Question Total graphics lockup has returned after a week without problems ?

[Fastfishy2]

I made a post about this maybe a week back, but I was running into an issue where occasionally a 3Dmark stress test run would freeze up my computer. First the benchmark itself freezes, then I can try opening task manager but all the tasks are frozen on 0% usage (and this window soon becomes unusable too), and if I leave it too long I also lose the ability to sign out / shutdown using the windows interface and have to use a hard PC reset via the power button.

Since then I have done a fresh Win10 install to a different SSD, there are clean GPU and chipset drivers, my BIOS is on the latest revision, I have run multiple memory tests including Memtest86 and found no errors, and I have frequently run SFC Scannow via command prompt and found no OS corruptions. I have stress tested the GPU in various games and ran 3Dmark Port Royal & Speedway stress tests about 15 times over the course of a week with no issues until today. I have put the CPU through heavy workloads too, without issue.

After a week of no problems whatsoever this lockup happens again. It doesn't seem to leave any information on windows event viewer and it also doesn't ever cause a BSOD, so I am left with no information on the cause.

Specs:
R7 5800X3D (w/NH-D15s),
Strix B550-A,
32GB DDR4-3200 CL16,
Samsung 970 EVO+ NVME (Boot),
Asus DUAL RTX 4070 12GB,
Corsair RM750X (2021) PSU.

All temps are well within happy ranges and no manual overclocking has been done (DOCP is enabled for memory).

 

[Colif]

if pc locks up, and isn't causing a BSOD, its not a windows problem.

So we can discount the ssd since you on a new one.

tried running Prime95 on the CPU?

could disable DOCP since its the same as XMP and realistically an overclock of the ram. Memtest can't see if its a timing error. Is ram on Motherboard QVL?

 

[Fastfishy2]

if pc locks up, and isn't causing a BSOD, its not a windows problem.

So we can discount the ssd since you on a new one.

tried running Prime95 on the CPU?

could disable DOCP since its the same as XMP and realistically an overclock of the ram. Memtest can't see if its a timing error. Is ram on Motherboard QVL?

I did some digging. This particular G-skill kit is not in fact on the QVL for this board. I'm an idiot.

I have F4-3200C16D-32GVK and it has other related kits but not that one. D'oh.

I'm getting another kit of RAM tomorrow, this time a set of Trident Z 3200 cl16, as it was suggested on 2 other forums that the culprit may be a memory issue.

 

[Fastfishy2]

if pc locks up, and isn't causing a BSOD, its not a windows problem.

So we can discount the ssd since you on a new one.

tried running Prime95 on the CPU?

could disable DOCP since its the same as XMP and realistically an overclock of the ram. Memtest can't see if its a timing error. Is ram on Motherboard QVL?

OK so, I put a new kit of ram in today, same frequency and timings 3200 CL16-18-18-38, but Trident Z RGB memory this time, just to rule memory out. Since late morning into the evening I've had a number of heavy stress tests running, including multiple runs of Unigine Superposition 8k (with and without OCCT medium CPU stress test in the background), I ran OCCT GPU stress test as well, I had 3Dmark Speed Way running on loop for 3 hours, Port Royal for over an hour, as well as a complete stress test run of Time Spy extreme and a complete run of Speed Way stress tests. The freeze didn't happen once. That's not to say it won't happen again, but it's possible it went down with the previous kit of RAM I removed. If not, I'm very interested to know why it's so hard to reproduce with stress testing software if it's a hardware issue, so I'll have to touch base with the team at UL if it happens again, given that it has only been happening in 3Dmark's software, and rarely at best, since this issue happened months ago with The Last of Us and nothing else since.


In any case, barring some super weird type of hardware fault that I've never heard of before, I think I can rule out the GPU being unstable or compromised, at least phsyically. I just don't see how it could sit through multiple runs of Superposition 8k, and about 5-6 hours of cumulative back-to-back heavy benchmarking, without so much as a stutter. I'm getting stability scores of 98.6% and 99.9% in Speed Way and Port Royal respectively.

If anyone has any other ideas, though, I'd really like to hear them. I have a feeling this isn't ogre. I will also note that on one occasion, I was able to tell the PC to log out of my windows profile upon which I regained full functionality of the PC and simply logged back in as if nothing had happened.

 

[Colif]

The freeze didn't happen once.

have you crashed since replacing ram? if new ram on qvl, that could be all it was. Ryzen is more picky about ram, intel CPU more likely to work with untested ram.

How often did it crash before?

 

[Fastfishy2]

have you crashed since replacing ram? if new ram on qvl, that could be all it was. Ryzen is more picky about ram, intel CPU more likely to work with untested ram.

How often did it crash before?

Replaced RAM on thursday morning and ran heavy benchmarks for the rest of the day without a freeze, including having 3Dmark speed way running on loop for over 3 hours without freezing. Ran speed way stress test this evening and it locked the computer up again and didn't generate a log file, there's also nothing in event viewer that corresponds to it, except one: it has a critical error at 6:48PM about "unexpected shutdown", which is when I got back to the computer and noticed 3Dmark had frozen up (and while I could open programs on my second monitor, they would quickly lose functionality too, although I could open HWinfo64 and scroll up and down to look at the values it opened with - nothing odd), and had to hold down the power button. However, the dialog on the unexpected shutdown message noted a time of 5:58 PM - the computer wasn't actually shut down at that stage, and was still semi-responsive up until I forced it to shutdown at 6:48. The 3Dmark test was started at 5:53PM and froze some time after that.

In regards to QVL Ram, I wasn't actually able to find any in my country that exactly matched the QVL list for my board, so I got G-Skill Trident Z RGB at 3200-16-18-18-38, 32GB.

 

[Colif]

its a hardware problem of some kind. Windows has no idea why its crashing

its odd it keeps working after windows thinks its shutdown. that shouldn't work, the pc can't keep working without windows. Unless the time in windows is wrong?

I would run the bootable version of Prime95
Prime 95 Bootable - https://www.infopackets.com/news/10113/how-fix-bootable-prime95-stress-test-hardware

Prime 95 Instructions - https://appuals.com/how-to-run-a-cpu-stress-test-using-prime95/

Prime only real way to test AMD CPU. It also tests ram, and PSU (sort of). It would be unusual if its a timing error on 2 different ram configs. Prime may show that.

 

[Fastfishy2]

its a hardware problem of some kind. Windows has no idea why its crashing

its odd it keeps working after windows thinks its shutdown. that shouldn't work, the pc can't keep working without windows. Unless the time in windows is wrong?

I would run the bootable version of Prime95
Prime 95 Bootable - https://www.infopackets.com/news/10113/how-fix-bootable-prime95-stress-test-hardware

Prime 95 Instructions - https://appuals.com/how-to-run-a-cpu-stress-test-using-prime95/

My only hesitation to running Prime 95 is that, even with an NH-D15S and Arctic Silver ceramique 2, my 5800X3D reaches 90c steady state and drops clocks by about 100mhz but otherwise stays stable. I'm a bit hesitant to have it run at that temperature for an extended period of time.

 

[Colif]

fair enough I guess. Prime can run over 18 hours.

if ram still not on listing, was it tested by G SKill to work on board? Board makers aren't only ones who test the ram, https://www.gskill.com/configurator might help

What country are you from?

 

[Fastfishy2]

fair enough I guess. Prime can run over 18 hours.

if ram still not on listing, was it tested by G SKill to work on board? Board makers aren't only ones who test the ram, https://www.gskill.com/configurator might help

What country are you from?

I haven't seen that G-skill has tested it on this specific board although they have tested their RAM with the B550-F, which is a very similar sub variant.
I'm from New Zealand.

I'm thinking that maybe the RAM kits I've had simply arent stable at their particular timings because they're both G-skill 3200-16-18-18-38 kits but with different aesthetics. Or at least they're not stable at those timings on this board. I don't mind reducing the clock speed and loosening up the timings a bit to see if that helps. How would I go about setting it down to 3000mhz instead, and what kind of relatively safe timings could I put down. I'm not after peak RAM performance here, especially not with the 5800X3D.

 

[Colif]

Couldn't you order ram from Australia? We have a few shops over here. I am not sure which ones ship to NZ though.

I am not sure about setting up ram timings on your board. This might help - https://rog-forum.asus.com/t5/amd-5...ure-correct-timings-for-ddr4-rams/td-p/726722

 

[Fastfishy2]

Couldn't you order ram from Australia? We have a few shops over here. I am not sure which ones ship to NZ though.

I am not sure about setting up ram timings on your board. This might help - https://rog-forum.asus.com/t5/amd-5...ure-correct-timings-for-ddr4-rams/td-p/726722

it's a long way to ship some RAM. I've also asked around on other forums and been roundly told that, as long as it's within the general spec of the processor and motherboard, the QVL shouldn't matter and is extremely restrictive.

 

[Colif]

motherboard makers don't test everything but if G Skill also didn't show ram as compatible it becomes more difficult. They are the ones more likely to try.

See if the ROG link helps. Setting up ram manually isn't something I have done.

 

[Fastfishy2]

motherboard makers don't test everything but if G Skill also didn't show ram as compatible it becomes more difficult. They are the ones more likely to try.

See if the ROG link helps. Setting up ram manually isn't something I have done.

I'll have a look at that one.
I also ran OCCT's power stability test last night, and that ran the cpu and gpu concurrently at their absolute limit for half an hour. No errors, no freezing. I've also turned off global C state in bios.
I've run speedway stress test repeatedly this morning with no freezes so far. I'll let you know if that changes.