[SOLVED] PC turns off on all core load after CPU upgrade

[RcCookie]

I recently upgraded my Ryzen 5 4650g to a 5950x. Ever since the PC has been turning off when under heavy CPU load, but not perfectly consistently (not a problem before). When it turns off, the rgb and the fans stay on. Sometimes it reboots, but most times nothing helps but turning the PSU off and back on again (not holding the power button, nor the hard reset button).

My specs are:

• Asrock B550m Pro 4
• Ryzen 9 5950x
• Cooler Master 212 Evo
• MSI 3070 Suprim X
• 2x16GB@3200 & 2x8GB@3600 Crutial Ballistix
• Corsair RM750

I am aware of the fact that the 212 Evo is not really enough to properly cool the 5950x, and an upgrade is planned. However, this doesn’t seem to be the problem. The PC also crashes when the CPU is not overclocked, running at ~3800mhz under all core load, with <70°C.

Interestingly, when I first installed the new CPU, it used to run @~4500mhz under all core load, eventually throttling down to ~4000mhz, but always running at the 90°C thermal limit, and using ~150W. It used to crash there as well, but not too regularly, I managed to run a 10 minute Cinebench run just fine. Since then I have updated my bios to the latest version, installed new chipset drivers and Ryzen Master. I am not sure what of these made the difference, but now the CPU is immediately throttling to 3800mhz as soon as there is an all core workload, even though it then hovers around 62°C, with an average power draw of 120W. Like this it also doesn't crash during a Cinebench run, but still during gaming like previously (I like playing BeamNG.drive, that does stress all CPU cores).

I have been playing around with Ryzen master, and what I noticed that both AutoOC and manual overlocking cause a quicker crash. For AutoOC lowering the power draw limit seems to make it more stable, but at 150W it is still crashing during Cinebench unlike before. Manual overclocking has similar effects, even just a constant frequency of 4100mhz causes a crash during Cinebench.

A quick conclusion over what I've tried without any success:

• Clear bios
• Update bios
• Update chipset drivers
• Windows (11) is up to date
• Run with only one type of RAM
• Check Windows event viewer: critical error 'Kernel-Power' with id 41 (didn't find anything useful online)

So I wonder: what could be the problem? I thought maybe its the PSU, but then everything should power off at once, right? Like I said, the fans keep spinning, as does the RGB stay on, and besides, even if the GPU was to draw 400W and the CPU 250W, which is not even remotely the case, the 750W PSU should be fine.

My last two guesses are either a CPU failure itself, or maybe the mainboard? I am not very familiar with power phases and that stuff, however, based on OpenHardwareMonitor the mainboard components never pass 65°C, even when forcing a power draw of >170W using manual overclocking. However, it is only a B550, so if there is a realistic chance that that is the point of failure, I will replace it with an X570. However, I can't really affort spending money on multiple components right now, that would mean the AIO had to wait.

 

[drea.drechsler]

...

• 2x16GB@3200 & 2x8GB@3600 Crutial Ballistix...

...

Try clearing CMOS while also taking the battery out.

What temperature is your VRM operating at? That's as important to keep an eye on as CPU temperature with a 16 core CPU.

For now at least uninstall RyzenMaster and do it before resetting CMOS with the battery pull. Not until you've got the system running stable should that be installed, if at all.

And that mix of memory is very troublesome. I wouldn't imagine even 3200 to be stable with 4 sticks and 3600 is doubtful at all. I'd expect 2666 is probably a best case scenario without some clever memory tweaking.

 

[RcCookie]

based on OpenHardwareMonitor the mainboard components never pass 65°C, even when forcing a power draw of >170W using manual overclocking

 

[drea.drechsler]

based on OpenHardwareMonitor the mainboard components never pass 65°C, even when forcing a power draw of >170W using manual overclocking

Suggest you get HWInfo64 as it works much better with AM4 motherboards and processors.

ADDED: oh yes, and the Kernel Power event is just Windows telling you the system shut down unexpectedly. Which is something you already knew...but is useful to know if running a kiosk system (for example) that restarts whenever it shuts down unexpectedly.

 

[RcCookie]

Suggest you get HWInfo64 as it works much better with AM4 motherboards and processors.

I tried again with that, I forced a quick crash by overclocking manually to 4300mhz using Ryzen Master (I will uninstall it and clear the cmos eventually). I actually started logging the sensors just before the crash, so I got a CSV with a single line, but no suspicious temps; all mainboard sensors at most like 70°C. This is the whole CSV.

 

[drea.drechsler]

I tried again with that, I forced a quick crash by overclocking manually to 4300mhz using Ryzen Master (I will uninstall it and clear the cmos eventually). I actually started logging the sensors just before the crash, so I got a CSV with a single line, but no suspicious temps; all mainboard sensors at most like 70°C. This is the whole CSV.

If it's crashing when fixing the clock that low then something else is wrong IMO. Have you also fixed the voltage, or taken it off AUTO?

Definitely uninstall Ryzenmaster, reset CMOS with a battery pull. Restart the system while making no changes to BIOS settings and check for stability in fully defaulted settings before doing anything else.

BTW, never overclock 5000 series CPU's manually. At best it's useless and most likely wasting performance potential unless on sub-ambient cooling.

 

[RcCookie]

BTW, never overclock 5000 series CPU's manually. At best it's useless and most likely wasting performance potential unless on sub-ambient cooling.

I didn't plan to overclock it, the reason I tried is because of the weird trottling behavior since I updated the bios and co. Previously it would just slowly decrease the clocks over time when getting too hot, now it keeps the target frequency and then starts to briefly reduce the clocks down to like 500mhz (or less, maybe it's a sensor resolution artifact). Ever since my cinebench scores have dropped significantly (like 3000 points or something) without any other changes to cooling or something. No surprise though if you see that it is throttling way early before even close to 90°. I will uninstall Ryzen Master, clear the CMOS and then monitor the boosting behaviour.

Have you also fixed the voltage, or taken it off AUTO?

I haven't changed anything. In Ryzen Master, what I've tried is:

• PBO (decreases performance even more, way more 500mhz spikes)
• AutoOC, with default (basically unlimited) power limits and decreased down to 150W on all 3 options, also actually with very low or negative boost override (25 is the lowest possible, or negative, I tried -100)
• Manual OC, simply set the frequency of all cores, nothing else touched

 

[drea.drechsler]

....
I haven't changed anything. In Ryzen Master, what I've tried is:

• PBO (decreases performance even more, way more 500mhz spikes)
• AutoOC, with default (basically unlimited) power limits and decreased down to 150W on all 3 options, also actually with very low or negative boost override (25 is the lowest possible, or negative, I tried -100)
• Manual OC, simply set the frequency of all cores, nothing else touched

Definitely remove RM and reset CMOS. Operate it in completely default settings to see if it's running stable before attempting anything else.

It also sounds like you're misunderstanding Ryzen performance. It's perfectly normal to be "spikey" as it uses an aggressive boost algorithm, very much different from the way Intel processors work. It's not a sign of poor performance.

The best way to tweak performance of 5000 processors is to use PBO with Curve Optimizer. And do it in BIOS, without using RyzenMaster since it's service is known to cause problems. Also take note that it's designed to runs upwards of 90C, which is perfectly normal.

 

[RcCookie]

Also take note that it's designed to runs upwards of 90C, which is perfectly normal.

So, I cleared the CMOS, haven't even opened the bios since. I'm not even running XMP. The thing is, I still get the same behavior where it doesn't even try to may out the temps. I've been running Cinebench for 6 minutes, and the CPU temperature is 60°. That doesn't make any sense! By default, the fan curves are even set to "silent", they wouldn't even go to 100% if the CPU was 100° hot. And after all, it's only a 212 evo, it shouldn't be able to cool the CPU. So why is it behaving like this?

In terms of stability, it's too early to say anything, I'm going to have to observe this for a day at least.

 

[RcCookie]

It's perfectly normal to be "spikey" as it uses an aggressive boost algorithm, very much different from the way Intel processors work. It's not a sign of poor performance.

All I can say is that the benchmark scores are terrible now in comparison to how it performed before the change of behavior (for whatever reason).

 

[drea.drechsler]

All I can say is that the benchmark scores are terrible now in comparison to how it performed before the change of behavior (for whatever reason).

Uninstall RyzenMaster? If you're really enamoured of it you can always reinstall it, but I'd suggest not.

But it does seem strange to be so low temp running Cinebench. I'd seriously suspect the VRM is throttling except a B550M Pro 4 isn't really that bad of a board.

 

[RcCookie]

I have uninstalled Ryzen Master (before the CMOS clear, the previous reports are after it was uninstalled. It didn't change anything).

But it does seem strange to be so low temp running Cinebench. I'd seriously suspect the VRM is throttling except a B550M Pro 4 isn't really that bad of a board.

What I don't understand is why this behavior changed over time. Like I didn't have better VRMs at the beginning either, yet it had no problem running at 85-90 degrees constantly without unnecessary throttling.

 

[RcCookie]

Do you think it could just be "bad silicon"? If so, what should I do? The way it currently is really isn't usable. I can use it for testing and for some minor work, but it crashes even on not so extreme loads, an hour ago even just from running Firefox and some monitoring tools.

 

[RcCookie]

This is the current throttling behavior:

 

[RcCookie]

(Pink is the CPU, red the mainboard sensor, whereever that is)

 

[RcCookie]

In terms of stability, it's too early to say anything, I'm going to have to observe this for a day at least.

Never mind, crashed again.

 

[drea.drechsler]

(Pink is the CPU, red the mainboard sensor, whereever that is)

I've no idea what your monitoring utility considers the "mainboard" sensor. It could be the chipset or a discrete sensor in the vicinity of the GPU. That's why I recommend looking at HWInfo64's sensor screen since it's very good at explaining each of the myriad of sensors readouts when it can.

Crashes with default BIOS settings and no overclocking software could be something Window's related too...except the processor simply isn't performing right. Would be really good to get a sensor screen reading from HWInfo to see what's going on with both motherboard and CPU voltages.

Also, it's not clear what you mean when you say it never did this before: with the 4650g processor in it? or when first installed the 5950X?

 

[RcCookie]

HWInfo64 doesn't know anything more:

It says this for all entries except "Motherboard" which just says "Mainboard temparature" and "CPU", which is irrelevant. Mainly though, these values are also not higher than the values from above.

 

[RcCookie]

Also, it's not clear what you mean when you say it never did this before: with the 4650g processor in it? or when first installed the 5950X?

I meant it didn't crash with the 4650g.

Crashes with default BIOS settings and no overclocking software could be something Window's related too...except the processor simply isn't performing right. Would be really good to get a sensor screen reading from HWInfo to see what's going on with both motherboard and CPU voltages.

Which exact values would you like? Do you want a log file? I suppose a single instant isn't really enough. I can also try showing you some specific graphs.

 

[drea.drechsler]

I meant it didn't crash with the 4650g.

Which exact values would you like? Do you want a log file? I suppose a single instant isn't really enough. I can also try showing you some specific graphs.

4650G, 6 core 12 thread, max boost 4.2G, is definitely a lightweight compared to 5950X, 16 core 32 thread, capable of boosting to 5Ghz. It could run wide-open on even the weakest of VRM's and not stress it much.

There are no sensor readings above Motherboard? that's odd that an Asrock board doesn't have a VRM temp sensor reported. I thought that's only something Asus does. But maybe that's it: I'd put load on and off and see if there's corelation with the motherboard reading to gain confidence in it or rule it out.

So I'd like to see the entire CPU section. First without a heavy processing load (Cinebench is best) then with it. You don't have to expand each of the core sensors. What's important is to look at current, min and max. Also average for a while.

You can right click on each sensor reading and put a graph on the desktop to monitor it visually too.

 

[drea.drechsler]

I meant it didn't crash with the 4650g.
...

Oh yes..and one thing to check:

Open MSConfig, type it in the search box. Go into BOOT tab, the ADVANCED Options button and make sure the NUMBER OF PROCESSORS box does not have a check in it. For some reason it can get a check with a reduced number of processors with a CPU change out.

If you do clear a check, restart Windows and test again.

 

[RcCookie]

Here are some screenshots:

Idle:

Cinebench:

Prime95 (interestingly, it boosts higher here, yet not to the 90° mark):

Here is a selection of graphs (The idle stats are before any CPU load, the first major CPU load is Cinebench, the second, less consistent load is Prime95) (I don't really have a good idea what this is, so it may help, it may not):

Also note that I did run Prime95 for longer before the shown graphs, and it did start throttling a lot more (fair enough). Didn't include it here so that all tests fit onto a single graph.

 

[RcCookie]

Oh yes..and one thing to check:

Open MSConfig, type it in the search box. Go into BOOT tab, the ADVANCED Options button and make sure the NUMBER OF PROCESSORS box does not have a check in it. For some reason it can get a check with a reduced number of processors with a CPU change out.

If you do clear a check, restart Windows and test again.

Wasn't ticked.

 

[RcCookie]

Did some more prime95 testing, this might be more interesting. I also added all the "current" stats I could find.

 

[drea.drechsler]

Here are some screenshots:
.....
Also note that I did run Prime95 for longer before the shown graphs, and it did start throttling a lot more (fair enough). Didn't include it here so that all tests fit onto a single graph.

There does not appear to me to be a VRM temperature readout which is sad. One thing I note is the CPU Vcore (SVI2 TFN) never gets very close to 1.5V which I'd expect with full default settings in BIOS and no RM profiles trying to influence things. That can have a significant impact on system stability when power delivery from a VRM isn't stable or reliable.

It's hard to say for certain but I still wonder if VRM is throttling the CPU: just look at how core clocks are steadily decreasing with time into the test. If you have an IR thermometer handy I'd check temps all around it while running Cinebench. Especially the capacitors.

And oh yes, definitely definitely absolutely disable running AVX when testing with Prime95. That's a sure fire way to damage that motherboard's VRM with a 5950X running on it. Or better yet, just don't use it as it's largely irrelevant with modern systems anyway.

ADDED: and I just noted something that's curious since the system is supposedly in full default after a CMOS reset: it's using a 1600Mhz IF clock so 3200Mtps memory. It's typical to see 2133 or 2400 default DDR4 memory settings after a CMOS reset.