POST A COMMENT

107 Comments

Back to Article

  • hot120 - Monday, September 30, 2013 - link

    Awesome article! Reply
  • blanarahul - Monday, September 30, 2013 - link

    Hmm.. Can you try cooling those 780s alone? Overclocking the CPU seems pointless on Haswell. Reply
  • valkyrie743 - Monday, September 30, 2013 - link

    overclocking haswell is not pointless. just is a pain (same with ivy bridge) cause intel decided to be cheap and not solder the IHS to the cpu. if you do a mild overclock its fine give or take how bad the tim on the cpu/ihs is. but if you plan on doing high overclocks and water cooling like this. you might as well de-lid the cpu and apply your own tim. temps on air (if done right) drop a good 15 to 20C under load. I've seen people hitting 90 C and go down to 70 or less underload. and thats on air.

    the reason why i have no upgraded from my sandy bridge 2600K. @4.5ghz right now at 1.28 volts and my max temp running intel burn test was 70C (air)
    Reply
  • The Von Matrices - Monday, September 30, 2013 - link

    Please read my post in response to NeatOman. The result is correct but the reasoning is incorrect. Reply
  • gandergray - Tuesday, October 01, 2013 - link

    For information about removing the cpu lid or integrated heat spreader, see the work performed by Idontcare: http://forums.anandtech.com/showthread.php?t=22618... . Reply
  • katherine0james - Wednesday, October 02, 2013 - link

    my parents in-law recently got an awesome red Lincoln MKS Sedan just by part time work online. site here..... http://CuttR.it/tvtmbce Reply
  • iTzSnypah - Monday, September 30, 2013 - link

    You are cooling way too much with only 600mm worth of radiators and your deltaT is obscene. Take out 1x GTX780 and retest if possible. Reply
  • NeatOman - Monday, September 30, 2013 - link

    I think the thermal paste between the cpu and the lid are the limiting factor here, i believe that not only will 4770K do better with better thermal paste in between the lid and cpu on just air cooling alone but also might have a larger difference between the air and water cooling.

    And of course there is also a full delid which i think wont be much of a threat because with water cooling you don't need the motherboard to support a large heavy cooler.
    Reply
  • NeatOman - Monday, September 30, 2013 - link

    Sorry, i meant that you wont need to put a lot of pressure like if you where supporting a large air cooler with the motherboard. Reply
  • The Von Matrices - Monday, September 30, 2013 - link

    The issue is not the composition of the thermal paste between the die and the lid; it is the thickness of the thermal paste between the die and the lid. It's widely reported that in Ivy Bridge and Haswell there is way too much of a gap between the die and the lid due to the thickness of the glue used to secure the lid to the package. You can solve this by removing the lid, using a razor blade to remove all the glue, then put on new TIM and place the lid back on the package. No matter what new TIM you use you will get drastically reduced temperatures.

    Either way, Haswell runs hot due to its FIVR, and there's nothing that can be done through beefier heatsinks, delidding, or changing thermal paste that will make it cooler than an equivalently modified Ivy Bridge.
    Reply
  • dragosmp - Monday, September 30, 2013 - link

    Still, it is incomplete. The thermal transfer formula is simply Rth=rho*L/S, more thermal resistance (Rth) more the temperature delta is high between the source and ambient: deltaT=Power*Rth
    Asuming the power is constant, to decrease deltaT you need to decrease the thermal resistance, so:
    *S is the die surface, can't change that
    *L is the thickness of paste - you're right, it needs to be as thin as possible; put 2x too much and you have twice the deltaT
    *rho - thermal resistivity (1/lambda) - it depends on the material; Intel does use cheap paste with a conductivity around 3; were they to use fluxless solder or at least some AS5 they'd decrease the thermal resitance by a factor of 2 easily, thus offsetting a thicker than needed layer of paste.

    My 2 cents: for performance the paste must be removed and replaced with something better plus as you say remove the glue to reduce the thickness. Of course one should be careful not to chip the die, but these two things really help.
    Reply
  • Sadrak85 - Monday, September 30, 2013 - link

    And one more thing is the addition of onboard voltage regulation, which is a lossy process almost by definition. Meaning, even with the better TIM, I seriously have my doubts that the thermals could hit the level of IVB or SB. Unless Intel somehow has some magic way of using the voltage and amperage they're scrubbing off.

    *small note, a good motherboard input, in terms of electricity, would pass through the filters pretty cleanly, but because Intel sets the specifications for the input, I have my doubts that they require such a thing, since the feature they added to their chip was to save money for the motherboard vendor.
    Reply
  • leafonwind - Monday, September 30, 2013 - link

    The thermal interface material is minor compared to the interface distance. Thermal resistance is L/kA. Going from a millimeter of thermal paste to an 10 micron gap (typical of paste when applied correctly) will give a 50x improvement. The difference in k between a good thermal paste and a bad thermal paste is typically a 5x difference unless you get into exotic materials like cadmium. http://forums.anandtech.com/showthread.php?t=22618... Reply
  • gandergray - Tuesday, October 01, 2013 - link

    To bolster Von's point, see the work performed by Idontcare: http://forums.anandtech.com/showpost.php?p=3405318... . Reply
  • merikafyeah - Monday, September 30, 2013 - link

    Super tiny correction: While it is true that liquids draw away heat much better than air, one must be cautious not to mistake water as a good CONDUCTOR of heat, aka something that "transfers" heat very well. Water is in fact an INSULATOR of heat, aka something that "absorbs" heat very well. Reply
  • merikafyeah - Monday, September 30, 2013 - link

    Note wording on first page, third paragraph. Reply
  • ShieTar - Monday, September 30, 2013 - link

    Correct, but to be precise, neither air nor water will conduct heat quickly enough for PC cooling purposes, both are only used to absorb the heat before being transported away from the heat source.

    Which makes you wonder how a closed-loop, compressed air cooling system would fare against a water-cooling system. Heat capacity might still be lower for air than for water, even at increased pressures, but I assume that you can produce higher flow rates for a compressed gas than for a liquid. And you could use the required compressor in order to:
    1) Reduce the air temperature below room temperature before sending it to the heat sources.
    2) Increase radiator temperature over the CPU/GPU temperatures, thus achieving the same heat transfer with lower air flow rates through the radiator. Though temperatures above 100°C may be unsafe in a consumer device for several reasons.

    Does anybody know if such a system has been considered and tested anywhere?
    Reply
  • Death666Angel - Monday, September 30, 2013 - link

    Considered? Probably. Used? Not to my knowledge. If you have a compressor it makes more sense to cool the water used in the loop to just above freezing or even below freezing with the right additives. Of course, if you cool it that much, you have to worry about condensation, so most people I read about who use compressor cooling for their liquid (instead of large radiators) keep the water around room temperature and have the cooler in another room, to not be bothered by the noise.

    The stuff that is used to conduct heat away from the components inside the PC is the metal heatsink. In the case of pure air cooling you then push air through the metal heatsink fins. Because of the delta T you have the air warming up, the metal cooling and being able to absorb heat from the CPU/GPU etc. again. In case of water cooling, you have the water running through the heatsink (usually some very fine canals inside that increase surface and flow rate) which absorbs the heat from the heatsink and gets transported to (large) radiators where air is again pushed/pulled through the radiator fins in order to cool it.
    Reply
  • Sadrak85 - Monday, September 30, 2013 - link

    Used all the time; Nitrogen is the most common component of air; it is compressed so much as to become a liquid. Then, thanks to the Carnot cycle, cooling the liquid to room temperature results in it boiling and becoming ultra-cold air, which cools a processor.

    A similar thing happens with your refrigerator.

    These coolers, however, require massive power to get them to that level, so they're only really useful for very niche-applications, but the equipment isn't really that hard to find. An evaporator will cost you something like $200 to $300, and then the Nitrogen.

    Now, if you're talking about keeping the air gaseous, then what you'll find is it just isn't possible. Cooling it very much with pressure on it will result in it condensing to liquid. If you just compress it, without the cooling, you'll heat it up, of course, which is how your diesel engine works.
    Reply
  • ShieTar - Monday, September 30, 2013 - link

    Fair enough. I am fully aware of the cooling concept via liquid nitrogen boiling itself, but I was considering a much simpler concept. Maybe I should describe it in a bit more detail.

    Imagine a closed air (or just nitrogen) system where the air pressure is about 3 bar within a radiator and about 2 bar when it circulates within the cooling blocks. You can have temperatures around 200K at 2 bar without liquifying, and not that much higher at 3 bar.

    So you offer your GPU/CPU coolers 2bars of air at 200K, maybe heat it to 220K, compress it to 3bar/330K, cool it back down to 300K (close to room temperature), decompress back to 2bar/200K.

    What needs a little more math is, just how much volume of gas do I need for this to transport 600W or so of power by this concept. And how much additional energy do I waste on the compression process. And probably, just how horribly noisy will this setup get with 2bars of air at high velocities getting pressed through the cooling blocks at high velocities.

    Yeah, the more I think about it, the worse the whole concept sounds. Nevermind it.
    Reply
  • UltraWide - Monday, September 30, 2013 - link

    Excellent article, I enjoyed reading this journey into water cooling. Keep up the great work! Reply
  • bojaka - Monday, September 30, 2013 - link

    And right about THERE I lost my interest in water cooling - Thanks :) I actually believed water cooling would be a viable solution to make my computer silent, but apparently not. Thanks for a great article!! Reply
  • utnorris - Monday, September 30, 2013 - link

    You can make a near dead silent water cooled system that allows you to overclock. Unfortunately, the fans you are using in the article, while good, are not the best, regardless what marketing says. I have built probably 40 to 50 systems over the last 5 years using water cooling and noise was never an issue. If you are not overclocking then a full blown water cooling setup is not worth it, but for the added performance, it's well worth it. Also, there is no need for additives like the Hydrx, distilled water with a silver coil or some anti algae drops from your local pet store is all you need and would be less toxic. Last, as already mentioned, delidding would have seen about 10c drop on the processor and if you are going to water cool, you might as well delid your chip. Reply
  • Aikouka - Monday, September 30, 2013 - link

    You REALLY shouldn't take a SINGLE article on water cooling as the end all, be all on whether water cooling can provide a silent build. Dustin was NOT pushing for a silent build, but rather just a build with whatever parts he was provided. I have a custom loop with an i7-4770k and two GTX 680s, and it isn't noisy at all. It is going to be a bit noisier at idle because of having significantly more fans (I have 7 120mm fans just on the radiators), but it doesn't get any noisier at load! I also use a fan controller that lets me set the fans to the speed -- and consequently, the noise level -- that I find acceptable. Reply
  • piroroadkill - Monday, September 30, 2013 - link

    Yeah, some people think this, but since you're shifting water at the same time as pushing air, there are simply more moving parts involved. Just get a giant tower heatsink and run those fans sloooowly, and choose a GPU with a really good factory heatsink. Reply
  • tim851 - Monday, September 30, 2013 - link

    Depends. There are much better, i.e. quieter, pumps available. The best one imo is the Aquastream by AquaComputer. Even then, a single Cpu-single Gpu system is quieter on air. A Thermalright HR-02 and an Accelero Xtreme will take care of it. Once you go SLI, however, you're running into space and heat evacuation problems. Watercooling is your only chance of acceptable noise at load now. Reply
  • piroroadkill - Monday, September 30, 2013 - link

    Yeah, multi-GPU rigs are a different beast. At that point a custom water loop is probably your best bet.
    But for single CPU/single GPU rigs, water is basically a vanity project, not one that will actually give you lower noise.
    Reply
  • piklar - Monday, September 30, 2013 - link

    Great article Dustin! It just so happens I have been considering a watercooling project this summer (new Zealand time) using the Corsair Air540 with 4770K and SLI GTX 780s. Ironicly it appears The Raven RV03 is doing just as good if not better job of the cooling with Corsair H80 and reference card cooling. You have shown how luck of the draw Haswell can be since the 4770Ks Ive come across all do 4.6ghz on 2.8- 3.0v no probs. Still your article was very useful and much appreciated. Reply
  • Death666Angel - Monday, September 30, 2013 - link

    2.8V to 3.0V? That's not right. :P Reply
  • Razorbak86 - Monday, September 30, 2013 - link

    LOL. My thought exactly. I'm sure he meant 1.28-1.30V. ;-P Reply
  • piklar - Monday, September 30, 2013 - link

    soz was half asleep, thanks for clarifying that I meant meant 1.280 - 1.30V for 4.6ghz with 4770K with H80. Reply
  • Razorbak86 - Wednesday, October 02, 2013 - link

    No problem. I'm pretty familiar with the voltage range, so I understood what you meant. I have a de-lidded 3770K OCed to 4.8 Ghz (offset) under water, and my voltage maxes out at 1.40V at load. Reply
  • Gigaplex - Monday, September 30, 2013 - link

    The power/temp/sound results aren't particularly useful without a comparison of OC air vs OC liquid at the same OC levels. We can't tell if the CPU is hotter on liquid vs air due to higher voltages or the GPUs heating the liquid. Reply
  • prophet001 - Monday, September 30, 2013 - link

    I thought this too. It would be nice to see metrics for identical OC settings on air and liquid. Seems like that would give more of an indication about how the cooling setup itself is performing. Reply
  • The Von Matrices - Monday, September 30, 2013 - link

    I enjoyed reading your article. I have owned a custom loop for three years, and after recently upgrading to Haswell, I can say that my overclocking experience mirrors yours. There is just no advantage to using a custom liquid cooling system versus a good air cooler or closed-loop liquid cooling system.

    I have also had the same experience with PWM splitters, and I think the blame comes down to the splitter itself. The splitter I have (it looks very similar to the one you tested) doesn't have the pins fitting very tightly in their connectors. Frequently, inserting the female connector into the male socket will result in the pins of one or the other pushing themselves out of the connector instead of mating with the other side, and you have to push the pins (or wires if the pins are just slightly out) back into the connector in order to establish the electrical connection.

    Two note as well is that are is worth mentioning: First, all video card full cover water blocks are designed with a particular PCB layout in mind, usually the reference PCB, so you have to be very careful which video cards you buy since they are not all compatible with the full cover blocks available. For example, for about the past six months there have been no more reference design AMD HD 7970s available on the market, so if you wanted to water cool the new 7970 you bought then you were out of luck (unless you wanted a MSI lightning, which one or two manufacturers make water blocks for). Second, while clear tubing looks great at first, it usually does not age well and begins to either turn yellowish or show water cooling fluid as it gradually gets dirtier over time. i would recommend solid color tubing if you want a system to look good with minimal maintenance.

    The biggest issue with this article is that you never stated the costs of the water cooling components. The system you built I estimate has about $800-$1000 in water cooling components. It is important to note that the system does not have to be that expensive, particularly if you're not especially concerned with aesthetics. Barbs cost about half the price of compression connectors, and if you have 10 of them in the system that's about $100 saved right there. If you frequently upgrade graphics cards, you can use core-only water blocks and separate VRM and memory heatsinks. The initial cost is about the same as a full cover water block (~$120 for a good full-cover block versus $80 for a MCW82 + $40 for the VRM/RAM heatsink plate) but every time you replace the graphics card you only have to replace a $40 VRM/RAM heatsink plate instead of a $120 full cover water block. And you can eliminate all the quick disconnect fittings for another $80 savings.

    I appreciate you documenting your experience!
    Reply
  • DanNeely - Monday, September 30, 2013 - link

    Unfortunately Swiftech has stopped selling monolithic plates for vram/vrm cooling on nVidia cards (allegedly due to their deal to make OEM watercoolers for eVGA). They still sell them for AMD cards; but on the nVidia front you're back to taping/gluing individual sinks onto each component. Reply
  • r3loaded - Monday, September 30, 2013 - link

    I enjoyed reading this article, but I'd add that you'll see a huge drop in CPU temperatures (like, 20 degrees) if you're able to delid your 4770K and apply a good quality paste under the heatspreader. It's risky but worth it if you're careful.

    If only Intel hadn't decided to cheap out and abandon soldered heatspreaders as on Sandy Bridge and earlier.
    Reply
  • kasakka - Monday, September 30, 2013 - link

    Those load temps seem awfully high for the CPU. I can't help but think that maybe the CPU waterblock isn't properly mounted, has too much or too little TIM between it etc. Back when I was overclocking a watercooled C2Q 9550 the idle and load temps didn't change much and were around 40-50 degrees. Surely the current Intels would provide better temperatures?

    I stopped watercooling mainly because it would've been expensive to upgrade all the blocks when I went to Ivy Bridge. Turns out there is barely any noise difference between the water cooling and today's best air coolers, meaning I'm very happy with the near-silent setup I have now (3570K @ 4.5 GHz, Phanteks PH-TC14PE, Asus GTX660Ti w/ stock cooler) so purely from a noise level point of view water cooling isn't necessary anymore.
    Reply
  • Jackie60 - Monday, September 30, 2013 - link

    Well that was a very informative monumental waste of time and effort and would have been money too if it wasn't for Anandtech. This tells me don't even ever think about water cooling unless in a small prison cell for a long time. Reply
  • Death666Angel - Monday, September 30, 2013 - link

    I'm glad you finally tackled custom water cooling in an article here on Anandtech! :D And for the first one, it wasn't half bad.
    There are a few things I would add though. It is highly unlikely in my experience that someone investing the time and money into a custom water cooling rig would not delid the CPU or use a IVB-E CPU. That should help increase OC headroom quite a bit and bring down temperatures. That doesn't mean that it can't be done with air cooling, but since that is infinitely easier to do, fewer people would invest the time/money to delid a high price CPU or get a high price platform like IVB-E.
    But then again, watercooling just for the CPU has been dead since we stopped having 100+W CPUs. If you ask in any WC forum whether to custom cool the CPU, you will always get the response "It's not worth it just for that".
    I also think that people have a wrong idea about custom WC. If you are not interested in the challenge it poses, then it isn't for you. The benefits (quieter running while providing better performance) are there, but they aren't great enough to change from air cooling if you don't enjoy the risk and the building and selection process.
    If you plot a chart with performance/noise ratio, the water cooling rig can always be better than any air cooling that is available. It can also run components cooler at the same overclock, which increases their potential life span. However, these things come at a steep cost and may not be as necessary for many.
    But it is very nice to have my AMD 7970 run at 1250/1650 at ~44°C and barely have any high noise than in idle. :D
    Reply
  • Gigaplex - Monday, September 30, 2013 - link

    IVB-E are 100+W CPUs. Reply
  • C'DaleRider - Monday, September 30, 2013 - link

    [q]...the HydrX PM is actually 90% distilled water (which contains no impurities) and a mixture of chemicals to prevent algae and corrosion. This is very similar to how coolant/anti-freeze works in your car; water carries heat very well, but is also corrosive, so chemicals are introduced into the mixture to counteract that effect.[/q]

    While the above is technically correct, the reasons auto manufacturers use glycol derivatives for coolant is because using dissimilar metals within a cooling system causes galvanic corrosion (in autos, aluminum radiators used, combined with the steel, iron and other metals in the cooling loop, causes a mild electrical charge to develop in the coolant, something glycol helps combat), something a water cooling loop(s) should not encounter given almost all radiators are copper/brass based as well as the cooling block(s) you'll use. Distilled water in and of itself isn't particularly corrosive when used in a cooling loop, as experience has shown by hundreds of people using pure distilled water. Go to XtremeSystems, for one place, and begin to read.

    Another reason glycol is used is its ability to help raise the boiling point of water under pressure. Not exactly what you find in a computer water cooling loop....really doubt you see a 16psi pressure in a loop like you see in an automobile cooling system.

    And, yes, I understand Swiftech will "void" their warranty on their products if not using their HydrX crap, but in reality, that cannot be enforced......the Magnuson–Moss Warranty Act, while primarily being directed at vehicles, could be successfully enforced in this instance, meaning Swiftech would have to provide the HydrX crap for free if required for their warranty to remain in effect. Otherwise, it'd be like Toyota or GM or Ford requiring the owner of one of their vehicles to only use the manufacturer branded oil to keep their vehicle's warranty intact, something that's illegal....and so is Swiftech's requirement to use HydrX to keep their cpu/gpu blocks and radiator warranties intact.

    And those fans.....such fail. Get some GT's and leave that crap from Swiftech behind. Find out what low speed fans with real performance is all about.
    Reply
  • Death666Angel - Monday, September 30, 2013 - link

    "something a water cooling loop(s) should not encounter given almost all radiators are copper/brass based as well as the cooling block(s) you'll use."
    Yes, but I haven't heard of any detrimental effects using these chemicals, so why take the chance? And in any case, you'd need biocide anyway, this way you have a 2 in 1 solution.
    Reply
  • DanNeely - Monday, September 30, 2013 - link

    Something in the anti-freeze I used in the loop I setup when I built an LGA1366 box caused the tygon tubing to slowly break down and after about 4 years crack drowning my GPU in the process. When I took it apart I found the pins on my CPU block were completely clogged with tubing fragments (explaining the several C temp increase I'd been wondering about). Reply
  • Aikouka - Monday, September 30, 2013 - link

    The usual reason given to not use an anti-corrosive agent in your loop is because it reduces the effective cooling potential since you're replacing water (what actually removes the heat) with an additive. Dustin stated that HydrX is 90% water, which means 10% is anti-corrosives, biocide and green dye. I mentioned in another comment that I use distilled water and two forms of biocide. The Dead Water biocide that I use is one drop per liter!

    So, why don't I worry about anti-corrosives? Because it's a custom loop and I know what went in it! I know that the actual flow portions of my loop are only made out of copper or brass (brass is copper + zinc). Some parts do use aluminum for construction purposes.
    Reply
  • Adamantine - Monday, September 30, 2013 - link

    Your article has a couple significant problems with the cooling setup. I understand some of it could be due to being given the parts free for review/article purposes, but I am going to ignore that for the most part and critique the article for improvements.

    First is the fans. While the fans provided will get the job done, there are basically two categories of fans, free air case fans and heatsink/radiator fans. Free air case fans perform terribly when mounted on an HS or rad due to a lack of static pressure (combination of fan blade design and fan motor). Of the radiator fans, the metric you want to use as the comparison tool is cfm per db ratio when mounted on a rad. When using this metric as the benchmark comparison, there are basically two groups of fans: Gentle Typhoons and everyone else. If you're using a fan controller (which is almost required), then the ideal model would most likely be the Scythe Gentle Typhoon AP45 (2150rpm). On my Lamptron FC5 fan controller the AP45's can get down to about 500/600 rpms or so. With 10 fans in push/pull (360+240 rad), switching from Scythe S-Flex (1900rpms) to the AP45, was about the equivalent of turning off 4 fans (one radiator) while at max speed. Higher rpm and still getting lower noise.

    Second, fan orientation. You don’t mount the fans on the rads like you’re doing air cooling which takes into consideration balancing intakes vs exhaust and/or directing air flow path(s). All fans mounted on radiators, when doing an all internal cooling loop, need to be mounted as intakes unless you don’t care about cooling performance. If you don’t care about cooling performance, you shouldn’t be doing a water cooling setup in the first place as better cooling is the ENTIRE point. Radiator performance is directly affected by intake air temperature and any increase in intake temps will reduce radiator efficiency. All of the air being pushed inside the case will be forced out whatever exit is available and the highly unbalanced air pressure has a side effect of being able to give just enough air to most of the remaining components that might get warm. They’ll probably be a few degrees higher than pure air cooling with balanced air flow, but you’re not pure air cooling so it doesn’t really matter unless the components are hotter than what they are spec’d for, or if it affects the overclock and/or stability. At max fan speeds, probably not an issue.

    You also picked a bad choice for a case. If it was the only one you had around, then I guess it’s what you HAVE to work with. However, you just recently reviewed the Phanteks Enthoo Primo. If you had to send it back, then this critique point isn’t that important. However, you could/should have mentioned that or why you choose the 540 over the Primo. The Primo, once it actually ships, will be THE best water cooling case on the market for the money (~$300 or less). Better than the venerable TJ07, Cosmos/Cosmos S, or newcomer 900D.

    Some people probably think the water blocks are the most import parts in the loop, but in the current market, most of them perform fairly similar, so it’s probably more about looks, and radiator choice is more about what rpms you’re going to be running your Gentle Typhoons at. Pump choice is pretty much down to which (Rebranded) Laing model you’re going to use and how many.
    Reply
  • superericla - Monday, September 30, 2013 - link

    On a similar note, the orientation of the front radiator makes bleeding out air a pain. It's extremely likely that there's an air pocket in that radiator causing it to do almost none of the cooling if it wasn't bled properly. There is a lot I would have done very differently if I were the person writing this article, but years of water cooling experience in everything from a test bench to a CaseLabs case tends to help me with that. Reply
  • DanNeely - Monday, September 30, 2013 - link

    An FYI for anyone thinking about installing a rad in that orientation. Either bleed the loop before installing the rad so it can sit flatish on the table (barb end higher), or unscrew the pump/res so you can tip the case on its front again putting the rad flat while rotating the pump/res combo enough to keep the res above it to assure inflow. Reply
  • cbgoding - Monday, September 30, 2013 - link

    Since you covered fan orientation and speed, there's two other factors that make a great deal in improving noise levels:

    Decouple your pump. If the bottom of it touches the bottom of your case, it's going to hum. 2 Good methods to fix this are: hang it off of the top radiator or make a shoggy sandwich (google it) between the pump and the bottom of the case.

    Fix your PWM settings. An mcp35x has its lowest speed at 20% PWM, and it's highest speed at 80%. To ensure that the pump never spins up at idle, set 50C as the low bounds at 20%, and have it on a linear scale to 75C at 100%.

    Finally, regarding fans, if ball bearing noise is something that bothers you at low rpm, then yate loon d12sl12's at 5v or 7v are inaudible. They push less air than GT's, but if you have enough radiator, it doesn't matter.
    Reply
  • cbgoding - Monday, September 30, 2013 - link

    Oh, and if you do go the yate loon route, top off the sleeve bearings with lube, the factory isn't reliable about giving you enough. There's a good post on how to do that on overclockers.com Reply
  • Death666Angel - Monday, September 30, 2013 - link

    Wait, you want to regulate your pump depending on the (CPU, GPU or water) temperature? No one does that, that is absolutely useless. 60L/h flow rate have been shown to be enough, above that you get a maximum of a few K better cooling performance. I run my system at 85L/h because that is the ideal point of noise, below that and I get case resonance and above that and it is simply too loud on its own. I have tested it at max with 120L/h and saw no noticeable gains in cooling performance. Set you pump once and then let it be. Make sure you have a water thermometer and see if you can set your fan speeds according to temperature. But let your pump in peace. :P Reply
  • cbgoding - Monday, September 30, 2013 - link

    It's a PWM controlled pump that's advertised to be able to scale speeds with temperature. I'd say at least a few people do it. "Absolutely useless" is a bit of a stretch, you can have it barely spinning along at 1200 rpm at idle, and be able to really crank the pressure when it counts. Some blocks, notably all of Swiftech's CPU blocks, respond well to increased pressure due to their absurd internal pressure drop. Now if you were using a Raystorm or HF or something similar, no, just leave it at the lowest setting and forget it. Reply
  • mum1989 - Monday, September 30, 2013 - link

    sorry, but i'ts a big fail ! :
    http://images.anandtech.com/graphs/graph7363/58372...

    Noise levels are worse with Liquid Cooling.
    Reply
  • ShieTar - Tuesday, October 01, 2013 - link

    Yes. As anybody who ever built a liquid cooling system can explain to you, and as is clearly described in the article itself, the pump operation means that you always have a higher idle noise with a Liquid Cooling system.

    Not saying it can't be improved upon by picking a different pump, actively controlling it so it won't run at full speed all the time, and maybe placing it in a sound-damped compartment within your case.

    Of course, the air flow setup looks excellent for the air cooling case, but far less than optimal for the liquid case. I mean, first off you have the two radiators basically in series, the top radiator will have to handle the air which has been already heated by the front radiator and additionally by the components within the case. That will be much less efficient than having 2 radiators in parallel, each with his own supply of fresh air. On top of that, there seems to be a rear exhaust fan installed which pulls out the same air pocket which the top radiator is trying to feed off. Given that the rear fan has no added resistance of a radiator, it will probably completely starve off the top radiator.

    So yeah, I think a lot of steps could be improved about the setup described above, but calling it a "big fail" is just excessive. And most certainly, even the perfect water cooling setup will not show lower idle noises than the equally perfected air cooling system. Load: sure, Idle: never.
    Reply
  • malkolm - Monday, September 30, 2013 - link

    Nice article, thanks for the work and effort you put into this.

    As a few of the commenters here pointed out this is only one article with one cunclusion and one can easily find hundreds well written articles elsewhere.
    I builded like ~100 custom-watercooled systems in the past 12 years as liquid cooling here in germany is by far more a topic than in the US. A reason for this might be that the sensitivity to noise and the corresponding need for a very silent (to absolutely inaudible!) consumer systems evolved earlier due to the lack of noisy air conditioning. Whatever...

    What i want to point out is: Watercooling IS the only way to make a hardcore gaming system absolutely silent. The one i use atm consists of a i7-3930K and 2xGTX580 what maxes out to ~800Watts of thermal power while overclocked. So far i didn't figure out how it could be possible to cool such a monster reliable and very silent at the same time only with aircooling. I say its impossible, but maybe some might tell me how.

    "Watercooling" itself is quite a lie, because in the end its again air that interacts as the media between the hot system on the one hand and the cool environment on the other. What the water in the watercooling really does is to overcome the limited spacial offering right inside the machine to actually cool. Esentially the water only transports the heat to a place where it can be cooled much more efficiently: To the border of your case or even out of it (using an external radiator).

    Dustin you're right pointing out that you essentially have to add a pump to the system as a noise contributor when using watercooling, but you're wrong with the assumption that this automatically leads to additional noise. Of course youre screwed with a big pump like a laing D5, but there are a lot of pumps out there that are designed not to be heared in a completely silent environment.
    Back ~15years as the whole pc watercooling stuff started people were looking for a decent pump. What they ended up with were waterpumps designed for aquaria in livingrooms. Esentially one of the best available waterpumps nowadays are (modded) waterpumps from Eheim: A specialist for aquaria ;)
    Derivates from this (like the Aquacomputer Aquastream XT) can't be heard outside of the case. In a 0.5m distance apart my noisemeter shows 0.0sone , so i would call this inaudible.

    As for the radiators, what you need ist area. The only thing better then lots of cooling area is...even more cooling area. As a rule of thumb systems i sell as silent have !at least! (100mm)² radiator area per 100 Watts of expected maximum thermal output.
    Cooling area is the only real advantage of watercooling over conventional aircooling, so use it! For a conventional single CPU / single GPU configuration a few radiators mounted to your case are enough. For a system like you use here in this article i would strongly recommend an external radiator (like the Watercool MoRa3).

    Considering these things (pump and area) everything else than what you found out and concluded would have been magic. If you REALLY want to go for silence one has to rebuild the cooling parts.

    At the end a few sidenotes regarding watercooling:
    -> watercooling helps save energy! each of my GTX580 uses about 40Watts less energy when used in my watercooled system (approx. 40-45°C GPU-temperature) than aircooled (80+°C). This more than makes up for the lousy ~5W needed for a waterpump.

    -> You learn to allways cary a pipe wrench in your emergency bag for pc repairing;)

    -> External cooling with an external radiator opens the field for put every(! except for optical drives if you need one) component with moveable parts away from your computer, your desk or even the same room. Once the tube system is completely filled one single small pump can drive meters of tube and many liters of water. In summer my radiator is planted outside my house in a shadowy corner so my room doesnt get heated by my computer at all.
    Reply
  • malkolm - Monday, September 30, 2013 - link

    Oh and what i forgot to mention and comes into my mind while reading the comments and complains about no fullcover parts for nvidia and so on:

    The hardware situation in the EU seems to be completely differnt from the US. Here we have plenty of companies in the small market of watercooling. A brief look at newegg forced me to post this link: http://www.aquatuning.de/index.php/language/en
    Not for advertisement, but only to show some of you what you really CAN buy, provided you know where to look at it.
    Reply
  • Death666Angel - Monday, September 30, 2013 - link

    Phobya DC water pumps are also great, cheap-ish, powerful enough for most normal systems (CPU+ dual GPU), small and quiet. I would never use a Laing without some serious noise dampening involved. Reply
  • mapesdhs - Monday, September 30, 2013 - link


    Given the level of experience you've referred to, I would be most interested if you could
    write an article about these issues, covering the additional aspects of watercooled
    builds you've mentioned which are not dealt with in the article, such as external units, etc.

    As it happens, I'm in the process of building a 3930K setup with four 580s (for AE), only
    air cooled atm, but I was thinking about switching to a watercooled config next year, mainly
    for the GPUs (the enormous Phanteks cooler already does a good job of running the CPU
    at 4.7). So, would you be up for writing some kind of article? I'd be happy to include it on
    my site, full credit, etc., or maybe Anand would be interested?

    Dustin has done a nice job of writing an introduction to this field, but there are always so
    many more questions someone new to watercooling will want to ask. Starting off with a
    closed loop kit isn't complicated (I've built two with H100s) but moving to a custom loop
    can be rather daunting. The number of 'additional' relevant points made by numerous
    posters shows just how broad this subject is.

    Btw, re choice of CPU, once again I'm ever more convinced that a used 2700K is still
    a better buy. :D Last one I bought (150 UKP on eBay) took mere minutes to get it
    going at 5.0 with an old used Venomous-X on an ASUS M4E. Delidding is not for
    the faint of heart...

    Ian.

    PS. A heartful thanks to all those posting comments! When it comes to sensible/useful
    comments worth reading (as opposed to the usual CPU/GPU flame wars), this has to
    be one of the best comment threads I've come across. Hmm, perhaps those who end
    up meddling with watercooling are just more sensible in general? :)
    Reply
  • mapesdhs - Monday, September 30, 2013 - link


    NB: I was referring to malkolm.
    Reply
  • cjs150 - Monday, September 30, 2013 - link

    A few points about water-cooling

    1. It is fun. You get your hands dirty (or wet!) in actually building what you want not want the manufacturer wanted for the price.

    2. It is not cheap.

    3. Pick your case carefully, look at forums for what others have done.

    4. A water cooled system is spot cooling: ie it cools only those spots where you have a water block. Air cooling is more general - as long as you have air movement you have a cooling effect. It is vital you have some air movement to cool things such as memory, hard drives.

    5. Custom loops should be designed with both air bleeding and drainage in mind.

    6. If you have a powerful GPU than a water cooled system will be a lot quieter than running on air.

    7. A dremel is your friend !
    Reply
  • willis936 - Monday, September 30, 2013 - link

    Great article. It's a very rational look at things that are all too often romanticized. Everyone wants that 5GHz 4770k and a fancy water cooling kit but it's important to stop and ask yourself what you're really interested in. While I'd love to see an article analyzing everything about delidding from when it started, the myths, the logic, and of course lots and lots of testing, I doubt I'll ever see it done because it comes close to stepping on some political toes.

    Personally on air I could hit 4.2 @ 1.2v on a 4770k before running into thermal boundaries and a delid got me to 4.5 @ 1.3v with stress test temps in the 70s. To get to 4.6 stable I need literally at least 1.45v which results in thermal boundaries. Even if a delid with a custom water loop would keep that temp under control it's a long ways to go for a tiny bit. Money would be better spent on a binned chip that could hit 4.8 on air. That's the nature of the beast and thank you for addressing it and giving people a realistic look at what haswell OC is like. All of the reviews always talk about engineering samples which doesn't give a down to earth idea of what a retail sample will be like.
    Reply
  • St. George - Tuesday, October 15, 2013 - link

    If CPU temps can be controlled with good cooling, and if I am after max CPU clock, is there a max voltage for Haswell? ASUS says 1.35v, various threads suggest between 1.4 and 1.5v, but many others seem to suggest max voltage is only limited by how much heat you can take away. What am I missing? You seem to suggest that the reason you won't run at 4.7Ghz is because you pass from what feels safe, 1.3V, to what feels risky, 1.45V. I am running archi apps that need CPU clock. Thanks. Delidding netted me 31C. Reply
  • spidey81 - Monday, September 30, 2013 - link

    I appreciate immensely the effort and thought put into this article. But to have someone new to the watercooling experience doing this wasn't the best way to go about it.

    For instance, and I'm sure this has been brought up, you compared results from different overclocks like it was apples to apples. Also, the radiators and blocks you used (although I have and use swiftech components) are inferior compared to other options available. Here's ( http://goo.gl/YJ8f13) a titan/780 waterblock comparison that clearly shows not all waterblocks are created equal. In fact, the swiftech block was one of the poorer performers in the roundup.

    Although good for basic loops, the radiators used could have been seriously beefed up! There are plenty of thicker rads available that use slower fans and still outperform these. Using slower fans will obviously create a more silent build. I've been using RX series rads from XSPC and they are always near the top for most efficient and pure performance.

    Basically, what I'm saying is that watercooling isn't strictly slap a block on and "voila" it's a super overclocking silent beast. The component selection from blocks to pumps to tubing to fittings are all critical in meeting the goal of the build. Research goes miles and finding good resources in overclocking/watercooling communities are a must.

    I was excited to see this article and left disappointed after reading it. Watercooling is far superior to air cooling in nearly every instance. Silence and performance are there to be had, you just need to have the proper selection of components and the subsequent proper arrangement of them.
    Reply
  • mapesdhs - Monday, September 30, 2013 - link


    I understand what you mean (paragraph 1), but in a way it's very useful, as it shows exactly
    the kind of issues someone new to watercooling is likely to encounter. Most guides and
    forums tend to assume a degree of existing familiarity. There's an information gap between
    being comfortable with top-end air cooling and dealing with reasonable custom water loops.
    As a newcomer to watercooling, half the problem is knowing what questions to ask. Forum
    threads can help, but they're frequently hindered by the odd one or two people who are
    very vocal but incorrect in their opinions, and of course it can be tedious in the extreme to
    wade through hundreds of posts to find all the various relevant bits of info (I know, I've done
    this with 3930K oc'ing).

    Your comment about component selection being 'critical' is a good example of what I mean.
    Unless one already has some experience with this, how does one make the right choices?

    Thus, I can understand why those already familiar with watercooling (especially someone
    with considerable knowledge/skill) might be disappointed with the article, but for beginners
    it will definitely be useful, ie. discovering potential pitfalls and mistakes.

    A parallel article by someone with more experience but using the same CPU/GPUs/mbd/etc.
    would be fascinating.

    Ian.
    Reply
  • Menetlaus - Monday, September 30, 2013 - link

    I would also love to know what the retail cost of the watercooling parts/kit used in this build.

    If the cost is $800-1000 as estimated above, then the overall cost for the full system would be in the $4k range.

    Bang for the buck is terrible on a full, custom, watercooling setup as the $4k watercooling system was typically less than 10% faster for 33% more money.
    Reply
  • The Von Matrices - Monday, September 30, 2013 - link

    I'd say your estimate on the entire system cost is about right. My guess is that Switftech sponsored this article and the author was either barred from stating the price or not curious enough to price out the components he was given. Custom water loops are a pretty terrible value if all you want is increased performance. Even though there are a lot of people who praise them, few who build them really consider the extreme cost since it's easy to get hooked into assembling it and forget about the price. Custom liquid cooling should only be used in three situations:

    1) You want the absolute best performance you can get at the current time.
    2) You want an extremely powerful but quiet system (mid-range systems can be quieted much less expensively with better air cooling).
    3) You care about aesthetics and showing off your case (the "cool factor")

    If you're only looking for increased performance, the best option to save money is just to forego the custom liquid system and use the money to upgrade the system's components more often. That will provide more consistent performance increases for the same amount of expenditure. The quietness argument is only valid if you have multiple video cards; any system with a single video card can be quietly cooled by replacing the stock video heatsink and using a better CPU heatsink and system fans. The aesthetic argument can't be reasoned with because that's a personal judgment, but generally people who use custom liquid loops for aesthetic purposes aren't particularly concerned about the price.
    Reply
  • sunnyjuicedrink - Monday, September 30, 2013 - link

    Thank you for the great article. I have one question that many people seem not to address in water cooling guides: how do you take a build apart? I've never built a water cooling system before, but I like keeping my PCs for a long time and upgrading individual parts. Is this possible, or is the deconstruction process cumbersome? Reply
  • cjs150 - Monday, September 30, 2013 - link

    Sunnyjuicedrink: Upgrading can be easy as long as your loop allowed for draining the system. Obviously if you change motherboard then components may be in slightly different places so your tubing may need replacing, simply changing the CPU is straightforward. Changing GPU can be awkward as you will probably need a new GPU waterblock and location of inlets on the block may not be the same as previously (although that just means some new tubing). All components (usually) work on a g1/4" socket Reply
  • Sadrak85 - Monday, September 30, 2013 - link

    After reading the whole thing (you did a lot of work here!), here's some points from the peanut gallery:
    1) I don't think it's just the lousy TIM under the IHS for the 4770K. I think part of the blame is to the integrated voltage modules, which more and more, I believe was included solely to allow partners to buy cheaper motherboards for their laptops and desktops.
    2) Thanks to IPC, that 4.4 GHz Haswell should perform (roughly) like a 5 GHz Sandy Bridge, and for what it's worth, 5 GHz Sandy Bridge chips were also rare.
    3) To easily attach tubing to barbs, boil water, put the end of the tubing into the water until it also reaches the boiling temperature, then slip the hot tubing easily onto the barbs. When it cools, it will be on there VERY well; in fact, you would likely have to cut it to remove it, but the huge forces need not be applied.
    Reply
  • DanNeely - Monday, September 30, 2013 - link

    The FIVR has more to do with Intel's mobile first strategy. It lets the CPU ramp/drop voltages much faster (order of magnitude???) which helps stretch battery life out by increasing the amount of time the CPU can spent at minimum power levels. Desktop systems get them by default because it's easier for Intel to use the same power regulation design everywhere. Reply
  • mokre - Monday, September 30, 2013 - link

    Thanks for the article. I'd like to suggest a couple of things:

    First, have a look at the fans you were using; I suspect for the radiators the fans do not have enough static pressure to move air through the rad, which is likely why you had to turn the fan speeds up. A buddy of mine has the same setup, but two gtx titans instead, in the same case, with slightly lower radiator capacity, and his temperatures are in line with yours, but noise-wise his system is much more quiet.

    Second, using a wrench on the fittings is probably not a good idea. While you can certainly tighten them properly, it could damage the fittings or worse yet, strip the threads on the acetal material used on the waterblocks (but only if you overtighten).
    Reply
  • pws328 - Monday, September 30, 2013 - link

    I think this is a great article, with a ton of helpful info. My only real question is why you setup your loop the way you did, chaining all of the components and the radiators together. I would think to get the best cooling possible you would use the following:

    *Pump
    Motherboard voltage circuitry.
    CPU waterblock (Apogee HD).
    *Top radiator (240mm).
    First GeForce GTX 780 (KOMODO-NV).
    Second GeForce GTX 780 (KOMODO-NV).
    Front radiator (360mm).
    Back to pump

    Since you have two radiators why not split up the load, or even run them separately? Is there an advantage to chaining the two radiators together? Again I have no experience with this, just a guess at what I would think would be the most efficient.
    Reply
  • nitemareglitch - Monday, September 30, 2013 - link

    I think if you had separated your GPU and CPU loops or added a rad you would have made out better. Also, DangerDen would have rocked your socks off. Reply
  • spidey81 - Monday, September 30, 2013 - link

    DangerDen went out of business earlier this year :( Was a sad day when I heard that.

    Separating the GPUs and CPU into separate loops would help, but it wouldn't be very beneficial unless you went with larger rads as well. Which would help with the single loop too. Also, setting up fans in a pull configuration, instead of push, increases the efficiency of weaker fans.

    Essentially, you want the largest surface area of heat dissipation as your case can afford to get the most out of liquid cooling. For example, if you can't fit a long radiator but a thick one will fit, use the thickest one you can afford to get.

    Another thing that wasn't addressed that I saw (but has been touched on in the comments) is loop order. Strictly speaking, it may help with temps out of the gate to evenly distribute the radiators between the loads. However, the longer the components are under a certain set of conditions (i.e. load or idle) the more the temperature in the loop evens out. Thus negating the perceived benefit of placing rads between heat generating components.

    I saw someone state that to achieve the best temps you need to set all your rads to intake air. This may be true, but you'll sacrifice case airflow efficiency. Keep in mind that you won't be watercooling every component in the case. Having balanced flow will not only help cool parts of the motherboard, GPU, and drives but you'll also have less dust buildup to clean out. Not to mention that you may gain up to a degree or two with the fans intaking air.

    Also, the thing with watercooling is that it not helps with cooling, but it looks amazing! Oh did I also mention that it's addicting... :)

    I currently have two systems under water. One is a 2700k with dual HD 6950's in a switch 810 with a 600mm worth of 60mm thick rads. The other is a m-itx build with 360mm worth of 35mm thick rads cooling a 3570k (delidded) and HD 7950.
    Reply
  • geniekid - Monday, September 30, 2013 - link

    Dustin nailed it when he said watercooling shines in GPU cooling - with good case airflow and a decently sized heatsink you can almost passively cool CPUs these days.

    Unfortunately, there aren't many CLC solutions for graphic cards and the ones that do exist (e.g. HydroCoppers) command a high premium compared to a DIY solution which is already pretty expensive.
    Reply
  • SkyBill40 - Monday, September 30, 2013 - link

    Very well written and informative. I can understand and appreciate the feeling of satisfaction and/or accomplishment when building something like this. You do make a good point in that "...watercooling may just not give up the performance you were hoping for." This seems to be one of those particular instances seeing as the differences between the OC air and water are not as high as one might expect and certainly don't make the extra expenditure of funds appear all that worth it. Personally, I'm leery of water cooling seeing how water and electronics never mix well. Pre-sealed systems are significantly safer, but your're limited specifically to CPU cooling if going that route. Perhaps a hybrid water/air system might work best for most budgets end produce similar results with far less fuss. But, if money isn't really much of an object and you're a tinkerer, go for it. Reply
  • JDG1980 - Monday, September 30, 2013 - link

    This was supposed to be an article about water cooling, but what you really demonstrated was that Haswell sucks on the desktop (at least the LGA1155 version with a badly attached heat spreader). I think that IVB-E (maybe the 4930K) would have shown off water cooling's strengths better. Reply
  • Hrel - Monday, September 30, 2013 - link

    Fantastic article!

    Pretty much just confirms what I already knew though. Liquid cooling is a hobby, not a practical means by which to build a PC today. Air cooling has just gotten too good. You see gains in benchmarks but nothing so substantial as to translate into being noticeable in real world usage scenarios.

    On top of the insane amount of work, and expense, you have added noise. CPU's and GPU's have built in overclocking today. So any real world gains are nothing but a pipe dream. 11% faster? You won't even notice that, it certainly won't be the difference between playable and not playable. On top of that with a $20 aftermarket CPU cooler you can get your PC totally silent. Don't have to suffer through all the noise of a pump and multiple, air restricted, fans. But even if you're deaf, or live in a way that you soon will be, the performance benefits of a .5%-2% increase in performance translates into literally nothing for real world usage.

    It's a neat idea, cooling a PC the same way we came to cool internal combustion engines. It even had a pay off for a short period of time in computing history. But the idea that anyone's going to get real world benefit from this hobby has reached the point of absurdity.

    Farewell liquid PC cooling.
    Reply
  • Shinobi_III - Monday, September 30, 2013 - link

    If you get MORE noise with water, you are clearly doing something wrong.

    And really, if you're going the DIY route, get a real pump, the thing in this article is a toy. Barely.

    The all in one kits are excellent for the money. But for true DIY you need real parts. Not this.
    Reply
  • BrightCandle - Monday, September 30, 2013 - link



    This review unfortunately is total rubbish, there are so many inaccuracies its actually worrying.I'll start with the biggest problem which is the amount of cooling you chose.

    A 120mm worth of cooling with this radiator, that is silent (800rpm) is about 115 Watts. So in total you put just 575 Watts of cooling into your loop at a water temperature of 10C above ambient. With the fans on full its going to be a bit more than around 800 watts of cooling. The components you are trying to cool:

    4770K - 150W
    2x 780 overclocked fully - about 400W each = 800W

    Grand total of 950W.

    So the loop is underspecced in terms of cooling, quite dramatically. There simply isn't enough radiators there even when the fans are running on full power. Further to that its no wonder that when the fans are turned right down the temperature shoots so far up, because the water temperature is DANGEROUSLY high, its actually going to be high enough to be damaging the tubing on full load.

    Realistically to cool a loop like this you need decent radiators (about 130W cooling at 10C delta) and around 8 slots worth of it, ie 2x360 + a 240 with fans designed for low speed fans (high fin density) such as the gentle Typhoon's in an actual watercooling case where those radiators and fans will fit, not this air cooling case that just about fits some thin radiators.

    The second major mistake is that you are cooling a 4770k. These chips are not really thermally limited. The thermal material under the heat spreader sure helps to make them hoter than they are meant to be but delidding them and replacing it and reducing the problem can net 20C but not really any extra overclocking. 22nm simply doesn't overclock as well and 4.4 is about where this chips normally top out. But additionally the 4770K doesn't pull much power even when overclocked, so its questionable if water cooling would ever help, because its only worth it if you chip will pull greater than about 200W, below that air does a better job. Realistically custom water is only worth it for overclocking on SB-E/IB-E and other high power chips, the low power mainstream ones we see today just don't net much benefit especially considering the thermal issues the chips themselves have out of the factory from Intel.

    So realistically you have underspeced the loop making all the components hotter than they should be due presumably to some bad advice and are cooling a processor with water that doesn't benefit from it. If that was it for errors I would stop here but it isn't. I have to say something about the inaccurate statements made in the review as well because they are going to seriously mislead.

    "The overwhelming majority of closed loop coolers employ aluminum radiators instead of the copper and brass that are used in custom loops, and the pumps tend to be on the weaker side, presumably to both keep noise down and because there's really only one component to cool"

    They use aluminium because its cheaper. The pumps are weaker because the loop is shorter and you don't need more, but most pumps aren't noisy at all, they are very nearly silent when a loop is setup correctly. Actually the all in one coolers tend to have noiser pumps because they are attached to something hard - the motherboard. With a separate pump you can mount it on some foam and eliminate all the vibration and noise for $5.

    "For pure thermal-to-noise efficiency, they're basically unbeatable, but if you want absolute or near absolute silence, you actually have to go back to conventional air cooling. The reason is that watercooling necessitates using a water pump, and while they can be tuned down for efficiency, they're never going to be dead silent. An air cooler will always be a fan plus heatsink; watercooling adds a pump."

    Near silence on a high end kit can be achieved with watercooling if you actually spec the loop correctly, which in this case you didn't. The reality is that if you don't have enough cooling and you don't mount the pump on foam then yes there will produce noise, but that is like saying a normal air cooler doesn't work very well if you don't clip it to the CPU properly, its user error. Water pumps can and are quieter than any fan inside of a PSU, they are quieter than most hard drives.

    "Water transfers heat exceptionally well, and radiators in turn will be massive, densely packed arrays of copper fins."

    Water does not transfer heat well, its actually a very good insulator. But what it does do is take quite a lot of heat to increase its temperature and its better than just air in terms of pulling the heat away from the component. The downside of water is its another material the heat has to go through, you need to pay attention to making sure that the water doesn't get hot by making sure there is enough exchange to the air via the radiators to ensure that doesn't happen, otherwise water performs worse than air.

    "Swiftech's HydrX PM (pre-mixed) coolant. the HydrX PM is actually 90% distilled water"
    "This is very similar to how coolant/anti-freeze works in your car; water carries heat very well, but is also corrosive, so chemicals are introduced into the mixture to counteract that effect."

    This isn't the right choice for coolant. Nowadays everyone in the watercooling world is using distilled water with a silver coil. You don't need anti freeze and anti corrosive because it actually performs worse than basic distilled water. The only reason for having anti corrosive in the loop is because you put a metal in there that corrodes, aluminium is the main offender. No one should be putting aluminium in a loop, even with anticorrosive it will slowly eat away at the metal in the components and realistically its already expensive to setup a loop, no point ruining it over a few years. The anticorrosive is pretty nasty stuff.

    "one with a normal fin density designed for low speed fans (these), and one with a higher fin density designed for high speed fans."

    Actually its the other way around. The high density fins work better with lower speed fans and designed for silent running systems whereas the low fin density is for the higher speed fans. Xtremesystem have plenty of charts showing this exact effect and its largely about the static pressure of fans and impedance to the air.

    "Next to them is one of the five Helix-120 PWM-controlled fans intended to be used with the radiators, and the odd-looking dongle is actually a SATA-powered PWM-splitter."

    I have never seen a professional watercooler review these fans, so chances are they are rubbish. There are plenty of fan choices that are good for radiators but not all fans work well for watercooling. The very best can be about 50% better than the worst unfortunate;y. Since I haven't seen a review of them comparing to other fans on a radiator its a bad choice of component.

    "Swiftech sent along their Apogee HD"

    Would not have been my choice of waterblock, it doesn't exactly perform very well compared to all the other choices. Its pretty dated as a waterblock and has never performed very well comparatively. A HK 3.0 will knock about 5C off the CPU temperature in comparison.
    Reply
  • scook9 - Monday, September 30, 2013 - link

    Good to see you got some experience in watercooling! There have been decent comments made already so I wont repeat what they all said. I will say that people are ridiculous when spec'ing out radiators for loops (kind of like how most people go crazy high on a PSU compared to what they really need). You have more than enough radiator for this build - my proof point is my HTPC. It is an i7 2700k and 2x GTX 570 all on the same loop cooled by a 30mm thick 240mm radiator without out a worry in the world. Yes they are all stock but it still is a non-trivial amount of power being dissipated and cooled. That all runs like a gem on a 650W PSU by the way without coming close to its limits (verified with a kill-a-watt). Also, loop order is pretty much irrelevant so just build your loop in whatever way is the most convenient from a tubing standpoint. The coolant is moving so fast with a good modern pump that the whole loops coolant temperature will normalize pretty quickly regardless of order. Just keep the coolant from getting saturated with heat and you are fine (and with that much radiator you are).

    The fans are probably the biggest chance for improvement - as mentioned you cannot do better than Gentle Typhoons usually. They are made by the Nidec servo company and distributed by Scythe. An important note on fans though when it comes to noise (or at least the GTs) - do not use them as intakes if they are right against a mesh opening in a case - this is loud! In that instance you are better off having them pull air through the radiator rather than push into it (at least better off from an acoustic stand point - have not tested cooling difference).

    The comments on de-lidding and the lackluster result of CPU cooling is spot on - unless you are directly touching the die there will be more heat than you want that you cannot get rid of. Back in the 130W Nehalem days there were decent improvements from watercooling but now-a-days the sub-100W CPUs don't benefit much over air cooling. I do it for aesthetics and because the benefits for cooling a GPU are absurd so adding the CPU to a loop is pretty trivial. My first loop was cooling an i7 965 and 2x 5870s. The 5870s were sandwiched together and both got up to 93c on the stock coolers - once watercooled they were down to 39c and no matter what I tried could not get them over that.

    My final note is on the actual coolant used. There is a lot of fear about water and it being damaging to electronics - while this is valid and true, pure distilled water is an insulator, it does not conduct electricity at all. In addition to having a better heat capacity it is non conductive, WIN! It is also a lot cheaper than the stuff these companies will try to sell you. Just don't mix aluminum and copper in your loop and don't use any dyes and distilled water and a single drop per liter of killwater (or whatever it is called) is all you need.

    Hope this was helpful!
    Reply
  • madisoncarter127 - Monday, September 30, 2013 - link

    hi Reply
  • madisoncarter127 - Monday, September 30, 2013 - link

    My last pay check was $11000working 10 hours a week online. My brothers friend has been averaging 17k for months now and she works about 18 hours a week. I can't believe how easy it was once I tried it out. Reply
  • dgingeri - Monday, September 30, 2013 - link

    You've sold me on the case, and convinced me to not try water cooling for a while longer, and possibly never. It looks like a lot of hard work for little to no reward. I'll stick to my H100. The case looks great, though. Reply
  • Shinobi_III - Monday, September 30, 2013 - link

    If you put the pump in a reservoir, you will no longer have a "noisy" pump.
    Noisy being a rather absurd wording, considering how incredibly little noise a EHEIM 600l/h pump produces.

    That pump you used in the article is ridiculous, what is that, a pump for ants?
    And compression fittings are complete suicide, they fail sooner or later.
    Perfectly regular ridged slide-over fittings, and a zip tie if you're nervous, will never fail.

    And the coolant is ridiculous too, get "long life" automotive coolant, mix it 1:10 and you will never need to worry again. Those bottles are just rebranded, ready-mixed glycol with a giant price tag.
    Reply
  • HisDivineOrder - Monday, September 30, 2013 - link

    Watercooling GPU's makes sense, but how long before CLC is common on those? It can't be long since everyone's used to CLC's on CPU's where you don't even need water cooling at all. I imagine it won't be long before GPU's too don't need watercooling anymore since they'll be integrated into the CPU.

    Given that Intel will make its own version of Mantle in short order that they'll pay entire publishers to use, I'm pretty sure we can kiss any advantage discrete cards have goodbye. I'm sure Intel will send AMD a fruit basket for both the idea and making everyone okay with it.
    Reply
  • ImSpartacus - Monday, September 30, 2013 - link

    Hot damn, this is comprehensive as fuck.

    Good article!
    Reply
  • livingplasma - Monday, September 30, 2013 - link

    Thanks for the great article, it's good to see a major tech site like Anandtech cover the often misunderstood art of water cooling. Just skimmed through the comments and a lot of my fellow water cooling guys/gals made some very important points. From my experience it is most important for radiator fan orientation to act as INTAKES. Flipping those top fans to intake can decrease your load temps ~8 degrees celcius from my experience as the way it's currently set up the top 2x120 radiator is using the air from the 3x120 radiator (which is already saturated with heat). Properly decoupled pumps will minimize noise/vibration transmission to the case and with the MCP35X's PMW control it can run very quietly when slowed down 50% which is more than enough flow for most setups. Radiators themselves are low flow resistance, full cover gpu blocks are probably next and CPU blocks tend to have the most but even maxing out a pump only decreases temperatures CPU by a degree or two (plus now the pump is dumping in more heat as well). Removing the stock GPU cooling setups is probably the biggest contributions to water cooling being more quiet than a regular air setup and even with a modest setup the temperature differences are huge as shown with the author's setup. Reply
  • 1Angelreloaded - Monday, September 30, 2013 - link

    Dustin, Do you know the difference between a Serial and Parallel loop? or how the High flow and Low Flow pressures effect heat dissipation on certain blocks? I understand this may be your first time setting this up; however, in the computer field research is everything, you already know this just like modifying a car without research you end up with an underperforming and underwhelming creation. This is not how a liquid cooled setup should be set up at all, and BTW your Proc temps are higher because the Air source feeding it is already heated from your GPU Rad outlet, try reversing the flow on intake to exit Flow and your CPU temps will be better, CPUs on average run at lower temps on full loads compared to GPUs that can average 70-80 C especially while gaming. Reply
  • hot120 - Tuesday, October 01, 2013 - link

    Man, there are quite a few fake 'experts' critiquing this fine article. It seems everyone wants the article done their way. Delid this CPU, change that fan, move this radiator, use that Kill Coil. Enough is enough. It is a basic article on watercolling, and is not meant to cover EVERY SINGLE POSSIBLE configuration. Only the most extreme of the extreme computer users will consider delidding their CPU's. Only the most insane of the extreme computer users will spend $800 on a custom loop to shave 5-10 degrees off their CPU/GPU temps. I'll take my $100 and by a CLC and live happily ever after with my CPU that is +10 degrees (over custom) and my air-cooled GPU. Some of you are tripping! Reply
  • pandemonium - Tuesday, October 01, 2013 - link

    Thanks for the introduction article to watercooling. To me - verified according to the results - this says it's still not worth it for a general user and gamer. The cost of increased power, thermals, and noise do not appear to outweigh the performance increase. I know other setups produce better results, but this is a good sample for not bothering with it.

    Watercooling is still for benchmarking...and that's about it.
    Reply
  • meacupla - Tuesday, October 01, 2013 - link

    The only thing this article proves is...
    1: corsair cube case can't mount enough radiators
    2: poor radiator choices
    Reply
  • ShieTar - Tuesday, October 01, 2013 - link

    Yeah, despite the description by Corsair, this case is most certainly not designed to work well with water cooling. I mean, it has a 330mmx460mm foot print, that should be enough for 6x120mm or even 6x140mm radiators on top. Set it up like that, add 360mm worth of intake fans to the bottom front, and you will look at much more cooling power than the current solution.

    On the other hand, the air flow looks great for the air cooled solution, with the drive cages nicely out of the way. Thats probably hurting the comparison number in the above review, the fact that we compare a great air cooling case with a mediocre water cooling case.
    Reply
  • HisDivineOrder - Tuesday, October 01, 2013 - link

    "The last problem I ran into had to do with Swiftech's PWM splitter. While it's supposed to take the PWM signal from the motherboard and split it among all of the fans connected to it, the fans all wound up just running at full speed. It's tough to figure out exactly where the blame lies or if it's just an incompatibility. Honestly, I think a rig like this is pretty much the ideal situation for a Corsair Link kit, where extremely tight and flexible software control of multiple PWM channels is vital. Gigabyte has made a fantastic motherboard here in the G1.Sniper 5, but their fan control is still woefully lackluster."

    Sounds like you were trying to split a (supposed) PWM fan port other than your main CPU PWM port. On my P8Z77-V I had this brilliant idea to split the PWM fan ports I needed and let the motherboard run the rest, including the CPU separately. To let some fans go up and down more frequently than others.

    Asus and other motherboard makers will tell you that all their 4 pin fan ports are PWM, but that's only half-true. Because they want them to be compatible with 3 pin fan control mechanisms, they are not true PWM controlled, so when you try to use a PWM splitter with them (all but that single 4 pin PWM fan port they have dedicated to CPU (and not the extra one either)), then they go full speed.

    I had to wind up splitting all my fans off of just the main CPU 4 pin. This works, but it's not ideal because ALL my fans go up and down with what my CPU fan is doing. That's more noise than I'd like, but it's still less than the alternative.

    All this to say, it's the motherboard maker's fault and you guys have been letting them get a free pass for years now on this. Perhaps you should start testing it and take them to task over it. I wouldn't blame the 8 pin splitter as I have it and it works fine.
    Reply
  • The Von Matrices - Wednesday, October 02, 2013 - link

    That's exactly what was true with my previous ASUS P6X58D Premium board. My new ASRock Z87-Extreme6 has only two PWM fam connectors, but both are wired correctly and are independently controlled. I agree that reviewers should test for this issue, but it seems that people who specifically want PWM fans are a minor component of the fan buying market and most will settle for 3-pin voltage controlled fans. Reply
  • mohamed Siran - Wednesday, October 02, 2013 - link

    pls can you tel first photo corsair Case model ? Reply
  • Kathrine647 - Wednesday, October 02, 2013 - link

    like Gregory said I am alarmed that a stay at home mom able to earn $5886 in 1 month on the internet. visit their website............B u z z 5 5 . com open the link without spaces Reply
  • alison_lenihan - Thursday, October 03, 2013 - link

    what Eric said I am shocked that some people can profit $4550 in one month on the computer. see post....... CuttR.it/tvtmbce Reply
  • don_k - Friday, October 04, 2013 - link

    "The reason is that watercooling necessitates using a water pump, and while they can be tuned down for efficiency, they're never going to be dead silent. An air cooler will always be a fan plus heatsink; watercooling adds a pump."

    Excellent article, very thorough, one thing though, I would beg to differ on this point. Pump noise rather obviously depends on the pump. The MCP35X that you chose, while a very good pump, is not the most silent. A D5 on the other hand can be completely, note not relatively but completely, silent when decoupled meaning not sitting or attached directly to the case so it does not cause vibrations on the case, even on its highest setting.

    Similarly, how much fan noise you have depends on the fans you get and how fast you run them. The benefit with watercooling is you have a much larger surface area from which to dissipate heat, which gives you the option of several slow fans without sacrificing cooling capability.

    Sure, you would get an few less points Celcius with higher RPMs but why bother when you're already at such a cool temperature and you need to double your fan speed and go from silent to noisy to do it.

    To give you some concrete numbers from my own testing, a 3x120/140 radiator with a low FPI that is designed for low speed fans and 3 good quality fans at 700RPM is near silent. At 500RPM or less, completely silent. Pump noise is a non-issue, a D5 can be completely silent as I mentioned earlier.

    Going from 500RPM to 1300RPM, the max on my fans, gives me a max of 5-7C decrease in water temperature under continuous load on both CPU and GPU, meaning running at 100% load for as long as necessary until temperatures stabilise at their maximum point.

    Some watercoolers like to run high RPM fans and high FPI radiators so they can get really good temperatures at the expense of noise. Others do the opposite to get a silent system and still get better temps than on air. The choice is yours.
    Reply
  • Shelbino - Monday, October 07, 2013 - link

    Dustin, thanks very much for this article. I'm glad you supplied so much detail. My only big problem with the specific design here is the cost. I did a tally of the liquid cooling parts you used, and came up with a total of nearly one thousand dollars! IMHO, that's w-a-y too much for the relatively modest performance gains. Despite that, thanks for this detailed piece on liquid cooling. Reply
  • Volnaiskra - Friday, October 11, 2013 - link

    So.......three days of work and hundreds of dollars for more noise, more power consumption, more maintenance, and about a 2% performance increase?

    I was seriously considering watercooling before, but you have thoroughly talked me out of it!!
    Reply
  • expatswe - Saturday, October 12, 2013 - link

    hmm, I liked the article and how the system was put together. Except I can not understand why it was so important to put the pump in the crammed space behind the powersupply when half the case was empty anyway.

    I was though really disappointed on how the over clocking was done and the results being shown. To me it looks like the writer ran out of steam and thus simply settled on the basic results on overclocking with liquid compared to air. It would be easy to get the system up to atleast 4.7GHz in that setup. I doubt it was the processor (which was given the blame)

    30 db is the "noise" level in a quiet bedroom at night. 40db is a quiet library. Thus in the test it would be more interesting to know what the temperature levels seen on for example the water with CPU over clocked at different speeds without the GPU's being overclocked and then with one overclocked vs 2 overclocked to max level. Since the system was in series the water was effectively pre-heated before going into the GPU's. This is a common solution and since the article is written for beginners that is what would be interesting. Temperatures then also shown at no load vs full load at the relevant speeds would have been relevant.

    This was not really a test of game speeds and software speeds as such but that of a build of cooling systems.
    Reply
  • egotrippin - Monday, October 14, 2013 - link



    I'm about to nerd out so forgive me...

    Some of the conclusions from this article aren't representative of what you can truly do with water cooling but rather the limitations of the equipment you selected.

    For starters, the addition of a pump doesn't automatically mean more noise. The pump you selected can sound like an angry wind up toy and it generates a lot of heat and vibration. A Laing D5 Vario/MCP 655 pump is whisper quiet, especially with a Koolance PMP450 top on it. It can be running at full speed pushing 4.6 liters per minute through the system and so long as it's decoupled properly from any hard surface in the computer case then you wouldn't be able to hear it if it was 6 inches from your ear.

    I run a 3930k overclocked typically to 4.4 GHz although I've done higher and if I'm running prime 95 or Intel burn-test my temps are about 66C TOTAL and not 66 DELTA?! If your room is a comfortable 24 C does that mean your CPU temps are 84C on water? Something isn't right. I don't know that chip but maybe it needs to be Delidded and Lapped. I had that same Apogee HD block before on a 3930k which is a much hotter chip and my temps never reached that high.

    The fans you are using have a familiar style... because they look just like the Scythe Gentle Typhoons which have the best noise/static pressure of any radiator fan I've seen tested. If you used Gentle Typhoons, temps would be lower and the whole thing would be quiet.

    The coolant you used is probably less efficient than distilled water. The only coolant I've used that delivers lower temperatures is Ice Dragon which is heavy and expensive and cuts your flow rate in half. Distilled water works great and it's $1

    I can't tell for sure but it looks like small diameter tubing was used. Using 1/2" inner diameter tubing delivers higher flow which means lower temperatures.

    Those radiators are anorexic. I don't think they come much thinner than that. Use something with a bit of heft. My rad is 80mm thick which is twice the thickness of those Swiffys. This allows for substantially more cooling and also higher flow rate which, again, increases cooling and lowers noise.

    My 3930k + my GTX 690 dual GPU card can both be overclocked and Folding or benching with all cores/gpus at 100% and it can be silent enough that if it weren't for the power light, you wouldn't know it was on. I briefly used air on my GTX 690 and it sounded like a hair dryer and filled the room with the scent of charred air much like the smell of turning your heater on for the first time in winter.

    If anybody reads this, you can expect better results if you buy better components. I started off with Swiftech because they were cheap and also they were carried at my local MicroCenter. I quickly graduated to better parts. The Apogee HD is an excellent water block and the Swiftech MCP35X2 and MCP655 pumps are both excellent pumps (but neither were used here).
    Reply
  • prismatics - Wednesday, October 16, 2013 - link

    Why did you only post OC Liquid benchmarks? I'm interested in Non-OC liquid numbers. I have no interest in overclocking, I just want the quietest, most efficient system. Reply
  • mc2k4 - Tuesday, December 31, 2013 - link

    Terrible article, would put off anyone from watercooling. Those results are horrendous. Reply
  • woogitboogity - Wednesday, January 22, 2014 - link

    I did a custom build CPU/Northbridge/GPU with the cooling loop going outside the case to the reservoir about 5 years ago...

    I will admit that when the thing ACTUALLY worked for a while it was insane... granted it was 5 years ago but even then seeing lukewarm temperatures on essentially every component at full load was pretty impressive.

    BUT... I feel that one thing missing from this article is a reality check: I work in experimental physics and I have had to work water cooling and even liquid helium cooling for magnets. 5 years ago the vendors of water cooling hardware implied a LOT more than they could back up in practice... since then their claims have only gotten more extravagant. At least in the days when people did water cooling from scratch they did not have marketeers offering false assurances in the form of warranties that clearly do not cover damage due to other hardware.

    BOTTOM LINE: I think the subtext of this entire article that needs to be clear is that custom water cooling should be treated as a VERY expensive hobby. Expect to lose every component... period. If that is not an acceptable outcome don't do it. I say this because I deal with experimental cooling all the time and I got burned by the sub-par cooling hardware offered not too long ago (same hoses and cooling block designs... still using water).
    Reply
  • Drittz121 - Friday, February 28, 2014 - link

    Just do yourself a favor. STAY AWAY from this company. Yes they look good. But when it breaks and it WILL. All they do is give you the run around. They have had my system for over 2 months trying to fix the garbage they sell. Worse company out there for support. DONT BUY Reply

Log in

Don't have an account? Sign up now