HD 6990 GPU Temperature benchmarks

pcm81

Senior member
Mar 11, 2011
581
9
81
Hello all.
Just a quick heads up that the GPU temp reported by GPU-Z during 3DMv11 is not as hot as the card actually runs at its peak... I use BOINC client to donate clock cycles and I usually leave it running overnight. My 6990s in CF usually run up to 109*C. This is the hottest temp that is reported by GPU-z. When I run 3DMv11 on Xtreme setting also on 4 GPUs, in CF mode, the highest temp that GPU-z reports is 100C. I got 5036X 3DMv11 score on the run which gpu-z reported 100C as highest temp for gpus...

I have no reason not to trust gpu-z version 0.5.2 but it looks like 3dm11 does not heat up the card as much as massivly parallel computing models like MW@home or Collatz@home do.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Hello all.
Just a quick heads up that the GPU temp reported by GPU-Z during 3DMv11 is not as hot as the card actually runs at its peak... I use BOINC client to donate clock cycles and I usually leave it running overnight. My 6990s in CF usually run up to 109*C. This is the hottest temp that is reported by GPU-z. When I run 3DMv11 on Xtreme setting also on 4 GPUs, in CF mode, the highest temp that GPU-z reports is 100C. I got 5036X 3DMv11 score on the run which gpu-z reported 100C as highest temp for gpus...

I have no reason not to trust gpu-z version 0.5.2 but it looks like 3dm11 does not heat up the card as much as massivly parallel computing models like MW@home or Collatz@home do.

Get your water blocks installed, mate. The stock cooler just doesn't cut it for that type of use.
 

pcm81

Senior member
Mar 11, 2011
581
9
81
Get your water blocks installed, mate. The stock cooler just doesn't cut it for that type of use.

They shipped out today... i have ups tracking #...

EDIT:
The 110C from BOINC is with stock clocks of 830 MHz. In 3dm usually only 1st core is at 100% load/clock and rest are going slower, so card automatically undervolts them, in BOINC all cores run separate tasks so they all at 100% load and therefore 100% of stock voltage.
 
Last edited:
Feb 19, 2009
10,457
10
76
Keep it looping for a few days, see if it explodes. :D

I'm betting NO. There's a reason why those cards are so huge and crammed full of the best components.

ps. Nice setup.
 

pcm81

Senior member
Mar 11, 2011
581
9
81
Keep it looping for a few days, see if it explodes. :D

I'm betting NO. There's a reason why those cards are so huge and crammed full of the best components.

ps. Nice setup.

This is why I did not OC it yet, but it runs fine at 110C for over 24HR...
 

OCGuy

Lifer
Jul 12, 2000
27,227
36
91
110c? Holy smokes...that is going to lower the life of those cards. Water, ASAP!
 

pcm81

Senior member
Mar 11, 2011
581
9
81
110c? Holy smokes...that is going to lower the life of those cards. Water, ASAP!

I know, I know... not running BOINC any more, until i get my coolers. Bottom line is:

To test your HOTTEST performance, use BOINC not 3DMv11...
The reason is this:
In BOINC each GPU is working on his own task and so is always at 100% load and 100% clock so all GPUs are also at 100% stock voltage. In 3DM your GPUs are working together and only your 1st GPU is at 100% load, ther rest are waiting for work synchronizations... In this cse only 1 GPU is at full voltage and clock so the net thermal capasity of the card is lower than its 100% thermal capacity.
 

pcm81

Senior member
Mar 11, 2011
581
9
81
I guess I just added a data point that HD6990 actually works at 110C... Not going to try to test how long its going to work for... but it works...
 

pcm81

Senior member
Mar 11, 2011
581
9
81
Shameless bump, because people need to know that during video rendering benchies their cards are not running at the peak temps that could be at cards full load...
 

WelshBloke

Lifer
Jan 12, 2005
30,331
7,987
136
Is that the GPUs or VRMs getting to those temps?

Also I don't think those dual GPU cards (both Nvidia or AMD) are suited to tasks which involve them running at full load for extended periods.
 

pcm81

Senior member
Mar 11, 2011
581
9
81
Yeah I think the last figure (gpu 3) is for the VRM's.

Also GPUz doesn't report the actual voltages, it just reports what they should be.

I still hold that those class of cards aren't suitable for that sort of workload.

The cards are at stock voltage, I did not flip the switch... waiting for water cooling first. BTW 104C and 102C for 2 other temps is still pretty high.

I bet water will take care of it though. 6990s are pushing the air limits and then some... but the power circuit seems to be ok... I do not see any failures at the 110C either. Seems like a very robust design.
 

WelshBloke

Lifer
Jan 12, 2005
30,331
7,987
136
The cards are at stock voltage, I did not flip the switch... waiting for water cooling first. BTW 104C and 102C for 2 other temps is still pretty high.

I bet water will take care of it though. 6990s are pushing the air limits and then some... but the power circuit seems to be ok... I do not see any failures at the 110C either. Seems like a very robust design.

Yeah if you aren't forcing different voltages then GPUz should show the correct values.

The 6990's do seem to be built to take the punishment. I think VRM's commonly run hotter than GPUs, I seem to remember them being the limiting factor in overclocking some earlier AMD cards.

Anyway good luck and enjoy those cards.
 

pcm81

Senior member
Mar 11, 2011
581
9
81
Yeah if you aren't forcing different voltages then GPUz should show the correct values.

The 6990's do seem to be built to take the punishment. I think VRM's commonly run hotter than GPUs, I seem to remember them being the limiting factor in overclocking some earlier AMD cards.

Anyway good luck and enjoy those cards.

Water coolers should be here on Wed, so next weekend i should have them on water and will report those temps...
 

pcm81

Senior member
Mar 11, 2011
581
9
81
...and post pics :colbert:
By the way, I am running at 85 idle, because the ambient in my room is 80... i like it warm...

So, the post important thing to take from this thread is:
1. Your GPUs might be hotter in some cases, then they are under game load... Graphix benchmarks do NOT use 100% of all GPU clock cycles.
2. 6990 GPUs do not melt, flicker or crap out at 110C, but i cant speak for life expectancy at those temps...
 

SolMiester

Diamond Member
Dec 19, 2004
5,331
17
76
Does BIONC has a CUDA client setting?, I wonder if the NV 590 gets as hot (different components?)...110c cant be good for the card\s!
 

Keysplayr

Elite Member
Jan 16, 2003
21,209
50
91
I can try it if you like.
Link me to what you're running and I'll see if I can use the CUDA client if there is one.
 
Last edited:

Keysplayr

Elite Member
Jan 16, 2003
21,209
50
91
Wouldn't FAH work just as well?

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.