Interesting, so it’s during heavy viewport updates. Still sounds like a fairly high load. I’m guessing it is indeed the GPU then. Was still reading through the thread though, so not sure if you have tried another GPU in your system yet or not?
Do you have the chance to check how much Power your machine consumes when this happens?
My theory: Your Power supply is too weak, and when your graphics card starts getting busy with the scene, power supply breaks down completely, killing everything. Link to power supply stats:
As this seems to happen accross operating systems, I think it’s a valid root cause theory.
Solution would be: Get a stronger Power Supply Unit.
Damn i think it’s gigabyte h110m-s2pv ddr3.
Pc is turned off so cant check now
But i think its this one
they don’t have unpacked GPU to test on my motherboard they have actually on other pc.
They have a small cyber cafe I will request them to test those graphics card on my pc.
But I think if they wont if the cyber cafe is full.
But they have extra pc to test my GPU on it. But 2 days ago they tested my graphics card on a pc that was faulty itself. That computer was restarting even if it was idle
Yes I was monitoring all this time.
Can’t tell about blender but in substance painter when I clicked on export textures after couple of seconds GPU jumps to more than 150 watts.
It doesn’t restart sometime but I noticed that after jumping to 150 watts PC restarts.
Well maybe now you can say it’s the power supply problem. But GPU stress test pulls almost 185 Watts from the graphics card. Same as most of the games ,marmoset toolbag, unreal engine 4 and even with that scene with 2px tiles size
Where do you monitor your (overall & GPU/CPU) power consumption?
One of these wall socket measurement units might be an option.
Actually, I don’t get why your PC store put a 550 Watt PSU for testing, when your old one had more (650 Watt). I’d still try a stronger power supply and see if the problem still persists.
Well, similar to the other h110m boards, the VRM is certainly by no means impressive, but I would expect it to be fine-ish for an i5, especially if it’s a quad core, which it should be.
Ok, I’m pretty certain it is the GPU VRM. In a gradual buildup(of load and thus heat) the fans might be able to offer enough cooling not just to the GPU die but the vrm too, but in a sudden high load scenario the VRM thermal protection might be triggering before the airflow from fans can manage to cool it down.
What was the exact GPU manufacturer/AIB(add-in board partner)? I.E. GIGabyte, ASUS, MSI etc. Not chip manufacturer(AMD in this case, obviously). And what exact model?
It might be one that doesn’t have proper if any thermal pads for the VRM.
Because they’re stupid, and don’t have any unpacked psu to test, only had that one.
Also sounds like they don’t event want to unpack e new psu to test it
I don’t even know any friends who was better psu than me so I could give it a try
It’s a Sapphire rx 590 OC edition.
And even in other software too, like Marmoset toolbag which causes gpu jumps 35 to 180watt instantly but still pc doesn’t restart
But why only with these 2 things?
Sorry I’m not an expert about this, just asking
Who knows. Might be just up to tolerances. In UE and Marmoset it might be just barely, but a sufficiently gradual change in heat.
Also a possibility that blender and substance make use of some feature that utilizes the fixed function pipleine in a way that the others don’t, maybe theres a gpu defect affecting that specific feature. This is very borderline and frankly an unlikely case.
But if you can prove that the same issue happens with the same GPU in any and every other system too and you still have it under warranty… I’d just contact the manufacturer and ask for a replacement card.
Theres still one thing you can test that comes to my mind. You can try slightly underclocking and undervolting the gpu through the Radeon Wattman in the drivers control panel. Then retest blender and Substance.
Great idea I’ll do that tomorrow.
And another thing i forgot to mention.
I tried lowering the highest voltage usage of that gpu,
50% lower, after that gpu couldn’t even use 100w properly
But when i opened that scene again in blender pc restarted even after lowering the power usage
Edit : i didn’t do all of these what he’s doing in this video.
I just simply lower the power usage by 50%
Undevolting too much will not only negatively affect the stability, but can even degrade the silicon, same as overvolting. But it’s typically safe to undervolt by 30 or 60 mV when it comes to AMD GPU’s, you might even find that it stabilizes your clocks in games, and will on average find that your gpu runs higher frequency and more stable.
OK, it’s midnight now on my end too, and I’ll go to sleep. Will check back tomorrow.
I did not read if you found the issue BUT:
There are two kind heat sensors in some cpu or mobos. cpu itself and cpu socket.
I had this issue with my mobo that cpu’s heat would not go beyond 60 but my socket went beyond 90 and shutdown. I never notice this until I install some hardware monitors.
Also there is Mosfet heat and so many things on mobo.
I first fixed it by under clocking the cpu.
some month after bought a new cooler and no shutdowns anymore.
Thank you but no I didn’t find any solution.
It can be my graphics card on the motherboard but like you said also it can be temp issue that I don’t know.
I installed HWmonitor and it shows me temperatures of the motherboard.
About 35-42C i think.
but this problem shouldn’t be a temp issue cause sometimes when I change the tile size to 4PX PC restarts instantly before even heating up anything.
Some people mentioned about graphic cards Vrms could be facing some issue.
I also think that actually is still can’t be sure
Interesting. How old is your PSU? Do you have a speaker? Like this one:
If you have, install it according to manual and listen - maybe it will tell you via beep codes what the hell does it want.
I had the similar issue, coincidentally on Gigabyte mobo, with only difference that it wasn’t caused by Blender and it didn’t recover after the shutting down, endlessly beeping about power delivery failure. Otherwise it looked the same - same second of display freezing and constant, but not maxed out speed of fans. Unfortunately, it was resolved only by replacing the board. That board was 5 years old, so maybe it was some kind of degradation.
I tried the file Md_utsho shared and as soon as i rendered it in the viewport and tried to pan around the scene Blender crashed and my PC shutdown and rebooted. I use 50 viewport samples and the 4px tile size.
Here’s a list of my hardware.
CPU, AMD Ryzen 5 2600.
GPU, Nvidia GTX 970
PSU , Corsair 750 watt
I6 GB RAM
OS Linux Mint 19.2
