SLI GPU temp differencePost Date: 2011-02-03 |
Post Reply
|
Author | |
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply
Topic: SLI GPU temp difference Posted: 03 Feb 2011 at 12:29pm |
I have a new dual-loop watercooled DS system which, except for a few Win7 hangups, is all I wanted it to be. Being a die-hard tinkerer, I have already boosted the i7 920 CPU overclock to over 4.1 GHz with no difficulty, and am now working on overclocking the SLI GTX 580s to squeeze a little more performance out of them. When doing graphics-intensive benchmarks (i.e. Furmark, Heaven), I have noticed a significant difference in the temperatures of the two cards, as much as 20 – 25 deg C. Although the highest absolute temperatures are nothing to worry about (no more than 78 for the higher of the two), I can’t help but wonder why I’m seeing such a big difference when the cards are under load. I know they’re sharing the load OK (according to the Nvidia SLI indicator bar), and idle temps are fairly well matched at about 34 and 37 degrees. Since the cards are connected in series in the cooling loop, I’m wondering if the card closest to the pump would tend to run cooler because of cooler water coming out of the radiator and pump. You would think the water flow would be fast enough so that this wouldn’t be a factor, however. The only other possibility that I could think of is that the waterblock for one of the cards isn’t making as good a contact as it should. Although I don’t see this as a real problem and all my games and programs run fine, I’m curious as to why I’m seeing such a big difference. Any ideas from the watercooling experts out there? |
|
justin.kerr
DS Veteran Joined: 06 May 2008 Online Status: Offline Posts: 5084 |
Quote Reply Posted: 03 Feb 2011 at 12:45pm |
can be a few things.
1. the temp sensors in the GPU's are very poor, so they could be off that much.
2. if running furmark, make sure that both GPU's are under load, sometimes in windowed mode, or with some other options, only one GPU will be under load, causing the temp to rise much higher on one card.
3. as far as problems with the water cooling setup, usually if the system is fairly new, there can be some air pockets still, which will cause one card to run higher under load, it takes some time for all the air pockets to work out of the GPU water block.
as far as water temp rising, here is some math for you, water moving at 1.5 gallons per minute, should be appx, what yours is at, will require appx 250 watts to raise the water temp 1 degree C, so even if you card was overclocked to the nuts, 2C water temp between the 2 cards would be all you would see, and GPU temp scales linearly with water temp.
|
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 03 Feb 2011 at 1:13pm |
Thanks for the quick reply. 1. Is there any way to prove/disprove this? If they were off that much, wouldn’t there also be a big discrepancy at idle temp? 2. Furmark and Heaven both were run in fullscreen, with the SLI indicator bar showing balanced load for each. 3. My system is about two months old, probably enough time to work out the bubbles. I think I’m going to have to live with the temp difference, especially since it doesn’t cause any problems. The engineer in me though says something needs to be fixed, but I really don’t want to tear into the watercooling setup. What would you do?
|
|
justin.kerr
DS Veteran Joined: 06 May 2008 Online Status: Offline Posts: 5084 |
Quote Reply Posted: 03 Feb 2011 at 4:02pm |
Well then, I do not know what DS policy is on installing water blocks Most experienced installers know that a slight over-tightening can cause issues, so they usually are not on too tight, but I doubt an under tightened GPU block would be the issue, the other would be a poor T.I.M. job. here is a couple of my multi GPU's on water just for reference.
Here is one of my rigs running 3 graphic cards.with pretty darn high overclocks, and over volt-ed
4 GPUs
|
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 03 Feb 2011 at 5:35pm |
I suppose I should know this, but what's a poor T.I.M. job?
During DS stress testing, is this be something they would have picked up? I can't say if I've had the temp difference since day 1, but I remember not long after I got my system I looked at the GPU temps and saw that they weren't the same, but wasn't concerned about it at the time. Just looking at the cards and the way the plexi waterblock is mounted, there doesn't appear to be any place where I could mount a thermocouple as a crosscheck on the internal sensor. I sure don't feel a big temperature difference between the two cards, so maybe it is just a linearity error in the card sensor. I guess a call to DS tech support might be in order. |
|
ablahblah
DS Veteran Joined: 12 Jun 2009 Online Status: Offline Posts: 2312 |
Quote Reply Posted: 03 Feb 2011 at 11:10pm |
Poor TIM job means a bad application of thermal interface material (thermal grease, paste, etc. whatever you call it) on the waterblock that contacts any part of the actual heat spreaders on your processing unit. This can be a bad application on your RAM, GPU, or any other things that should have TIM, and will thus lead to poor heat transfer at the source. The only way to solve a bad TIM job is to carefully remove the waterblock, remove old TIM, reapply good TIM, then carefully reinstall the waterblock.
I doubt DS would do a bad job on it, but it's always a possibility. Usually, this is the sort of things you suspect after checking down everyone else. As justin said though, best to just let your system sit out and run through its paces for a little bit to clear up any air pockets and make sure the TIM is nice and settled. |
|
R4D4RPR00F
Core i7 920 @ 3.9Ghz Asus Sabertooth X58 EVGA GTX 570 Mushkin 6GB 1414Mhz |
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 15 Feb 2011 at 10:01am |
UPDATE -- for those who might be interested......
I called DS tech support, and Robert was very helpful but had no new ideas as to what could be causing the difference in GPU temperatures. When my system was initially being stress-tested at DS, they had run Furmark and he had a screen grab showing over 15 degrees difference at that time, with the higher temperature about 80. To me, this seems high for a watercooled system, but apparently nobody was concerned enough to look into it. Robert volunteered to send out a new waterblocked GTX580, but since I had already ordered some Arctic Silver 5 thermal paste, I decided to go ahead and redo the TIM myself. This of course meant draining the system, removing the waterblock from the suspect GPU, cleaning off the old white paste (which was excessive, BTW), and applying the Arctic Silver using the credit card method. It seemed like a good idea at the time to also redo the other GPU, so I wound up doing both. After reassembly and powerup, I was disappointed to see that the temperatures were the same, neither better nor worse with the new TIM job. At this point the waterblock or onboard temp sensor now is suspect, so I called Robert back and asked him to send out the new waterblocked and thermally tested card, which he did yesterday. When I get the card in about a week and get some tests run, I'll post the results. |
|
justin.kerr
DS Veteran Joined: 06 May 2008 Online Status: Offline Posts: 5084 |
Quote Reply Posted: 16 Feb 2011 at 8:41am |
the only way, is to keep removing the variables, and you are doing that.
TIM wise, artic silver is a poor choice, poor performance and poor longevity.
MX-4 is a way better product, in both aspects.
on GPU's, we are talking about small differences here, but I usually do a very small X pattern on the center of the GPU chip, about the total size of a green pea.
|
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 16 Feb 2011 at 9:58am |
I don't think a couple degrees difference between the best and the worst TIM (see http://www.tweaktown.com/reviews/3383/arctic_cooling_mx_4_thermal_compound/index.html) will make any difference one way or the other in my case. As far as what's best, it all depends on who you believe. There are plenty of "expert" reviews of Arctic Silver 5 that rate it very highly. The point being, TIM is apparently not the cause of the temperature difference that I'm seeing. It was the #1 suspect, but now I'll have to wait until the new card comes in and hope that solves the problem.
|
|
justin.kerr
DS Veteran Joined: 06 May 2008 Online Status: Offline Posts: 5084 |
Quote Reply Posted: 16 Feb 2011 at 1:49pm |
Actually there are no experts ranking Artic silver 5 at the top, and the biggest downside is its short life span.
some people spend a ton of $ trying to get that last 3C lol.
|
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 23 Feb 2011 at 9:49am |
UPDATE #2:
The new waterblocked GTX 580 arrived from DS on Monday, and the first thing I noticed was that they had used a waterblock different from the one used on the original card. The new one was an XSPC Razor GTX580, and the original I couldn't identify. As luck would have it, the fittings wouldn't line up between the new (bottom) card and the original good top card, so I couldn't join the two with the straight connector pipe. Back to Robert at DS, who said he will ship out another waterblocked GTX 580 ASAP. I did install the new card, and am running it by itself until I get the second one from DS late next week. Temps look good -- mid 40's for an extended single GPU Furmark run. The saga continues....... |
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 05 Mar 2011 at 10:11am |
UPDATE #3:
The second GTX 580 with the right waterblock (XSPC) arrived Wednesday, and I think the temperatures for an SLI configuration are about where they should be. Now the bottom card, which is the first to receive the coolant from the pump output (after radiator cooling) has the lower temperature, and the upper card is 10 - 12 deg, C warmer, Apparently the warmer coolant coming out of the bottom card plus a slight amount of convection (heat rises) both contribute to the temperature difference between the two cards. Benchmark results: Heaven 2.1 -- 64C for GPU1 (top) at 95 - 99% load 51C for GPU2 (bot) at 95 - 99% load Multi-gpu Furmark -- 54C for GPU1 at 99% load 44C for GPU2 at 25% load Kombustor -- 50C for GPU1 at 99% load 35C for GPU2 at 0% load If anybody has a similar configuration as mine, I would appreciate hearing if your temperatures are much different. Not that I'm obsessing about this or anything, but I'd like to think my setup is as good as it can get. It is certainly better than it was before the two cards were replaced, and my hat's off to Robert at DS for his help and support. |
|
DSnoob
Newbie Joined: 02 Jan 2011 Online Status: Offline Posts: 23 |
Quote Reply Posted: 05 Mar 2011 at 6:25pm |
what about the fan speeds? do you have a profile for them? my is not LC but ever since i implemented fan speed profiles my 570 sli has been around pretty stable
|
|
supertech
Newbie Joined: 13 Nov 2010 Online Status: Offline Posts: 13 |
Quote Reply Posted: 06 Mar 2011 at 9:50am |
Fan profiles are good when you want to optimize the speed of the onboard GPU fan in response to rising temperature. In a liquid-cooled system, that fan is removed and replaced by a waterblock that circulates coolant over the heat-producing areas of the board. I do have case fans, of course -- 8 total, 5 of which are for the radiators in the GPU and CPU cooling loops. All the case fans run full bore, as they should. I just don't have individual GPU fans.
|
|
Alex
Admin Group Digital Storm Supervisor Joined: 04 Jun 2012 Online Status: Offline Posts: 16314 |
Quote Reply Posted: 06 Mar 2011 at 2:40pm |
@supertech,
The temp differences you have now are good. |
|
Post Reply |
Forum Jump | Forum Permissions You cannot post new topics in this forum You cannot reply to topics in this forum You cannot delete your posts in this forum You cannot edit your posts in this forum You cannot create polls in this forum You cannot vote in polls in this forum |