Parallel or Serial cooling for GPUs?

[ih8spam]

I just received my Koolance PC4-1025BK yesterday.

In the kit, Koolance has provided two 3/8" => 1/4" splitters and some 1/4" ID tubing that can be used to connect to my GPU blocks in parallel instead of the usual serial arrangement.

My question is this: Is there a benefit for going parallel over going serial for cooling my 8800 GTS SLI CPUs?

At first glance, it seems to me that I would get a slower flowrate over the waterblocks if the diameter of the tubing stayed at 3/8", so I think the cooling effect may be less efficient. But, there may be other benefits that I'm missing here.

Laying out the plumbing would probably be easier with a parallel arrangement, but I want to make sure that I'm getting the best cooling possible.

On a minus side, there would be 6 more junctions that a serial arrangement, so there would be 6 more potential failure points.

IH8Spam

 

[mavalpha]

Personally, parallel is better. You don't have to worry about one card heating the other, and it just looks cleaner. Just make sure your GPU block can take 1/4" tubing.

 

[Nicepants42]

Personally, parallel is better. You don't have to worry about one card heating the other, and it just looks cleaner. Just make sure your GPU block can take 1/4" tubing.

I completely disagree with this.

1. As long as the graphics cards are in the same loop, they will always be 'heating each other' - order of components does not matter in a closed loop.

2. How does more tubing ever look 'cleaner'?

To the OP: smaller tubing will not have any (positive) effect on the speed of liquid passing through the GPU block(s), because flow rate is constant in a closed loop. The only way to accelerate the liquid through the GPU block is to change the GPU block's effective cross-section. 1/4" tubes will restrict flow more than 3/8" tubes (although parallelization may reduce or negate this) and running the GPUs in parallel will reduce the flow (by about half) through each GPU block, reducing the cooling effectiveness of the blocks.

In general, the performance differences between the two setups are probably small enough such that you wouldn't notice a temperature difference, but if you really want to optimize performance, serialization is the best way to do it.

 

[mavalpha]

1. As long as the graphics cards are in the same loop, they will always be 'heating each other' - order of components does not matter in a closed loop.

I disagree. In a serial loop, the first card may have slightly lower temps than a parallel, but the second will be significantly hotter because the water has already absorbed some 75-150W of heatdump. The radiator is supposed to be able to dissipate whatever heat is absorbed, so the beginning of each loop should have a zero net heat gain.

2. How does more tubing ever look 'cleaner'?

Sorry, I prefer symmetry. I don't like how an "unbalanced" loop looks, I always have this thought stuck in my head: "So, why did I make that one the primary and not the other one?

To the OP: smaller tubing will not have any (positive) effect on the speed of liquid passing through the GPU block(s), because flow rate is constant in a closed loop. The only way to accelerate the liquid through the GPU block is to change the GPU block's effective cross-section. 1/4" tubes will restrict flow more than 3/8" tubes (although parallelization may reduce or negate this) and running the GPUs in parallel will reduce the flow (by about half) through each GPU block, reducing the cooling effectiveness of the blocks.

That I'll agree with, although I did forget that different splitters behave differently. Koolance's designs are extremely restrictive, and the only place they belong is the trash. Either run your loop in series as Nicepants42 suggested, or get a good 3/8-to-2x1/4 splitter like Swiftech's.

 

[Nicepants42]

I disagree. In a serial loop, the first card may have slightly lower temps than a parallel, but the second will be significantly hotter because the water has already absorbed some 75-150W of heatdump. The radiator is supposed to be able to dissipate whatever heat is absorbed, so the beginning of each loop should have a zero net heat gain.

The water temperature does not fluctuate significantly throughout a closed loop. You need to either define 'significantly hotter' as 'maybe 1C hotter' or post a link where the second GPU was hotter in a closed loop. I have actually had two GPUs in series in a closed loop, and the temp difference was never noticeable (>1C).

 

[mavalpha]

http://www.forum.clan-tlb2.com/forums/showpost.php?p=153828
8 degrees.

http://www.ocforums.com/showthread.php?t=521888
5 degrees enough? And yes, that's centigrade.

http://www.hardforum.com/showpost.php?p=1030939058
4 degrees there...

http://forums.tweakguides.com/showpost.php?p=49373
Another 4.

And in a completely different, more extreme configuration-
Dell H2C Peltier+Water:
http://hothardware.com/cs2007/forums/t/33850.aspx
Almost 15!

 

[Erasmus354]

How about you back that up with science mavalpha....


Lets assume a horrendously slow 1lpm flowrate (it also makes the calculations easier). Now lets assume a ridiculously hot 150W heat dump. Want to take a guess how much the water heats up?

Water has a specific heat of 4.186J/g (energy required to raise its temp by 1C), 150W = 150J/s, 1lpm = 1000 grams per minute = 16.66 grams per second.

(150 J/s) / (16.66 g/s) = 9.003 J/g

9.003 / 4.186 = 2.15

By the LAWS OF PHYSICS in this near worst case scenario the water temperature increases by 2.15C. Most graphics cards really only dissipate ~100-120W into the loop, lots of energy is lost to secondary losses. Also most loops have between 2-4 lpm, which further reduces the temperature increase. A typical loop would be hard pressed to see even a 1C gain in temperature. You know what the beautiful thing is? This 1C gain in coolant temp does not diminish the waters ability to cool at all! Hurrah for science! Boo bad thermistors and voodoo superstitions!


In conclusion, they will both cool just about the same, do which ever you want! Listen not to the witch doctors of superstition.

 

[mavalpha]

Your math is sound (at least by my 4am reasoning), but I don't think you can attribute such discrepancies to "bad thermistors" when the vast majority of people do in fact pull temps within 2-3C; minor variations can even be attributed to airflow around the rest of the case, but to blame "inconvenient" results on defective hardware is unscientific. There is more at play here than theory and perfect-scenario equations, and I'm usually the first to jump on the idealistic bandwagon.

 

[ih8spam]

And in a completely different, more extreme configuration-
Dell H2C Peltier+Water:
http://hothardware.com/cs2007/forums/t/33850.aspx
Almost 15!

mavalpha - I can't speak about the other setups, but the Dell H2C system does not liquid cool the GPUs. They only liquid cool the CPU. (I was looking into this system when I was frustrated with my own setup). Thus, the extreme 15 Deg C temperature differential is from an air-cooled GPU setup where one card is getting trapped hot air from the other GPU.

That said, I agree with you that the second GPU in a serial loop should be warmer that the first GPU due to the liquid being warmer after the first GPU. I just don't know how much ...

It's a shame that I haven't found any postings of a parallel vs. serial setup for the same case. If I was more industrious, I would be the first candidate to try - but my setup isn't apples => apples. I have the 1/4" converters in the loop.

BTW, right now I am leaning toward doing the parallel configuration (more out of need than any other reason). I can't figure out how to run the loop serially between my two GPUs (separated by only one slot) without getting a serious bend in the loop. Of course, the silly Koolance 3/8" => 1/4" converters each have 90 degree bends, so I may change my mind on this.

IH8Spam

 

[Betrayer00GT]

http://www.forum.clan-tlb2.com/forums/showpost.php?p=153828
8 degrees.

http://www.ocforums.com/showthread.php?t=521888
5 degrees enough? And yes, that's centigrade.

http://www.hardforum.com/showpost.php?p=1030939058
4 degrees there...

http://forums.tweakguides.com/showpost.php?p=49373
Another 4.

Are you kidding?

Those links just show that those setups have a difference, they demonstrate nothing regarding why there is a difference. To come to any other conclusion is "unscientific" - to borrow your formating from a later post.

Is there no room in your argument for differences between different chips? You know this kind of "binning" difference is all that Intel uses to determine whether chips become an X6800 vs an e6600, right? Just like with CPUs, there are large variations between the "same" G80 chip from one example to the next. Lets not even get into differences between separate waterblock installations, cooling sensors, heatspreader flatness, heatspreader installation at the factory, and about a billion other things.

Erasmus354 clearly showed an example of why those differences would have very little to do with loop order. Not "nothing", just very little.

 

[Erasmus354]

Are you kidding?

Those links just show that those setups have a difference, they demonstrate nothing regarding why there is a difference. To come to any other conclusion is "unscientific" - to borrow your formating from a later post.

Is there no room in your argument for differences between different chips? You know this kind of "binning" difference is all that Intel uses to determine whether chips become an X6800 vs an e6600, right? Just like with CPUs, there are large variations between the "same" G80 chip from one example to the next. Lets not even get into differences between separate waterblock installations, cooling sensors, heatspreader flatness, heatspreader installation at the factory, and about a billion other things.

Erasmus354 clearly showed an example of why those differences would have very little to do with loop order. Not "nothing", just very little.

Pretty much, my calculations ignored all of those variable factors and merely focused on the issue at hand: Serial cooling heating up the water going to the second block. I proved that the GPU does not produce enough heat to make the water significantly hotter going from one block to the next. There is no way around that, and that is why serial or parallel doesn't matter.