How to measure GPU temp levels

[advancedkill]

Hi

Here's where I also posted this: http://www.pcpro.co.uk/phpbb/viewtopic.php?p=691047#691047

I have an ati radeon 9800 xt being water-cooled by a dual 120mm fan rad (the only thing in the loop being cooled is the GPU - no CPU cooling or anything). In ATI TOOL and RIVATUNER I get the temp level around 30 degrees C, but the Catalyst Control Centre records it as around 50 degrees. Strangely, its alway 20 degrees apart (so if control centre gives 51 degrees, rivatuner/atitool give 31)

Why is this??? (I can touch the dangerden maze4 gpu block and its barely hot)

 

[Thor]

One core temperature the other being a temp prob on the outside/casing of the gpu???

 

[advancedkill]

I dont think so - check out my link and have a look at my screenshot (look at the rivatuner temp monitoring)

 

[stocky]

bloomin software!

The only way to get an accurate measurement is to stick a probe in there, preferably a calibrated Fluke - you can then compare the true temp to those given by the software to see which is closest

 

[maverik-sg1]

Bad contact - tighten the block up a bit and check again

 

[advancedkill]

So why are 2 pieces of software saying one thing (ati tool and rivatuner say around 30 degrees) and another saying something 20 degrees more (always exactly 20 degrees - that the catalyst control centre)

Are there 2 temp monitors on the card, or just one??

Thanks!

 

[stocky]

I think that Mav is going by the fact that the block is not warm to the touch, which means that the heat is not transfering as it should be; indicating poor contact.

 

[advancedkill]

Yes, but I wouold have thought that a dual 120mm Radiator with 2 120mm fans would be pretty good at cooling it. - there is also a pci fan thingy under it blowing air very fast at it (its very noisy - probably around 70cfm)

What temps do 9800xt cards normally run at (I got 65 degrees when idle with the stock cooler, but that waws with the catalyst control centre)??

Heres some pics of my rig (its still in the same place with the side panel off)

http://www.pcpro.co.uk/custompc/phpbb/viewtopic.php?t=216337&highlight=

 

[Thickbrit]

I don't know if you are,but you should never open multiple monitoring apps at once or they will conflict,especially ccc and atitool.You need to stick a probe to the back of the card,right behind the core and insulate it from outside influences.Add about 4/5 degree's to the probe reading and see which software it is closest to.

 

[maverik-sg1]

I don't know if you are,but you should never open multiple monitoring apps at once or they will conflict,especially ccc and atitool.You need to stick a probe to the back of the card,right behind the core and insulate it from outside influences.Add about 4/5 degree's to the probe reading and see which software it is closest to.

Exactly, if you require that level of accuracy that is what you have to do.

I was referring to the block being cool to the touch, and maybe contact is an issue for temp fluctuation although again I agree with thickbrit - you cant run two monitoring applications at the same time and expect to get tangible results.

Typically a watercooling system will cool to a maximum of 10degrees above ambient - meaning that if your room temp is 20degrees, you will only ever achieve a best temp of 30 degrees and so on and so forth.

What I did when I set mine up was I used Rivatuner to measure the temps of air cooling (monitoring while playing a game) then did the same when I switched to water cooling. The sensor may be calibrated differently with each software pack you use - but the temp improvement using the same monitoring software is pretty accurate.

That is to say if (under air cooling) Riva Tuner says its 65 degrees under load - then says its 40 degrees under load - regardless of what the temperature actually is the improvement is 25 degrees.

But i would check the contact by making sure that the cooler is sat level to the GPU and then turn each screw 1/2 a turn clockwise and see if the temps are any better.

 

[K404]

@ OP...are you worried about temperatures or the fact that you are getting more than one reading? Stick with the piece of software you`re happiest using and take all references from that...as Mav said, if software X shows an improvement of Y degrees between air and water cooling, the watercooling is doing its job.

-> if the waterblock has a backplate, a thin sheet of neoprene and some tissue paper a few layers thick (between backplate and neoprene) will help contact no end

 

[advancedkill]

I had a look at this site:

http://www.digit-life.com/articles2/radeon/r9800xt-3.html

it said that their 9800xt ran around 50degrees witht he normal cooler - with my reference heatsink attatched i was getting 65 degrees min (but this was from the catalyst control centre, not atitool or rivatuner)

The water-cooling seems to have worked in the fact that even the catalyst control centre now reports 48 degrees (sometimes goes around 50, but that was what the linked website said they were getting from non-watercooled components)

Ok - How do I go about overclocking my video card? (Ill assume rivatuner and ati tool are right, as a)its water-cooled with dual 120mm rad b)2 apps say the same thing c)the maze4 block barely feels hot)

Thanks!

 

[advancedkill]

Ok - I ran 3dmark 03 whilst recording temps in Riva tuner - as you can see the temps did go up when I ran the program and then returned to normal afterwards (at least this shows the temp monitor works)

However, the temps only went up by 4 or 5 degrees (maybe this is because the water temp is the same as the rad can put it back down to room temp)

 

[advancedkill]

Ok - I used rivatuner to overclock my graphics card. I got to this level (see pic), but after that I got lots of random pixel flickers on my screen (like snow). I hope I havent damgaed my card (lowered the OC as soon as I saw this, and everything seems to run fine)



I benchmarked with 3dmark 03, 05 and 06:

OC:

http://service.futuremark.com/orb/resultanalyzer.jsp?projectType=10&XLID=0&UID=7458549

http://service.futuremark.com/orb/resultanalyzer.jsp?projectType=12&XLID=0&UID=7459817

http://service.futuremark.com/orb/resultanalyzer.jsp?projectType=14&XLID=0&UID=7460330

Non oc:

http://service.futuremark.com/orb/resultanalyzer.jsp?projectType=12&XLID=0&UID=7460394

http://service.futuremark.com/orb/resultanalyzer.jsp?projectType=14&XLID=0&UID=7460086

I still dont know why the temps are reported differently.

 

[stocky]

Nice results Make sure you get them entered for the team, here and at HWBOT :B

If I remember right those marks are usually Core errors, where as the black marks are Mem errors. Have a play and find a good compromise between the two.

It is common for the temps to be different in other software as they haven't been calibrated to every type of card and probe - even if they were there are still tolerances between them. Especially as each and every electronic component has its own tolerance, so as you can imagine they all start adding up!

If you're really interested get yourself a probe in there, either on the back or the card (& insulated) as Thicky suggested or with a bit of work you can get one close to the die.

 

[advancedkill]

Now how would I go about entering my results for the EP-UK team?

I tried a more dumb-**** moethod of finding the temperature: Whilst running HL2:Lost Coast (the last game I have installed, as I had to re-install windows), I touched the copper metal on the back of the card. I guessed that if it was running at 50-55 degrees, I would burn my finger, but I wouldnt if it was 30-35 degrees.

Well, it wasnt even as hot as a cup of tea after leaving it for 10 or 15 minutes (im guessing that its around 30 degrees then)

I was wandering (is wandering spelt wOndering or wAndering?) whether or not to add UV die to my coolant (I'm currently using de-ionised water from TESCO).

Good idea:

If I turn on the UV lights with my case I can a)freak out my friends even more and b)see if any of it is leaking over my nice rig - that would be useful

Bad idea:

I dont know what's in it and it says that it stains skin and is an irritant on the packet. I also dont know what it will do to the insides of my water-cooling parts. Oh, and it would be a bit hard to drain the system (I would totally mind having water spilt on me as much as uv die)

Heres where I got it from: http://www.kustompcs.co.uk/acatalog/info_2472.html

Thanks!

---edit---

Wow - the forum even scubs out A r sss e! *(better than most forums)

Hehe

---edit---