[solved] My gpu lets system crash when it is stressed

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Hello forum,

as suggested in wow thread, I tried to stresstest gpu and could reproduce the hardlock of my system that only happens when starting wow or stressing gpu.

When running glmark2 after a while, the system is hardlocked. Monitors are going to sleep and I can only hardreset the box.

The box is a small form factor pc with an integrated intel onboard gpu that is deactivated, when I installed this one:
http://www.club-3d.com/index.php/products/reader.en/product/radeon-r7-250-eyefinity-4.1604.html

Box:
http://h20564.www2.hp.com/hpsc/doc/public/display?docId=emr_na-c01570344

Its the sff version with E8500 Corde2 Duo @3.15 Ghz, 8 Gb of RAM and a powersupply of 240W.

I have changed disks, installed an ssd for system and two sata 1 TB disks for a mirrored raid1 drive for userhome. Is it possible that the powersupply is to weak for these hardware? Or is the gpu more likely overheating?

I dont get any crashes in everyday use or by emerging for a long time, so I guess, the cpu is not the problem.

How could I analyze this further in order to isolate this crashes on heavy gpu load?

I need help in how to underclock gpu, pcie or measure and find out if gpu gets to hot, or if powersupply is overcharged, or where the limitation is.

Thanks in advance for any hints, that I would really appreciate.

emerge --info

Zucca · Posted: Thu Oct 20, 2016 9:40 pm Post subject:

Hi.

I have R7 250E on my server. It's the E model that uses 75W of power max and is usually seen as passive cooled version. I used it on my desktop and worked just fine with a system that had only one case fan. No CPU fan no GPU fan. I had back then a 400W Silverstone Nightjar passively cooled PSU also.
The page you linked tell that the GPU has 61W TDP. So I believe it's the same 75W total power when under full load.
CPU-World tells that the CPU has maximum power usage of 100.31W.
So in a bad situation there's only 64W (rounded down) left to give to other components.

I'd say if there's any problems with the air circulation, your PSU is going to overheat and then maybe overload (because of resistance related things).

First I'd check the current PSU for dust and clean it.
Secondly I'd still buy the 350W PSU (if possible). ;) Since I was ok with same kind of setup, but with passively cooled 400W PSU. I'd bet that 350W with active cooling would be pretty good for the setup you have.

EDIT: I Found this.
Also you might try your luck with "CFX power supply" search.

I hope this helps.
_________________
..: Zucca :..

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Thank you for your information. I will see if I can get a more powerfull psu. Meanwhile I cleaned the box from dust, and testet what happens when disconnecting hardware like drives or removed 6 out of 8 GB RAM to limit powerconsumption. With lm-sensors I observed that as soon gpu reaches 112 degrees, system is locked (cpu temp is around 45 degrees). I tried to enable fanspeed in the hope of being able to turn fan to max anf see if it still locks or gets so hot. Sensors tells me that 120 degrees is crit, so I dont know if its normal that system locks at 112 degrees to not damage hardware. As underclocking cpu and/or pcie bus is not an option (to limit powerdrain of other components) because of limited HP bios, I would like to know if there is a possibility to measure actual powerconsumption of components. And I would also like to control fanspeeds of gpu, cpu and psu. Using OSS radeon driver because depricated fglrx does not support actual stable X server I also dont know if it is possible to underclock gpu.

So I still dont know if its a problem of overheating or too weak psu.

By the way while installing lmsensors I wanted to also activate tickless kernel but I could not find kernel option. was this moved or is there a prequisite I might have missed? And why I cannot adjust fanspeeds?

Zucca · Posted: Sat Oct 22, 2016 2:36 pm Post subject:

Whoa!
112°C is definitely too hot. And that's the reason for the lockups. You need to fix that first. Forget the PSU for now.
You should check that the GPU fan is spinning at all.
_________________
..: Zucca :..

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Hello Zucca,

ok, I see. The thing is, that right now sensors gives me:

russK · l33t Joined: 27 Jun 2006 Posts: 665

Elleni,

Just to double-check, the page you linked to with the Radeon R7 250 Eyefinity 4, on the "System Requirements" page says a Minimum of 500 Watt or greater power supply, and they link to a 600W recommended.
So I wondered if you really are seriously underpowered.

In addition to making sure the fan is working.

Regards

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Hello russK,

well card looks exactly like the one linked, and is about 3 years old. Thank you for mentioning the system requirements of the card, I was not aware of. :oops:

lspci -v gives:

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Is it possible that more than one connected monitor drains more power and generates more heat? I think I have read something like this. I now deactivated one monitor and while system is emerging stuff, I did run gl2mark, and system did not reach 112 degrees on this run, although it went up until 106 degrees. As soon as glmark run was finished, the gpu temperature went down quickly to 61 atm.

Reactivating second screen makes temperature climb to 75 degrees again

russK · l33t Joined: 27 Jun 2006 Posts: 665

Yes another monitor could mean more work for the GPU. I am not an expert with this type, I am more used to nvidia.

The product PDF sheet (cgax-r7256lm4_web.pdf) agrees with Zucca's comments in terms of the power consumption, it says 61W max, 16W idle.

The glxmark with more monitors may be spiking the power and hence the thermal load very quickly, in that case all of the thermal management needs to be working well or you are in trouble. If you are up to it you could disassemble the card and ensure good thermal grease and contact of the heat sink (if that's possible, I don't know the details). And the airflow needs to remove the heat and preferably out of the box. If it just swirls hot air around the heat sink it does not do much good.

I think the recommended power supply of 600W may be overkill if the rest of the box is relatively low power, they just don't want to recommend a supply that is too small and end up being an issue.

HTH

ct85711 · Veteran Joined: 27 Sep 2005 Posts: 1791

Heat generation that a cpu and a gpu is pretty close to each other, in that the more it has to work, the more power it uses and more heat it generates. Just rendering simple text to 2d graphics generally is pretty light on load, as the graphics card does not need to process it much. Just as russk said, having additional monitors will cause the graphics card to generate more heat, as it now has to process more.

Another way to think of it is like this, think of how hot you get just walking . (basic processing of text and simple graphics)
Now, try sprinting as fast as you can non stop (running the gl2mark and stuff). By sprinting non stop, it is almost certain you will be heavily sweating in a relative short amount of time. This relates directly to the amount of work you put on the graphics card, in that by running gl2mark, you are telling the graphics card to run full blast.

Now air circulation in the case, works right in too of by adjusting the environment you are in from outside to an enclosed room. If you need to, an easy comparison is by seeing the temperature difference in an weight room that has been empty for a while to one that's been crowded for a while.

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

I understand, thank you!

Since I ve bought this gpu (almost 3 years) I did not really feel like gaming, so this box was mainly used for desktop tasks that do not put alot of load to gpu, so I had no crashes. Now that I wantet to game a bit, I realized that there is a heat problem. I not even had lm-sensors installed before. First I had to figure out that the problem is not related to low psu capacity. As this was my second thought after thinking on a software bug. At last it effectively seems to be a heat thing.

Now with one monitor deactivated I gain about 10-15 degrees so I can configure the games gpu settings in a manner, that gpu temperature is about 90 to 95 degrees. Luckily in wow one can set a lock on max. frames so when locking it to 24 fps I can observe temperature not going higher then 93 degrees, even on high graphical settings. But not having frames locked, depending on the area ingame I can get more than 100 fps and then the temperature goes up quite quickly to the degree where system is locking for preventing melting.

I assume that heatsink is ok, as within minutes the temperature is reduced from 95 to 65 degrees when closing the game. The environement is also quite ok I think, as there is cpu fan sucking (cold) air from the front and blowing it straight to the back of the box, where warm air is comming out. The gpu is sitting right behind the fan so its warm air is going out at the back, so I think this thing is not overheating because of the pc, and there is no difference on heat developpement whether the cover is open or not.

https://www.youtube.com/watch?v=YciLgpnJy6Q

As cpu fan and heatsink seam to work well too, I wonder why the card gets too hot, when not limiting its outcome. Is the cooling system just not adaequate enough to cool it down on full load? Or can this still have to do with my weak psu; I meen, can the gpu fan just not get enough power to spin quickly enough? Or is it still somehow defective. Nevertheless I am happy to have a workaround if running with some gpu load is needed. I will just turn 2nd monitor off nd limit fps to 24 within WoW. Tough I am wondering how hot it would get when installing two more monitors as with one monitor its about 65 degrees warm and with two 75 @ no gpu load.

Thank you all guys, for your very appreciated help. I did ask the producer for support, in order to see what they suggest.

Buffoon · Veteran Joined: 17 Jun 2015 Posts: 1369 Location: EU or US

Renew the thermal paste between GPU and heatsink.

russK · l33t Joined: 27 Jun 2006 Posts: 665

Elleni · Veteran Joined: 23 May 2006 Posts: 1291

Hello guys,

thanks alot. After having replaced thermal paste, gpu operates as intended. On very high/maximum load temperature does not get over 95 degrees most of the time its even under 90 degrees. Within minutes after having switched of gpu load by closing game, temperature falls rapidly and gpu operates at about 65 degrees. So case is definitively solved. Especially as even with highest possible settings, I can use it for long period without overheating at all!

And deacivating 2nd monitor temperature is reduced further to ~52 degrees when gpu is not on load, so I guess replacing thermal paste did the job and cooling sytem operates effecive enough to hold the card in needed temperature range.