HD 6990 GPU Temperature benchmarks

[pcm81]

http://milkyway.cs.rpi.edu/milkyway/
or
http://boinc.thesonntags.com/collatz/index.php

All of these projects max out your GPU, just change preferences to use GPU always.
Reason why all these programs max out the GPU is because they run separate task per gpu, so 2nd gpu does not wait for work from main one, like it does in 3d benches.

I suggest you use GPU-Z 0.52, this is what I used and the consistency of the app taking the temp reading is the most important part of the test.

 

[WelshBloke]

What's the use of a controlled experiment with different programs? Why even suggest I use a dissimilar program than the exact BOINC projects the OP used? Unless of course they aren't available. I don't get it.

Well its a compute program that fully loads the GPU, its no biggie though. Dont worry about it.

Edit: Also I thought you used FAH anyway.

 

[pcm81]

I'd also suggest that we compare the full load to idle temps, rather than just absolute temp at load, because it is effected by ambient temp in the room... Its 80F in my room...

 

[Keysplayr]

http://milkyway.cs.rpi.edu/milkyway/
or
http://boinc.thesonntags.com/collatz/index.php

All of these projects max out your GPU, just change preferences to use GPU always.
Reason why all these programs max out the GPU is because they run separate task per gpu, so 2nd gpu does not wait for work from main one, like it does in 3d benches.

I suggest you use GPU-Z 0.52, this is what I used and the consistency of the app taking the temp reading is the most important part of the test.

Allright, I have GPUz 0.5.2 up. I'm trying to d/l the Milkyway Boinc client but man is it buggy and slow to d/l. Gimme a few.

 

[Keysplayr]

Well its a compute program that fully loads the GPU, its no biggie though. Dont worry about it.

Edit: Also I thought you used FAH anyway.

I do, but there is no way to tell if the load would be identical between projects such as MW@Home or F@H. Never know. It could give me a light load and result in much cooler temps for the 590. Cant have that now.

 

[Keysplayr]

I'd also suggest that we compare the full load to idle temps, rather than just absolute temp at load, because it is effected by ambient temp in the room... Its 80F in my room...

Heh, I am not cranking up my heat to get this room to 80F. Sorry, but I draw the line. LOL. :biggrin:

 

[WelshBloke]

I do, but there is no way to tell if the load would be identical between projects such as MW@Home or F@H. Never know. It could give me a light load and result in much cooler temps for the 590. Cant have that now.

Meh, ballpark figure would have been fine.

You both have different ambients, cases and airflow (presumably) anyway.

 

[pcm81]

Dont crank up the heat.

1. Record the temps GPU-Z reports at idle
2. Load the BOINC clint and connect to a project. (Make sure that BOINC has GPU use enabled. Just change from "Run based on preferences" to Use GPU always and run always.
3. You will see in GPU-Z the load on your GPUs go up to 100%. Let it sit there for couple minutes or until the temperatures stable out.
4. Please report the idle and full load temps.

 

[Keysplayr]

Meh, ballpark figure would have been fine.

You both have different ambients, cases and airflow (presumably) anyway.

So, should I not bother then?

 

[WelshBloke]

So, should I not bother then?

Its up to you, I'd probably chill out a bit about though if I was you.

 

[pcm81]

So, should I not bother then?

It would be interesting to see peak temp of 590 even though we cant directly compare our setups, I have 2x6990s side by side, so I suspect that your temps will be lower.

The biggest asset from test like this will be you knowing how hot your card in your case in your room can get under full load, because 3DM11 does NOT simulate full load...

 

[Keysplayr]

GPUz temp at idle (well I'm running 3 monitors so the temps are up a bit from 45C) is 49C.
Running BOINC Milkyway@home client. GPU Always, Run Always. GPU load at 84%?
Temp 87C. Fan speed 77% or 2610rpm.
EDIT: GPU load jumped to 94%. 89C. Fan 87% or 2910rpm.

 

[WelshBloke]

GPUz temp at idle (well I'm running 3 monitors so the temps are up a bit from 45C) is 49C.
Running BOINC Milkyway@home client. GPU Always, Run Always. GPU load at 84%?
Temp 87C. Fan speed 77% or 2610rpm.

Before you got chippy with me I was going to question if the programs would load up the same on different architectures?

Doesn't FAH load Nvidia cards differently than AMD ones?

 

[insurgent]

80F is not warm! >.>

I'm from the tropics and 80F is "comfortable". 86F, now that's warm!

 

[Keysplayr]

Holding firm at 94% GPU load on both GPU's. 87% Fan speed and 89C. Steady as she goes.

 

[Keysplayr]

98% GPU load now. 92C Fan at 95%.

 

[Keysplayr]

Before you got chippy with me I was going to question if the programs would load up the same on different architectures?

Doesn't FAH load Nvidia cards differently than AMD ones?

Would you grow up? Nobody got "chippy" with anyone, until probably NOW. Back off Welsh.

 

[WelshBloke]

Would you grow up? Nobody got "chippy" with anyone, until probably NOW. Back off Welsh.

Calm down dude. After you jumped down my throat for suggesting you used FAH I didn't want to point out that both architectures run both those programs differently so it likely wouldn't make any difference as long as you loaded the cores up to 100%.

 

[Keysplayr]

OP, how do I know how much work I'm accomplishing here?

 

[Keysplayr]

Calm down dude. After you jumped down my throat for suggesting you used FAH I didn't want to point out that both architectures run both those programs differently so it likely wouldn't make any difference as long as you loaded the cores up to 100%.

Oh my dear lord have mercy on my aching head.

 

[WelshBloke]

Oh my dear lord have mercy on my aching head.

Have the voices come back? just ignore them.

 

[insurgent]

Keys do you plan on getting a waterblock for the 590? Would be great to see what improvements can be gained using WC.

 

[hawtdawg]

You should see what kind of temps you can get with bitcoin. It gets my GPU's several degrees hotter than Furmark can.

 

[pcm81]

Your numbers sound reasonable. I suspect you have 1 GTX590 or they are not in slots next to each other?

Only 1 of my GPUs is reporting 110C, the rest vary from 83C to 100C.
See another screen shot with 4 CPU-z apps shown at the same time.
http://magicmrv.com/heat.jpg

 

[pcm81]

OP, how do I know how much work I'm accomplishing here?

Usually your card does a task, it sends results back to server and you get X credits, which are used for bragging rights. The number of credits per task depends on the amount of time it takes for multiple machines to complete tasks of that size. This gets averaged out across CPUs and GPUs. All projects run on CPU and some also run on GPU. There are about 2 million people participating in this BOINC grid.

EDIT:
The interesting part is that GPU1 runs at 500MHz and GPU2 is the hot one. I am guessing that card can throttle back all except the master GPU? I have 2 cards side by side, so the main card, the closest to CPU has worse air in take than the 2nd card. This would explain why GPU2 is hottest and if its also the "main" gpu it would explain why GPU1 is set back to 500MHz.