fine in regards to the IMC, and the RAM is B-die, so that can easily do 1.45v as well. I am fairly confident in my assumptions at the moment. We were right about the IMC being affected by RAM voltage but everything I have found says that 1.45v is safe for B-die and the AMD IMC. I am still gonna run it as low as possible while keeping the XMP stable tho, I just need to give it some time to see if it's truly stable, it is passing the stress tests I have been throwing at it, so now it's just a matter of time to see if I get any app crashes or failed updates.

I appreciate the assistance more than you can imagine, I have been dealing with this for a long time and I don't know anyone personally who knows anything about this that can help me reason through this.

 

[Karadjgne]

Intel extreme xmp2, as tested by Intel, says 1.5v max for DDR4. That's not a recommendation, it's a hard limit regardless of ram die. Intel determined that DDR4 voltages above that increase chances of deterioration or death to unacceptable levels.

As I said, it's 2 different voltages, one does not affect the other, dram voltage is seperate from imc voltages, there's no connections. However, running ram at higher speeds can require higher voltages in the imc to compensate for the speed increase.

So you could run 2400 at 1.25v or 1.5v and it wouldn't affect the imc, vccio etc but bumping that to xmp rated speeds like 3600 often requires bumping the dram to 1.35v, but also may require bumping vccio, vccsa etc. Not because of the voltage increase, but because of the speed increase.

There's no set policy for OC. Each cpu is different, each set of ram is different, what one setup gets may be similar or even the same as others, but there's no guarantee.

VCCIO is the operating voltage of the imc, dram voltage is the operating voltage of the ram, VCCSA is the operating voltage of the communication between the imc and ram. They'll have different names in different bios, different brands, but that's the essence. So far you've only mentioned dram and imc, nothing about what's between. I'll bet that's where you are losing your stability. It all has to balance, can't just bump the ends.

 

[David_676]

Intel extreme xmp2, as tested by Intel, says 1.5v max for DDR4. That's not a recommendation, it's a hard limit regardless of ram die. Intel determined that DDR4 voltages above that increase chances of deterioration or death to unacceptable levels.

As I said, it's 2 different voltages, one does not affect the other, dram voltage is seperate from imc voltages, there's no connections. However, running ram at higher speeds can require higher voltages in the imc to compensate for the speed increase.

So you could run 2400 at 1.25v or 1.5v and it wouldn't affect the imc, vccio etc but bumping that to xmp rated speeds like 3600 often requires bumping the dram to 1.35v, but also may require bumping vccio, vccsa etc. Not because of the voltage increase, but because of the speed increase.

There's no set policy for OC. Each cpu is different, each set of ram is different, what one setup gets may be similar or even the same as others, but there's no guarantee.

VCCIO is the operating voltage of the imc, dram voltage is the operating voltage of the ram, VCCSA is the operating voltage of the communication between the imc and ram. They'll have different names in different bios, different brands, but that's the essence. So far you've only mentioned dram and imc, nothing about what's between. I'll bet that's where you are losing your stability. It all has to balance, can't just bump the ends.

Everything else is set to auto, which is what I meant by the voltage "connection". I know they aren't physically linked and they are all different, but raising DRAM only and letting ASUS work out the rest is what I was worried about. It seems that ASUS requires VCCIO to be at least half of VDRAM. But as I said, the 1.45v I had it set at shouldn't have been a problem as even you said above 1.5v increases the chances of deterioration. I'm not saying that nothing could happen below 1.5v but that limit by intel and many other things I have read all say that 1.45 should be fine. Especially with my B-die RAM.

The instability I was having was a bad IMC, AMD RMA'd the chip and the new one has no problems at all at XMP. And again, I never touched the IMC voltages. So the IMC went bad for some reason and I just want to make sure that the 1.45v DRAM wasn't the reason, and like I said everything I have found says that 1.45v shouldn't cause that.

If you have a source that says 1.45v will cause degradation of the IMC then I would absolutely take that into account, I just haven't seen anything like that yet. It's all 1.5v, 1.65v and higher that they don't recommend for daily driver.