GPU Overheating?

[Darkfire Fox]

Hello. I recently purchased an ATI RadeonHD 3650 (AGP interface) for my 5-year-old Gateway 614GE desktop PC. I wanted to test the new 2D/Xv acceleration support on Kubuntu 9.04, and it works wonderfully so far (can't wait for OpenGL 3.0 acceleration!!)

However, I think the fan is not spinning fast enough to properly cool the GPU. My screen intermittently starts glitching out, then returns to normal. I felt the back of the GPU and I actually pulled back a blistered pinky. OUCH !!!!!!!

So.......Is there some way to increase the fan speed? I'm currently using the "ati" open source driver that comes with Kubuntu 9.04

[bridgman]

There are some power management options to lower the GPU clock, which should help. I think it's Option ForceLowPowerMode in the Device section. The power management options went in just after 9.04 locked down so you'll need to either build the driver yourself or find a package, hopefully someone else here can jump in with advice (maybe the edgers repo ?).

We aren't programming fan speed yet in the open source driver; I believe the fan speed / power relationship is programmed at power up by the VBIOS but not 100% sure about that yet. As a first step I would update the X driver and try the power management options.

[agd5f]

I suspect the glitching may be display watermark related. Use of the 3D engine could be causing underflow to the display controllers due to the extra demands on the memory controller. I just added support for properly programming the display watermarks, but you'll need the driver from git for that as well.

[ForexGuides]

The GPU is running that hot when I am not playing any games. I was thinking that my reboots are because of my PSU but now I am not quite sure. I got a fairly cheap 450W PSU for my box when I got the card but have been having problems of the machine rebooting on and off, only when I play games.

I ordered a new Antec 550W PSU but now I dont think that it will help me resolve this issue. Do you guys have an idea on how hot the GPU should be normally? Also, how hot should it be during my playing games? Also, is there any cheap and easy solution for me (other than buying another graphics card)?
Forex Guides (http://www.forex-guides.com)

[curaga]

Cheap'n'easy: reflash the card bios on windows to use higher fan speeds. But on the other hand it's sacrilege to use windows, so do only if your conscience can take it ;)

[nanonyme]

Cheap'n'easy: reflash the card bios on windows to use higher fan speeds. But on the other hand it's sacrilege to use windows, so do only if your conscience can take it ;)Also it's a terrible risk and probably voids the warrantee if vendor ever finds out. ;)

[frische]

Before resorting to such measures, check if the fan is working at all. Either open the case or use [1] in Windows and see, if the temperature is going down at all.
( Maybe you could use Linux for that, but the driver has to provide access to the sensors and I'm not sure whether fglrx or xf86-driver-{ati,radeonhd} do that )

[1] http://www.cpuid.com/hwmonitor.php

[RealNC]

This is the reason I stopped using the open drivers on my old X1950XT. I can't believe they put out drivers that may actually result in hardware damage.

[nanonyme]

This is the reason I stopped using the open drivers on my old X1950XT. I can't believe they put out drivers that may actually result in hardware damage.I partly agree. I think the powersaving options should default to on and user should have to turn them off with an override until more dynamic measures can be coded..

[TechMage89]

I think it's the vendor's fault as much as anyone's. If the card they build doesn't have enough cooling to run at full clock, I count that as a serious design flaw.

That said, "just because it isn't your fault doesn't mean it isn't your problem," so the Linux driver really should have some sensors support, and clock down the video card if it starts to overheat.

[nanonyme]

I merely meant that in the meantime before the driver gets the sensor support and can dynamically underclock the card to get it cooler, it might make more sense to always assume it's overheating than always assume it's cool. Thus a more reasonable default setting would be all powersaving features at maximum until the time when the driver can actually dynamically adjust them.

[bridgman]

Setting power savings to maximum by default would just annoy a different group of users, and some of the power saving code is still problematic on a few systems.

It might make sense to have something in between by default, say slower engine clock but full PCIE bus width or something like that.

[nanonyme]

Setting power savings to maximum by default would just annoy a different group of users, and some of the power saving code is still problematic on a few systems.

It might make sense to have something in between by default, say slower engine clock but full PCIE bus width or something like that.Is not annoying someone seriously so important that someone else's hardware has to end up at risk because of that? ;) (Though fine, as long as it's not at *full* blow by default, it's probably sane. In any case, we're just talking about a temporary thing, neh?)

[agd5f]

I don't think there's any real evidence that anyone's card will overheat with the default clocks and fan setup.

[nanonyme]

I don't think there's any real evidence that anyone's card will fry with the default clocks and fan setup.We'd be talking about evidence if we tried to find a responsible party for something. I'm more talking about a precaution, nothing more.

[bridgman]

s/evidence/indication/

[nanonyme]

s/evidence/indication/Oh, right. That's another matter then. Carry on. :)

[RealNC]

I don't think there's any real evidence that anyone's card will overheat with the default clocks and fan setup.

Maybe not "overheat" but "too hot". The card lives shorter if it's running at its maximum all the time. There doesn't seem to be an "idle" mode in GPUs like there is for CPUs. My ATI card runs at ~93C regardless of whether it actually does something or is idle. The catalyst driver underclocks it though when it's idle, resulting in a temp of 78C.

I don't think it's healthy to run at 90C+ 24/7.

[bridgman]

Just curious, have you tried ForceLowPowerMode and if so what temperature did it give you ?

[Darkfire Fox]

@ adg5f: I'm not running any 3D applications on my HD 3650, so I don't really see how my screen-glitching is 3D related, but I wouldn't know.

@ frische: Yes, my GPU fan *is* operational, as I can feel it blow hot air out the exhaust vent and see/hear it running.


How can I quantitatively determine my GPUs temperature in Linux? There are only so many fingers I can continue to blister. :p

I'm not really a fan of installing stuff from source-code, but do I simply follow the instructions at http://www.x.org/wiki/radeon in order to install new driver from git?

[bridgman]

There is no 2D engine in 6xx/7xx GPUs, so 2D and video acceleration uses the 3D engine.

If you're getting exhaust air that *probably* means your fan is running but it could also simply mean that the box is pressurized by one of the other fans.

[agd5f]

@ adg5f: I'm not running any 3D applications on my HD 3650, so I don't really see how my screen-glitching is 3D related, but I wouldn't know.

On r6xx and newer chips there is only a 3D engine so EXA ("2D" accel) and Xv (video accel) are implemented on the 3D engine.

I'm not really a fan of installing stuff from source-code, but do I simply follow the instructions at http://www.x.org/wiki/radeon in order to install new driver from git?

basic guide:
http://www.botchco.com/agd5f/?page_id=2

[monraaf]

The 3D engine is used for 2D and Xv acceleration.

Here's a guide for installing radeonhd from git on Ubuntu:

https://help.ubuntu.com/community/RadeonHD

For radeon just use xf86-video-ati instead of xf86-video-radeonhd, and you can probably skip the drm part.