Dual Loop versus Single, the facts

[rge]

=

Why did I do it then - well, for aesthetics, practicality, symmetry. So, I am not against this, merely saying it's not really required - but I do it anyway, because I can.

Same reason as me, and likely many others.

[MrBean]

A little bit off-topic, a rig I had back in 2004/2005, competing against Macci and our own Fugger, and this setup I actually remebered I got no1 in it's class, which was held by Fugger - score was 14,768, and I know the CPU was a 3.4EE clocked at over 4.75Ghz, mobo was an Abit Max3, and I had a very nifty VTT auto-regulating VMod on the memory side, allowing me to clock the DDR Mushkin blacks to something like 292-298Mhz speeds, which was unbelievably crazy back then. Still have those sticks...

Those were the days, but just a quick photo or 2 to show the extremes we went through to really overclock and compete - this was my 24/7 rig btw, for benching, I would really ram up the dyi trimpots on Vgpu and Vmem, and normal operation, turn it down again. First versions of this system had the Vapochill (with higher gas-load) cooling the CU, and chilled water/acohol (-7 or somewhere there) chilling the GPU, but for 24/7 use, I reverted to closed-loop water on the GPU, with the Vapo freezing the electrons of the silicon-wafer on the CPU.

Here's verification from an old thread from South Africa, posted back in 2004, when I was the sole local boy running Intel Northwood with nVidia graphics, and holding my own against the then CPU holding the performance crown, AMD - I think this was the time Kingpin first stepped onto the world-scene, if memory serves.

http://forums.pcformat.co.za/viewtopic.php?f=29&t=8360

Here's a linky to a Northwood, with aircooled 6800 nVidia, getting open-class Top 10 - back in the day when overclocking actually relied on things like knowing what you were doing with electronics, modding the bejesus out of your hardware, doing speca bits with the software, like running W2003 server, trimed down to bare essentials, and restoring a fresh image after every bench-run - just to make it to the top 10.

How many of you guys do this today?

http://3dmark.com/3dm03/3396884

So, I just thought a bit of history would be in order, as my credibility was somewhat questioned yesterday -> so guys, if I give you my word that I attain specific results with a hardware combo, please believe it - no reason for me to bull, been around the block to many times.

Anyway, enjoy, and have fun.



and a close-up:

[matari]

It's not all about performance. For more than three blocks, I have always found it easier to setup and maintain dual loops instead of one giant loop.

Also, don't care if my GPU runs somewhat hot, however, I want my CPU to run as cool as possible. I will complicate the issue by wanting to run my fans at ~500 RPM. Much simpler to put the GPUs on separate loop than have one loop.

[Church]

Gabe/Stephen: btw, i got curious how you guys test things. I guess that some walk-through pics from your lab might be tasty treat . Do you use some testloops only or some fluid dynamics modeling software as well? And probably some more serious grade instruments instead of customer grade flowmeters/temp probes/anemometers..?

[Utnorris]

I have done three loops, two loops and single loops. I have done various builds since I got into this about 5 years ago. For me, doing different configurations was for learning, looks and because I could. Heck, right before I went to my current build inside my desk I had a UF2O bolted on top of an Ascension. The UF2O held the PC parts and the Ascension had the water cooling gear in it. It had 2 x Feser Quads, a Feser Triple and I was getting ready to add two PA120.3's to the loop, all in a single loop. Why? Why not is what I ask. If you have the room and the gear laying around and you like playing with it then do it. I have also done a phase on my CPU and water on the rest, not to mention I built a water chiller out of a window AC and used that for a while. So I have done a lot of different builds, more than I can count on my hands, feet, my kids' hands and feet, you get the idea, just because I enjoyed it. With my current setup I chose a single loop for my CPU and GPU's. Why? Because it was the simplest way to do it and the most compact. I have three PA120.3's with Triberk fans running at ~800Rpm and it keeps a 2-3c delta between the air intake and the water temperature. This is cooling a 2600K overclocked to 4.8Ghz at 1.42v, a HD6990 with XSPC FC water block at stock (880/1250) and a HD6950 (unlocked to HD6970) with an XSPC FC water block at 880/1250. IMHO I get this type of delta because I am over raded, but I also do not fold 24/7 or run furmark 24/7. I do normal, everyday stuff like gaming for 3-4 hours at a time or video conversions. My goal on this build was for it to be quiet and cool, which it is. Could I get better results with two loops? I doubt it, but it could be more efficient by being two separate loops, meaning I could probably drop one of the rads and still maintain except-able temps on my CPU and GPU's if they were in different loops. Every build is different and to say dual, triple or single is the best way is pure BS. There are too many considerations that are unique to the situation to say one is better than the other.

[gmat]

I have been through many different PC's and loop setup since i went into this 10 years ago, parallel, dual, etc. For my most recent build i finally decided on a dual loop:
- CPU + GPU's in parallel
- Motherboard
For the only reason that the motherboard full cover block is so restrictive, it kills the flow even with dual pumps. So i have dual pumps on the CPU+GPU loop, and a single pump on the motherboard loop.
Then the CPU loop has GPU's in parallel, and 3 radiators (2 triple 140 and 1 triple 120). So i can handle full load on everything, or turn off (completely) a lot of fans at idle. This makes a pretty silent PC in light use or idle, and it starts whooshing a bit at full load. And on medium loads (video encoding which is CPU only) it is still very silent thanks to the extra rads, which is the most important in regular use. The dual loop setup here was just to maintain enough flow (i have 1.4 Gpm / 310 lph in each loop) and convenience too, as changing my vid cards wont involve dismantling the motherboard loop, and tube routing is a lot easier.

[NaeKuh]

I've asked you several times to explain precisely what you mean with "holding capacity of water" because it does look like you are mixing things up.

Martin + Skinnee + Vapor + others i remember long ago HASHED this forumla out.


it translated to 300W per 1 gallon per minute flow in pure distilled h2o.
http://en.wikipedia.org/wiki/Heat_transfer_coefficient

From the above equation, the heat transfer coefficient is the proportional coefficient between the heat flux that is aheat flow per unit area, q/lund, and the thermodynamic driving force for the flow of heat (i.e., the temperature difference, ΔT).


if im lost somewhere please teach me stephan.. your one of the guys i know who knows LC and thermo.

But watercooling in its basics is..

Heat is being released in blocks... water picks up... carries.. and then dumps.
That means when water picks up heat, it picks up energy and MUST increase in temp.
That number of Energy is in a relationship between flow and medium that is being used.. ie.. distilled water.
Water is non compressible, so were not going have crazy values which compressed gasses like LN2 would have as there is no evap involved.

Saying water WONT go up 1 degree after touching a X amount of heat is breaking the laws of physics.

So how much of a gradient is acceptable? or better yet do you even have a gradient?
If your a new user.. typically your gradiant wont be larger then 1C.

If you have 3 580gtx under load, with a full board block with a cpu block and a ram block, and you watercool your aquero controlller along with an ARC-1680ix... well... as i said, you need to pay for somethings.

For the only reason that the motherboard full cover block is so restrictive, it kills the flow even with dual pumps. So i have dual pumps on the CPU+GPU loop, and a single pump on the motherboard loop.

yeah i wish people understood this..
the bridge on the nb -> Sb only allows so much without making it bulky.

Which is why the barb locations on most is made so you could connect a gpu to them without much trouble.

[MrBean]

For the only reason that the motherboard full cover block is so restrictive, it kills the flow even with dual pumps. So i have dual pumps on the CPU+GPU loop, and a single pump on the motherboard loop.

and

yeah i wish people understood this..
the bridge on the nb -> Sb only allows so much without making it bulky.

I have found on both my systems, that flow is not important - there is less than 1.0 deg C differnce, keeping all else constant, running the pumps at full tilt, vs 8.5 - 9.0 Vdc via the bigNG or TMS-200. This have the pumps very quiet, pretty low flow, but still doing a great job.

Let's look at this analogy -> take a standard air-cooled system, let it idle, and disconnect the fan. Keep your hand on the HS, and feel how quickly that warms up. Now, with a fan-controller, adjust the fan to be barely rotating, and see how quickly that wisks heat away - within seconds, the heatsink is cool to the touch.

With watercooling, this is even more efficient for low flow conditions, due to water's efficiency. Yes, super-high flow will hel a little, and high-flow mght be needed where you have blocks, like the Swiftech Storm of days gone by, that needed pressure to try and atomize the water.

In our modern systems, again, for the last 1-2% performance, that extra flow might help a little, but for else, not an issue. I have a 480 Feser on the Wife's rig, a single disconnected the one D5, then, Ek full-cover for the eVGA 4-way SLI board, EK on the 5990, and Swiftie GTZ on the CPU - pretty restrictive from the EK full-cover block, overclocked to past 4Ghz, and pump at low flow, fans low speed, and a very happy system.

But let me stop these long replies, I am sure they're tiresome to read, apologies for that.

Enjoy, gotta start rewiring the AX1200 with spaghetti noodles HP Blackbird is waiting.

[gmat]

With watercooling, this is even more efficient for low flow conditions, due to water's efficiency.

I dont understand that. Heat transfer is a direct function of flow, not an inverse one... More flow = more heat transfer.

In our modern systems, again, for the last 1-2% performance, that extra flow might help a little, but for else, not an issue.

It was an issue for me in my previous system (1 single loop for everything) where flow became so low that my GPU's over heated, even though i had enough radiators and excellent idle temps. But flow was so low in the parallel GPU blocks that they went up too high in temp.
Also i already have to use 2 pumps for my main loop. The mobo block with a single D5 already brings its flow down to 300 lph... (and there's nothing else but a radiator !). On my main loop there is already a lot of restriction (3 big radiators, a lot of tubing up & down the mountain mods case). If i disconnect one pump on my dual top the other cavitates and flow stops entirely.. So yeah, it's an issue for me. I better take no chances with that and have the critical 1GPM (or more)...

[paulbagz]

I have seen better temps going from dual loop to single loop with dual pumps and that is after adding a third GTX 480

-PB

[MrBean]

I dont understand that. Heat transfer is a direct function of flow, not an inverse one... More flow = more heat transfer.

Problem often is that people want to look at theoretical numbers only, vs practical experience - yes, theroretically, you will see an improvement, in practice, this is so small, it's hardly noticeable.

The second problem is selective reading -> please read my comments proper, and you will note this part:

I have found on both my systems, that flow is not important - there is less than 1.0 deg C difference, keeping all else constant, running the pumps at full tilt, vs 8.5 - 9.0 Vdc via the bigNG or TMS-200.

I did clearly state there was an improvement with higher flows, point is, it is not as significant as theory would like you to believe. Remember, the 1deg C I refer to here, is water-temp......

[MrBean]

One thing I would like to mention, too, is that, when I have a loop, where I'd prefer low-flow, i.e. run the pumps in silent mode, I like to add a few 90-deg bends, as they act as turbulators, breaking up the laminar flow normally associated with low-flow loops - and note, I am not referring here to low flow due to excessively restricting waterblocks, rather to conditions where you have a relative free-flowing loop.

In these instances, having a couple 90-deg fittings, normally gives me fractions of improvements in water-temp....

[gmat]

Well the difference for me between too low flow and decent flow is 45°C at full load with decent flow and over 85°C (then it went to crashing) with not enough flow. That's the GPU temp. Critical aspect here is "enough flow", as all practical measurements (what you'd call theory i suspect) show is indeed there are diminishing returns for high flows but under 1GPM (usually - rough figure) is where the efficiency curves starts to fall significantly. Having 2 separate loops in my case helped having enough flow in both loops, instead of "not enough" flow in one loop, and even afford to tune down the pumps a bit. Note that i'm talking about full load here, Furmark / Crysis / other intensive games (Metro 2033 was rather intense on hardware too i recall).
Also as i have several radiators in series in my main loop, i want to maximize the flow so i get the best deltaT for each radiator. Again, two loops helped a lot here by separating the motherboard radiator which receives its own air and water and is not affected by CPU or GPU activity.

[MrBean]

Point taken, gmat, thanx.

I also strive to keep around 1.1 - 1.2 gpm, pumps are quiet, and temps are really good - ar least in my system, no apprciable difference going from 1.2 to 2+ gpm, but there may be cases, like in yours, where higher flow is required.

This is what I luv about watercooling, and why I keep on telling people they will get the best answers by experimenting themselves - no 2 systems are the same

[stephenswiftech]

Martin + Skinnee + Vapor + others i remember long ago HASHED this forumla out.


it translated to 300W per 1 gallon per minute flow in pure distilled h2o.
http://en.wikipedia.org/wiki/Heat_transfer_coefficient

From the above equation, the heat transfer coefficient is the proportional coefficient between the heat flux that is aheat flow per unit area, q/lund, and the thermodynamic driving force for the flow of heat (i.e., the temperature difference, ΔT).


if im lost somewhere please teach me stephan.. your one of the guys i know who knows LC and thermo.

But watercooling in its basics is..

Heat is being released in blocks... water picks up... carries.. and then dumps.
That means when water picks up heat, it picks up energy and MUST increase in temp.
That number of Energy is in a relationship between flow and medium that is being used.. ie.. distilled water.
Water is non compressible, so were not going have crazy values which compressed gasses like LN2 would have as there is no evap involved.

Saying water WONT go up 1 degree after touching a X amount of heat is breaking the laws of physics.

So how much of a gradient is acceptable? or better yet do you even have a gradient?
If your a new user.. typically your gradiant wont be larger then 1C.

If you have 3 580gtx under load, with a full board block with a cpu block and a ram block, and you watercool your aquero controlller along with an ARC-1680ix... well... as i said, you need to pay for somethings.



yeah i wish people understood this..
the bridge on the nb -> Sb only allows so much without making it bulky.

Which is why the barb locations on most is made so you could connect a gpu to them without much trouble.

what exactly are you trying to show with this equation? "A" in your equation is the surface area on which you are calculating "h" - fluid properties have nothing to do with that. That doesn't explain where your "holding capacity of water" is coming from. This formula has nothing to do with the problem here.


When designing liquid cooling systems the only relevant equation would be: Q [W] = Flow [Lps] * Cp [J/Kg/K] * DT [K]

W is the amount of heat (in W)
Cp is the specific heat for Water (or Heat Capacity) and is equal to 4186 J/Kg/K.
Flow in Liter per second.
and DT is the difference in temperature between 2 points.

it only gives you the relation between the amount of Heat that is going through your liquid cooling system and the temperature delta between inlet and outlet.
you can use this formula to calculate the amount of heat that is going through any component of your loop: radiator, water block, etc (provided you precisely know your flow rate, and temperatures). Also note that formula is good way to decently approximate how much heat a CPU or GPU is putting into your LC system.

so if you take a CPU or GPU that puts out 200W - with a flow rate of 1 GPM (say 4LPM for easier calculations), your DT = 600 * 60 / (4 * 4186) = 2.15 C
Now say you run 4 of these: 3x GPU's and 1 CPU, that's 800W. Each of them increases the coolant temperature by 2.15C. let's keep things simple, say they're all in series, that's a total DT of 8.6 C.

Now we just gotta figure out what the coolant temperature is at the radiators outlet.

two MCR320-QP have a combined thermal resistance of 0.012 C/W at 4 LPM. For 800W heat load the coolant temperature will be 9.6C higher than ambient. If the ambient is 20 C, that's 29.6 C at the rad outlet (and at the radiators inlet you'll have 29.6 + 8.6 = 38.2 C).


See there is nothing here that remotely suggests a "limit" or even a sweet spot... If there is a limit, it is something you personally came up with based on your experience. But that is completely different than stating there is a limit and that is coming from the heat transfer formula you linked. And don't tell me water properties change with temperature because these variations exist but are irrelevant: Cp (I.E. water properties) is almost independent from temperature (4182 @ 20C ... to ... 4196 @ 80C) - so really water properties don't change - at least in the range that is usable by LC systems.



This was pretty much off topic... Anyways back on topic you disagree with the actual results shown in this thread - which is fine. But my problem is that you try to prove your point by stating there is a limit and that, basically single loop systems are going to be passed this limit and that is bad. Nope, there is no 300W limit with LC systems whether it's for computers, cars or anything. The only physical limit are coolant properties but here again they don't vary in any relevant way. In the 20 to 50C range of coolant temperatures (that probably covers 99% of the systems) it's all nice and linear: like you said xxx Watts = yyy C at zzz Lpm.

The only good point you make is bringing up the fact that Gabe's results prove single loops are the way to go for systems that take care of CPU and GPU when they are not all loaded at the same time. If we were to load up both CPU and GPUs all at the same time, then dual loops would probably end up each with a slightly higher flow rates, but not by much, which "could" result in slightly better temps but again not by much! Keep in mind that with the hardware to build 2 loops, you have at least 2 pumps and 2 radiators. There is no problem going in a CPU first then into a radiator, then into the GPUs and then into another radiator. This will smooth out the coolant temperature at the inlet of the blocks.

Also, unless you run 4 pumps, by going dual loops you will lose redundancy (as opposed to a single loop with 2 pumps in series). And anyway, who is loading CPU and GPUs at the same time? I am sure there are people who do that, I am one of them when I stressing our kits for pure testing purposes but is this the majority? I really don't think so.

[MrBean]

Good post Stephen - that's the same formula we used to use when I studied (and passed) for my DipTech in Instrumentation and Controls in the 80's.

And representative of my experience too, although, as I have stated, in practice, I see very little to no appreciable improvement in temps running pumps at 4.3 LPM vs 7.5 LPM - this as measured by the Koolance flowmeter, flows might not be 100% accurate, but for the purposes of the experiment, close enough....

[rge]

so if you take a CPU or GPU that puts out 200W - with a flow rate of 1 GPM (say 4LPM for easier calculations), your DT = 600 * 60 / (4 * 4186) = 2.15 C
Now say you run 4 of these: 3x GPU's and 1 CPU, that's 800W. Each of them increases the coolant temperature by 2.15C. let's keep things simple, say they're all in series, that's a total DT of 8.6 C.

unit conversion error?

Q watts (joules/second) = mdot * Cp*dt = 1 gal/min x 8.34/lbs/gal x (1min/60sec) x (.4536 kg/1lbs) x 4186 Joules/KG C x 1 C = 264W.

I get 264 Watts to heat water by 1C at 1 gal/min. For a delta of 2C, would take 528W

Just dont want people thinking that after 2 gpus = 4C difference, or will be constantly dealing with posts about loop order.

[stephenswiftech]

unit conversion error?

Q watts (joules/second) = mdot * Cp*dt = 1 gal/min x 8.34/lbs/gal x (1min/60sec) x (.4536 kg/1lbs) x 4186 Joules/KG C x 1 C = 264W.

I get 264 Watts to heat water by 1C at 1 gal/min. For a delta of 2C, would take 528W

Just dont want people thinking that after 2 gpus = 4C difference, or will be constantly dealing with posts about loop order.

Possibly, I made the approximation that 1gal = 4LPM (I used 4LPM in my post) and made the approximation that 1Kg = 1L
You get the idea

[stephenswiftech]

Good post Stephen - that's the same formula we used to use when I studied (and passed) for my DipTech in Instrumentation and Controls in the 80's.

And representative of my experience too, although, as I have stated, in practice, I see very little to no appreciable improvement in temps running pumps at 4.3 LPM vs 7.5 LPM - this as measured by the Koolance flowmeter, flows might not be 100% accurate, but for the purposes of the experiment, close enough....

Yes it is usually difficult to measure. With some lab equipment you have a slightly better resolution for this type of things where differences are small.
In general radiators will yield any visible improvement from 1 to 2gpm for example. Depending on the water block you are running you may start seeing some performance degradation as flow rate drops.

[OC Maximus]

Single Loop + Sufficient Raddage = WIN

No need to over complicate.

[RyNemesis]

Under typical computer use, the above test data suggests as a general rule that users would not benefit from setting up dedicated loops for CPU and GPU. Serializing pumps in the same loop also adds a redundancy factor that dedicated loops cannot provide. With superior reliability and lower temperatures at both CPU and GPU levels, single loops appear to win hands down.

Under extreme computer use, this setup recorded a notable advantage at the CPU temperature level for the dual loop, counterbalanced by the opposite effect at the GPU level. This extreme environment uncovered the critical importance of the respective load ratios generated by CPU class devices vs. GPU class devices, relative to the heat exchangers to which they are connected. Clearly, a CPU generating 150 Watts solely dedicated to a triple radiator will cool substantially better than when mixed with another 400 watts generated by two GPU's even with a second dual radiator in the loop.

Gabe with all due respect it appears your single loop apparently cures all that this thread has turned into, actually only applies to typical computer use mildly overclocked with a CPU and Single GPU in the loop and completely discounts the test numbers you ran yourself, in the first set of tests in your .pdf paper.

One single loop is based on the assumption all of us are satisfied with Typical Computer use, disregarding the fact that some of us water cool our CPUs for the maximum possible CPU overclock we can stably reach, some of us even went to water cooling specifically to cool the Sandy Bridge CPUs to be able to stably reach a 24/7 5ghz overclock.

Granted that's not typical computer use at all, I'll give you that!, but from my own experience, I could not reach a 5ghz CPU overclock running 2 full coverage overclocked 580GTX in SLI in a single loop, so for myself that's why I split my setup into 2 loops, and once the Sandy Bridge was on it's own loop, a 5ghz stable overclock was then attainable.

Your own first series of test result numbers plainly show a high CPU overclock and multiple GPUs put you at a disadvantage with your own i7-920 tested at 4095mhz a 1430mhz overclock, (Which was actually kinda on the tame side since some went way past that with air cooling), and I'll venture to say that with a 2nd,3rd,and 4th, generation Intel CPU a 1430mhz overclock from stock would show even higher numbers than you reached in your own tests.

This thread now seems flooded with single loop flag wavers completely discounting your own established test numbers, so shouldn't you make a distinction that this does not apply to those seeking as high a stable CPU overclock as they could possibly reach, especially with a Ivy Bridge or Haswell CPU, as these test numbers are already old information with an i7-920?

For the record we all know Ivy bridge and Haswell is an even hotter overclock than Sandy Bridge was, so for those truly seeking as high a CPU overclock as they can possibly stably reach using radiator water cooling, I am going to say combining that high CPU overclock with multiple GPUs in the same loop will limit the CPUs possible overclock.

Taking your own test numbers into account do you at least agree with that statement?

This is Xtreme Systems forum isn't it, maybe I'm in the wrong forum?

Thank you for your time and consideration. RyNemesis

[Church]

RyNemesis: On this my point was: if desired overclock couldn't be reached with extra 5C temps with gpus on same loop, i wouldn't use said overclock at all, as that in my eyes is not reliable/stable enough, with similarly loosing stability on eg. simple situation of raising ambient temps by same ammount.
P.S.
Ivy Bridge is not THAT hot .. once you replace it's flawed TIM1 beneath IHS.

[RyNemesis]

RyNemesis: On this my point was: if desired overclock couldn't be reached with extra 5C temps with gpus on same loop, i wouldn't use said overclock at all, as that in my eyes is not reliable/stable enough, with similarly loosing stability on eg. simple situation of raising ambient temps by same ammount.
P.S.
Ivy Bridge is not THAT hot .. once you replace it's flawed TIM1 beneath IHS.

Well that would be the time for a dual loop setup to isolate the GPU heat load from the CPU overclock, then that theoretical 5c would actually make a higher stable CPU overclock possible.

Theoretical meaning the Ivy is a hotter CPU overclocked than the i7-920 he ran his tests with, once you comparatively add a 1430mhz to the stock clock, that would put my 3770K at 4930mhz.

What are you running cooling specs and CPU overclock wise, and for the record my Ivy, 3770K has not been delidded.

I think some of the missing information is the relevance to the actual DeltaT an individual is satisfied running when higher CPU overclocks come into the picture.

http://www.overclockers.com/guide-deltat-water-cooling/

IMO dual independent loops make a lot of overclocked CPU performance reachable by isolating the CPU deltaT in it's own loop, thus allowing higher clocks.

I think Conundrum explains this very well in his thread.

[SNiiPE_DoGG]

If you are serious about overclocking the CPU, you make a CPU only loop with a 140.3, a 120.2 and dual mcp355's and aircool the gfx...

My cpu temps barely exceed 32C @ 4.6ghz

But seriously... my setup is overkill and was designed for 3x the thermal load I have it on.

[Johnny87au]

Nice testing indeed. . .but, all these results are with 355's. I'm curious what would happen to the results if 655's were used instead. There are those out there that flat out refuse to use DDC's for whatever silly reasons they believe in and this testing really doesn't help them.

lol dual loops just more hassle to be honest, always stuck with 1 loop but 2 pumps.. Thanks GABE