So when CPUs overheat and shut down...

[tRidiot]

Well, the unprovoked shutdowns the last couple of days were just random reboots. No crash screen or BSOD, normal behavior on reboot.

However when trying to OC and start a stress program, it's a hard crash to black with needing to pull the power cable to get it to reboot. Most times would require F2 to get into BIOS and everything would reset.

 

[tRidiot]

Maybe I'm missing something here, but the irregularity of it all, I would test out memory first. Could be something simple where XMP doesn't like your system. 3600C14 is not always easy to run. Since you have AIDA64 why don't you set to bios defaults (leave RAM at slow / stock speed) and run the memory/cache test for a few hours. You could even test drive it for a few days (the loss of performance won't kill you for a day or two). If all good, then go ahead and enable XMP and do the same.

My system just shuts off like that too if I have a bad mem OC. Mainly if things are just too tight.. or too fast I suppose.. I have to remind myself occasionally that this is not an Intel and it does not behave like one.. Is your ram on your QVL? I would assume it is being G.Skill.. for both my pairs of Tridents I checked the QVL on both sites and it was a go, and I haven't had any problems really other than what I have caused myself. But at first I thought maybe you cooked a CCD or X or whatever by not having TIM and 1.35v..

Ok, so I just got home and it looks like Aida64 has run for a number of hours without a crash. This says 7h elapsed time, but I don't know about that... it should show nearly 9h. My time changed on the computer, for some reason it never wants to stay in my time zone, so it's showing 2h behind what time it is, which would account for the lost time, but I would think Aida would just count elapsed time, not go by timestamp on start and on my current Windows time (which is wrong). Computer had gone into hibernate mode, but it looks like it was still running? Not sure if that really counts, lol. Was it truly still running and testing the RAM all this time? I forgot to shut off the hibernate thing.

I'll keep it up a few more hours, but that is something else. Makes me kinda pissed, I paid about $350 for this memory.

- - - Auto-Merged Double Post - - -

I do not think it was running all this time... when I started it up, the memory showed 31 and 32 degrees, now in 10 minutes it's up to 37 and 38 degrees. So leaving that test running didn't do me any good because I forgot to change the automatic sleep thing. So I have disabled that and will leave it running now for a few hours.

 

[Zerileous]

It's a delicate balance, as, freeagent said, it can be tricky to OC mem on AMD. My b-die kit will run 3600 C14 but at 1.5v (they are rated for 3200 C14 at 1.35v). You could probably adjust the timings a bit and achieve a stable OC at 3600 C14. There are also some Zen2 (3000 series) CPUs that struggle with 3600MHz, although most can do at least that. Is there a reason you chose this memory? In your signature it looks like you replaced a kit of 3600C16 memory with the C14 stuff, is that correct?

The difference in real world performance is going to be pretty negligible going from 3200 to 3600 (approx 1% FPS) so yes $350 is a lot to pay for the kit. If you're specifically benchmarking and OCing memory it can make a difference, but otherwise it's not much. For most uses, the value is not there to spend so much on memory. So I would definitely say returning it and using something else or even buying a less expensive new kit is an option if you don't feel like tweaking this kit. If you do feel like tweaking (since this is OCForums after all), we can help with that.

The thing with the system clock changing sticks out to me as well though and does also lead me to suspect the mobo. Is Windows set to "set time automatically"?

 

[tRidiot]

Yeah I went back and changed the time, it should hold now. I just forgot it after my Windows reinstall this week.

So an hour and a half in, it seems to be holding well. I think I'll change it back to 3600 and see if it'll stay up.

I did specifically get this kit because it is Samsung B-die and I wanted to get faster speeds with tighter timings than other out of the box kits. My other 3600 kit was CL16-19-19-39 I think, so I had hoped this CL14-15-15-35 kit would help me a tiny bit in benchmarking. It is a price to pay, for sure. If I can't run those timings and still OC my CPU and such like I want, then they're going back. It's just not worth giving up the other performance and stability.


So we'll see how well Aida64 will run this now at 3600 CL14. Honestly I'm still kinda hoping this is my PSU more than anything, but since it's holding at the moment, that seems less likely.

 

[freeagent]

Awesome man, glad to hear it

You should be able to do ok with your new sticks

 

[tRidiot]

Well, I'm more confused than ever now.

So I ran Aida64 on memory stress last night at the faster speeds for about 3 hours, no problems, so I went ahead and clicked the CPU/FPU boxes too. I noticed the CPU temps immediately went up about 7-8 degrees. So in spite of it showing 100% CPU load when testing memory, adding in the CPU stability test did increase the workload.

So I left it that way.

All night.

12 hours total.

No crashes, no shutdowns, no nuthin'.

So I close the test, reboot the computer and while I'm surfing the web, 10 minutes in - random reboot. <facepalm>

Now I've left it up and we'll see if it does this when just sitting there, I'll know when I come home if it rebooted or crashed.



But...

It's a little frustrating, yeah.


Still planning on changing out the PSU this weekend for my 850W Enermax unit. If I continue to have crashes, that's obviously not the problem, so I'll try swapping the memory back out to the Micron-die GSkill, I guess. I suppose I will figure it out eventually, just takes time and patience. Really wish it would show its true colors under stress, though, instead of randomly when surfing the web.

 

[Zerileous]

Yeah the CPU/FPU test is different than the memory test for sure. You can also try the cache test with memory which helps to stress test the CPU memory controller. Maybe the CPU doesn't like to run 3600 or maybe it just doesn't like something about the XMP profile timings.

Trying to represent CPU load as a percentage is misleading in a lot of ways. It has it's use but absolutely limited.

 

[tRidiot]

Yeah when I was testing, I did the cache and memory together. Then I added in the CPU/FPU later.


I'm just praying something, ANYTHING, is borked besides my CPU. That's the worst case scenario, if you ask me.

 

[EarthDog]

With AIDA, I would test CPU/FPU/Cache, and then test Memory alone.

You mixed memory and CPU cache which, as you found, is a lot less stressful on the CPU (as it tests memory and the cache on the CPU).

 

[tRidiot]

Hmmm, ok. Good tip. I can do that this weekend, too.

 

[EarthDog]

If you really want to cook the CPU, test FPU only.

 

[tRidiot]

If you really want to cook the CPU, test FPU only.

Is that sarcasm? Like... I don't truly want to fry it. I'm already trying to figure out if I've damaged it somehow by doing something I didn't understand.

 

[freeagent]

Was that 1.35v CCX voltage? Manually set? Seems like a lot for an AVX load..

 

[EarthDog]

Is that sarcasm? Like... I don't truly want to fry it. I'm already trying to figure out if I've damaged it somehow by doing something I didn't understand.

It was just another data point for future reference. FPU only is worse than all of those.

 

[tRidiot]

Was that 1.35v CCX voltage? Manually set? Seems like a lot for an AVX load..

My voltage has pretty much always been set at auto, and when Ryzens clock up (boost) their voltage typically goes DOWN, I thought.

I tried doing a negative voltage offset a few times and was never able to attain better stability.

And reading voltage in Ryzen Master (which I haven't reinstalled) isn't anything close to what BIOS says or HWM reads, so it's next to impossible to know what you're really running.

 

[bassnut]

Don't get ahead of yourself thinking you damaged the cpu .... Microsoft just did a bunch of updates it wouldn't be the first time they F##### something up.

 

[tRidiot]

Well, I ran Aida64 on memory test only at the faster speeds for 12 hours overnight with no problems. So I've done stability testing with the whole kaboodle (CPU, FPU, cache & memory) for 12h and with memory stress only for 12h with no problems.

Nothing apparent, anyways.

 

[krag]

To give some personal experience on the original topic of this thread about CPUs overheating and shutting down. When I first installed my 5950X and a brand new water loop. I plugged the pump power cable into the wrong mobo header. On my mobo it has a designated AIO header. I plugged it into that and my rig powered on and ran for a like 30 seconds and shut down. I thought it was the mem not being seated so I firmly pushed each stick in and tried again. same thing happened. It did this 3 times. Only to find out that my AIO plug was not compatible with the stupid pump cable. I then plugged the pump cable int a CPU fan header and the pump started to work.

I have zero problems with my 5950X and mem at 100BLK at 14cl 3600

Thankfully, the high temp shut off saved my CPU and it worked perfectly no issues. I am happy to run this CPU at factory "auto" setting for the rest of its life. I used to OC, but I kept smoking expensive computer parts trying to push the limits. (never say never hunh?) So "I think" that any degradation is probably due to OC'ing. Try flashing your bios. AMD has come out with new Agesa firmware updates for Ryzen CPUs. I have successfully flashed my mobo 4 times within this last year and the latest one from Asus has all the new firmware for AM4 and Ryzen Agesa