PC Restarting Under Load

[3x0]

I have two folders within LiveKernelReports - USBHUB3 and WATCHDOG but both have nothing inside

Are you using some system cleaner, CCleaner, Wise Disk Cleaner etc.? There should be .dmp files in there if the folders are there but some sys cleaner will delete .dmp files.

 

[RogueSquadron80]

I bought a new PSU and swapped out the GPU PCie cables and it still crashed.

i had this no end of times on x570 platform, tried, psu, ram, windows installs nothing worked in the end i swapped to z690 and 12700k and it never happened again, guy at work had similar issues as well, would work fine for a while then crash all the time

 

[Waldorf]

@_roman_
the RMx series (latest edition) is a pretty decent unit, single failure no proof of anything, i expected you to know that kind of thing.

@NameIsGreg
those memtests with 86 ending are useless for stability testing, they (+90% of the time) only detect defective sticks, as stated on the website.
the fact that you have CL18@3600 means you have at least decent dies (next time get at least some LPX, with the lowest CL you can get for reasonable money)

you want to test stock settings to rule out if its not coming from ram oc.
reset to stock/ turn off docp, disable fast boot, and test with hci in windows (untick the "low priority.." box, enter 32 for threads),
make sure to close all progs/stuff in the background, use ISLC to kill anything not needed, then run the test until +1600%.

if thats done without errors, you can look at manually setting clock/main timings/bus clocks and voltages for ram and soc.


HCI memtest
standby cleaner

ddr4 oc guide
occt cpu+ram testing

 

[NameIsGreg]

Are you using some system cleaner, CCleaner, Wise Disk Cleaner etc.? There should be .dmp files in there if the folders are there but some sys cleaner will delete .dmp files.

No not using any cleaner software. The only time I ever had and .dmp files was when I ran driver verifier - deleted the associated software. Then never got .dmp files again.

I did change some kernel settings once to try and get my PC to generate .dmp files on crash but I honestly cant remember what I did as it was over a year ago but it didn't change anything and have reset my PC since then so should be back to default settings.

 

[phanbuey]

i had this no end of times on x570 platform, tried, psu, ram, windows installs nothing worked in the end i swapped to z690 and 12700k and it never happened again, guy at work had similar issues as well, would work fine for a while then crash all the time

This happened with one my devboxes on the 470 -- also with corsair vengence sticks. No DOCP - bone stock. Was running as a dev database server and would just reeboot every 5-9 days with a cx05 error in event log.

Would pass stress tests all day long.

 

[NameIsGreg]

you want to test stock settings to rule out if its not coming from ram oc.
reset to stock/ turn off docp, disable fast boot, and test with hci in windows (untick the "low priority.." box, enter 32 for threads),
make sure to close all progs/stuff in the background, use ISLC to kill anything not needed, then run the test until +1600%.

if thats done without errors, you can look at manually setting clock/main timings/bus clocks and voltages for ram and soc.

Ok have set this up but am going to run it overnight as its going to take a lot of time to reach 1600% on every instance. I also dont have the option to untick low-priority and enter thread no. I'm assuming that's only available in the paid version?

 

[thesmokingman]

Nope, all are Corsair Vengeance Pro 8gb with a max speed of 3600MHz

edit: also in their correct pairs as well.

Corsair ram, snickers...

Check event viewer and look for whea errors. Better yet setup hwinfo 64 to show whea errors on your taskbar.

 

[NameIsGreg]

Corsair ram, snickers...

Check event viewer and look for whea errors. Better yet setup hwinfo 64 to show whea errors on your taskbar.

No WHEA errors in Windows > System then filtering by WHEA-Logger and no errors in hwinfo 64 either

 

[Princess Garnet]

I would be concerned with the fact that it was previously pulling a lot of wattage through a few specific wires, causing them to melt, to begin with...

Four memory sticks can be heavy, but you seem to imply you've been running them at stock JEDEC speeds, and since they are 8 GB DIMMs, they are probably only single rank. Lower frequency and less ranks should make the memory configuration easier to run. AM4 (at least with mid to late chipsets and CPUs) can often do four dual rank DIMMs at 3,600 MHz (I'm doing it now), but it's typically near the edge for it and officially its only rated for lower than that with that many memory sticks/ranks, so the first thing I would have suggested would be to scale that back.

Since you tried enabling profile speeds and it made it worse, it might suggest that you're still not stable at default speeds. IMCs can go bad, I guess. I agree with the other comments to move down to two RAM sticks and see if it improves.

This would help if you had another video card to test, especially give the prior issue with the cable melting. You said Furmark was stable but what about OCCT's "variable" graphics stress test? When I had Black screen restarts a year and a half ago in games/hardware accelerated applications, that was really the only synthetic/stress test that would cause it. Others (like Prime 95 plus Furmark, or the rest of OCCT's tests) were seemingly stable, even though the PC clearly wasn't.

Since it's an nVidia graphics card, have you tried going back to 566.36 drivers? The 2025 drivers aren't always stable for everyone, and sometimes they cause serious crashes like this. (If you already mentioned trying this and I missed it, I apologize.)

I'm a bit surprised there's no WHEA logs being made. If it's automatically restarting (and not BSODing), that can usually indicate it might be an MCE, and those tend to create such logs in there.

 

[Waldorf]

@NameIsGreg
yeah, seems they changed that.
for quick test if you still have docp on, use the occ ram test, if it shows errors, go to stock and test, occ passes, then hci.

@Princess Garnet
never had any issues with imc being picky on x950 (compared to x800/x600 chips).
if you mean jedec with "default", it will always do that on 4 sticks, even with DR, unless its a really crappy sample for board, but not a single asus i setup (AM4, for friends) did docp stable @3600 (4 SR/2 DR) for me.

 

[Koozwad]

From memory that RMx PSU cannot handle the transient spikes(if I remembered the name correctly) from the 30x0 series. I had the exact same issue, though I was using a 3090 with an RMx 650W(and 6700K CPU at the time). Upgraded to AX850 PSU(and later 10900K) and all problems disappeared. The 3090(STRIX) did end up failing but that was likely due to a design issue on ASUS' part. Do a websearch for '3080 transient spikes' and you'll see what I mean.

PS also make sure you are not daisy-chaining if possible.

 

[NameIsGreg]

give it a try - kahru ramtest and GpuMemTest - https://www.programming4beginners.com/gpumemtest -- just to rule these out.

volmgr - is volume manager - its usually just windows whining that the volume has data corruption from a crash.

It really sounds like the rig is stable at full load (prime, furmark) but not in the transient states - where there's either instability somewhere along the curve - or from something else - bad ram / incompatibility (Corsair Vengance strikes again) -- i've also seen bad SATA and NVMe devices do this - one time it was as simple as a loose sata cable.

GPU MemTest passed with no errors but karu ramtest is paid and there seem to be lots of free alternatives that @Waldorf has provided so will be giving them a go.

I've heard this about sata cables but replaced them with my new PSU, I haven't looked into NVME aside from testing drives with Samsung Magician

 

[sram]

If you get BSOD's, it maybe worth it to try WhoCrashed software. It helped me several times.

 

[phanbuey]

From memory that RMx PSU cannot handle the transient spikes(if I remembered the name correctly) from the 30x0 series. I had the exact same issue, though I was using a 3090 with an RMx 650W(and 6700K CPU at the time). Upgraded to AX850 PSU(and later 10900K) and all problems disappeared. The 3090(STRIX) did end up failing but that was likely due to a design issue on ASUS' part. Do a websearch for '3080 transient spikes' and you'll see what I mean.

PS also make sure you are not daisy-chaining if possible.

+1 - nothing in the dumps and just hard crashes like that means the kernel couldnt even get a chance to log the error. Usually power related.

even if your videcard went MIA from the system mid game that error would still get logged by the kernel -

CPU/RAM/MB completely failing/unstable or SSD/volume losing all functionality would cause a restart with no bugcheck - which is unlikely since those components pass tests. +1 to PSU.

 

[_roman_]

Four memory sticks can be heavy, but you seem to imply you've been running them at stock JEDEC speeds, and since they are 8 GB DIMMs

MSI X570 Tomahawk with a ryzen 5800x mainboard with 2x 16Gib crucial ballistix dram color a equiped. I gave that guy my other 2x16Gib crucial ballistix dram color b. That mainboard still runs with 4 x 16GiB dram with w10 with my overclock settings stable. When you know what you are doing some dram run with better timings as with the 3600 MT/S stock settings for DDR4. I applied only75% of my B550 gaming edge wifi settings for those DRAM. i build and clean those computers. I'm not responsible for the installed software for the x570 mainboard.

 

[Waldorf]

@Koozwad
iirc the rtx 20xx was worse, and i dont have trouble with the 750w, even if i put max load on gpu (binned, 2-2.2ghz constant boost).
the pcb has diagnostic leds, that would show if there is a problem power wise (that i did have with 2 different 850 units from other brands; one power led getting triggered during bench/gaming).

 

[thesmokingman]

No WHEA errors in Windows > System then filtering by WHEA-Logger and no errors in hwinfo 64 either

It's really strange that you get no whea errors. Something doesn't compute? Any instability and it'll get flagged for the builds I've made.

 

[phanbuey]

It's really strange that you get no whea errors. Something doesn't compute? Any instability and it'll get flagged for the builds I've made.

unless the psu trips when everything is working fine.

 

[NameIsGreg]

@NameIsGreg
yeah, seems they changed that.
for quick test if you still have docp on, use the occ ram test, if it shows errors, go to stock and test, occ passes, then hci.

So I turned on DOCP and couldn't even get to my desktop, had two BSOD's back-to-back then took out two RAM sticks and reconfigured them in the 1 and 3 slot. Then ran the OCCT memory tester and had no errors. Will do HCI tonight and see what happens, should I use two stick or four?

It's really strange that you get no whea errors. Something doesn't compute? Any instability and it'll get flagged for the builds I've made.

Yeah it's very strange, never had any hardware errors or .dmp files which is is one of the many reasons its been hard to diagnose this issue.

This would help if you had another video card to test, especially give the prior issue with the cable melting. You said Furmark was stable but what about OCCT's "variable" graphics stress test? When I had Black screen restarts a year and a half ago in games/hardware accelerated applications, that was really the only synthetic/stress test that would cause it. Others (like Prime 95 plus Furmark, or the rest of OCCT's tests) were seemingly stable, even though the PC clearly wasn't.

Since it's an nVidia graphics card, have you tried going back to 566.36 drivers? The 2025 drivers aren't always stable for everyone, and sometimes they cause serious crashes like this. (If you already mentioned trying this and I missed it, I apologize.)

I'm a bit surprised there's no WHEA logs being made. If it's automatically restarting (and not BSODing), that can usually indicate it might be an MCE, and those tend to create such logs in there.

Sadly don't have another card to test with but ran occt's variable graphics test for 1hr and reported no errors and the system was stable whilst running.

For drivers, the issue has been going on for so long (well over a year) that i've installed plenty of nvidia drivers over the time and nothing changed, also done DDU which also didnt fix anything.

 

[Waldorf]

@NameIsGreg
test 4 without docp to like 800%, should be enough to make sure its not the sticks.
if something is gross off, it usually tends to show errors within 5-20%, small timing/voltage stuff by 100%, testing to 800% should cover each stick twice.
without knowing if stock is stable, no sense in tweaking anything.
once we know its stable at stock, easier to confirm oc is as well, and dial stuff in.

 

[phanbuey]

run occt CPU and cyberpunk or a graphically intense (RT ON) game at the same time and see if the power supply takes a dump.

or furmark = pass.

then OCCT = pass,

then OCCT + Furmark = if power cycle then you know it's psu, since the other two passed individually.

 

[NameIsGreg]

@NameIsGreg
test 4 without docp to like 800%, should be enough to make sure its not the sticks.
if something is gross off, it usually tends to show errors within 5-20%, small timing/voltage stuff by 100%, testing to 800% should cover each stick twice.
without knowing if stock is stable, no sense in tweaking anything.
once we know its stable at stock, easier to confirm oc is as well, and dial stuff in.

So I reported no errors on 800%+ coverage, realised this morning that I should of made the testing size smaller and have more instances to speed it up but oh well.

Will do those gpu tests later and will update @phanbuey on the results, little confused of the wording on your post though, I get the first sentence but just confused about the wording beyond that?

 

[Contra]

Also try TM-5 with anta absolute config or equivalent

 

[3x0]

Also try TM-5 with anta absolute config or equivalent

GitHub - CoolCmd/TestMem5: TestMem5 - PC RAM stress test

TestMem5 - PC RAM stress test. Contribute to CoolCmd/TestMem5 development by creating an account on GitHub.

github.com

 

[phanbuey]

So I reported no errors on 800%+ coverage, realised this morning that I should of made the testing size smaller and have more instances to speed it up but oh well.

Will do those gpu tests later and will update @phanbuey on the results, little confused of the wording on your post though, I get the first sentence but just confused about the wording beyond that?

View attachment 405621

Basically test the CPU by running a Prime 95 or similar -- if that passes great.

Then run a GPU stress by running furmark -- if that passes, also great.

Then run them both together, at the same time, even start and stop the furmark test a few times with the CPU test running - this will pull max wattage from your PSU, and it will power cycle if the Power Supply cant handle transients.

The reason for this is your error mode (no log, no dump) is as if someone just ripped the power cord from the computer - that's really the only explanation for your hardware testing fine individually but the rig rebooting with no minidump or whea error or any other faults logged. It really might be as simple as the PSU overvoltage protection tripping.