A few months ago, Logical Increments reader Nintenrax built an incredible water-cooled PC for himself, and shared with us a very detailed recap of the build. (See all pics on PCPartPicker.)
I’ve been gaming on an Alienware M14X R2 with a GT650M for 4 years now. It was a purchase from a time in which I was much less knowledgable about PC hardware, and not the smartest buy when I look at it now. That’s all in the past, however, and in the middle of last year, the time had come that I build something a lot more powerful myself.
I wanted to do something special for my personal build. A normal ATX tower is cool and all, but I wanted something unique, or at the very least somewhat different. At first I was looking into Micro-ATX cube-cases like the Corsair Air 240. Later on I also began looking into some Mini-ITX cases. Eventually I decided I wanted to go with a Mini-ITX build, and I immediately fell in love with the Fractal Design Node 304.
At first I started planning for a normal build with air cooling, but later on I started to get a bit worried about temperatures, since I was definitely planning on overclocking. For example, I was planning on going with a blower-style cooler to keep heat out of the case, but the reference blower-cooler has trouble with throttling and I assumed that, by extension, any blower-cooler design would have the same problem. So I started looking into water cooling. After a while I stumbled on this and this build, proving to me that it definitely could be done. And so my quest began.
Planning out every tiny detail is enormously important when you’re trying to build in a smaller chassis, and it becomes even more important when you go ahead and throw watercooling into the mix. Most of the time I put into this build was probably spend planning everything out, followed by the time I spent benchmarking. The actual building, while still taking a good portion of my time, only takes a far third place. I will quickly go over my choices of PC hardware before getting into a detailed overview of my planning process and choice of watercooling parts.
I needed something that would drive the latest games on my monitor, so with a resolution of 3440×1440. Since no one benchmarks at that resolution, it’s a bit tricky to take a decision based on existing reviews. To have an approximation of how different GPUs would perform at 3440×1440, I made a humongous Excel file.
I collected benchmark data from a multitude of benchmarking sources (including AnandTech, TomsHardware, TechPowerUP, …) and put them in the Excel file. Following that, I made both linear and logarithmic approximations of how the FPS scales throughout resolutions. All that was left after that was to enter an x-value (3440×1440 or 4.953.600 pixels) and evaluate the y-value (FPS). This is all neatly done in the Excel-file by a little input field and a bunch of graphs.
For the GTX 1080 at 3440×1440, you get the following graph, which displays a bunch of games and their FPS:
From the graph, it’s very clear that a GTX 1080 is definitely needed for a lot of games at 3440×1440. It even falls short some times, but since the 1080 was the most powerful card at the time of the build and I didn’t want to deal with SLI (besides the fact that it isn’t even possible in a Mini-ITX build) I just went with the 1080. Why I specifically went with the Gigabyte GTX 1080 Windforce OC will be explained later.
I first wanted to go with just an i5-6600K since that should be enough for most games, but eventually went with a i7-6700K since some more recent games can be very CPU-intensive (GTA V is a prime example). I also plan on possibly doing some more CPU-intensive tasks like video editing/rendering in the future.
RAM is very basic. Basic 16 GB amount, basic 3000 MHz speed (decently fast, but nothing crazy). Got it during an Amazon sale.
At the time of purchase, there were not a crazy amount of Z170 Mini-ITX boards from which to choose. Enough to have some diversity, but not enough to be completely overwhelmed. I looked through all the boards, and in the end the ASUS Z170i Pro Gaming and the Asus Z170 Maximus VIII Impact were the ones that stood out to me the most. The Z170i Gaming:
- is cheaper, but I can deal with some extra cost; this build was going to cost a lot anyway
- has more USB ports, but I can deal with 2 fewer
- has more internal USB connectors, but the USB 2.0 one would go unused anyway
- has M.2 instead of U.2. This is honestly one of the major factors that made me hesitate between the Z170i Pro Gaming and the Z170 Maximus VIII Impact. I eventually decided that I could just use a U.2 SSD or a U.2 to M.2 adapter (there’s not a lot of information on them, but they DO exist) if I ever wanted to get a PCIe SSD. If you think about it really, it’s much easier to add/replace a U.2 SSD than it is to add/replace a M.2 SSD on the back of your motherboard, especially in a small chassis, so U.2 does have its advantages
Besides all this, the Maximus VIII Impact:
- has a better VRM design and has some really neat overclocking features build in, which is important since I definitely wanted to overclock.
- has a better audio design, and since you don’t have any extra PCIe slots to install sound cards, having some good integrated audio is important
- has more fan headers
- has some other nice features, like headers for temperature probes, voltage monitoring points, debugging LEDs, etc.
All-in all, while the Maximus VIII Impact has its disadvantages, it seemed like the best board for my build.
For storage, I went with a standard HDD + SATA SSD combo. No fancy PCIe SSD yet; that’s something for future upgrades. For the SSD I chose the 250GB Samsung 850 EVO, still one of the best performing value drives.
For the HDD I had to go with a 2.5” model, since fitting a 3.5” model into an already cramped build would’ve been really difficult. The HGST Travelstar 7K1000 was one of the cheapest 2.5” HDD’s I could find and according to tests, also one of the fastest.
Again with size in mind, I decided to go with an SFX PSU to make sure there was enough place to install a pump and for cable management. Corsair’s SF series are pretty much the best SFX power supplies on the market. I went with the 600W version because 450W would be cutting it close, especially when everything is overclocked.
To make sure all of the components fit nicely — especially the watercooling components — I made a detailed model in Sketchup. This model went through multiple iterations as components changed, until I finally ended up with the model that you can see in this render:
I modeled the case based on my own measurements. For a lot of the components (e.g. the motherboard, fans, some radiators, etc.) I used 3D-models I found online, which I sometimes modified if needed. I modeled the remaining components myself based on measurements and pictures found online.
I didn’t really know much about custom water cooling when I started planning for this build. To learn more, I read guide after guide after guide after guide. Especially helpful were Extreme Rigs, ThermalBench, Martin’s Liquid Lab and SkinneeLabs, which offer a lot of information and reviews for water cooling and water cooling parts. After a while I considered myself informed enough and began to assemble a list of components that I was going to need.
For the CPU block, I was first planning to use an EK Supremacy EVO since it’s one of the best performing blocks according to tests. I eventually went with an EK-FB Asus M8I Monoblock for the Maximus VIII Impact to make sure that the VRM would also be adequately cooled.
For the GPU, I used an EK-FC1080 GTX G1 Waterblock. I find them really clean looking and they also perform among the best for both the chip itself and the memory modules and VRM without being too restrictive to the water flow.
EK were also the only ones that had a block available for the Gigabyte GTX 1080 Windforce OC. Which brings me back why I chose this particular model of 1080. I was looking for a non-reference PCB so that I had a card with a power and VRM-design that would lend itself a little bit better to overclocking. But I also needed a card that was short enough so that it would fit in the case with a radiator in front of it. The Windforce OC was the only card (at the time) that filled these criteria while also having a waterblock available.
I managed to fit a dual 92mm radiator in the front and a single 120mm radiator in the back of the case. To decide which radiators to use, I created an Excel file very similar to the GPU Benchmark Database that I described above. It contains data for a good amount of radiators (mainly obtained from Extreme Rigs).
In the beginning I had set my mind on the Black Ice Nemesis M184 GTX. Later on I discovered that there is also a dual 92mm version of the Alphacool NexXxos XT45 which is slightly thinner. It performs a little bit worse than the Black Ice, but this is easily offset by the fact that it allows me to use standard 25mm fans, instead of the 14mm fans I would’ve had to use so that everything would fit. These thicker fans perform better (and quieter) than their 14mm counterparts.
For the back radiator I first planned to use the Black Ice Nemesis 120GTS, but I changed my mind based on the database. I eventually went for the EK-Coolstream PE 120. It performs similarly to the Black Ice but with much lower restriction to the flow. It is a little bit thicker, but still fits comfortably in the case. For the front fans I used two Noctua NF-A9’s. There aren’t really any reviews available for 92 mm fans so I just went with a brand that I knew was reliable (and quiet).
Pump and reservoir
I made an Excel file (again) to calculate what the flow rate through the loop would be for different pumps (and pump tops). Based on that I was fairly certain I wanted to go for a DDC 3.2 pump: Powerful enough for my build, but still small enough to fit in it. I also decided I definitely wanted a PWM-controlled pump so I could easily adjust the speed of the pump.
I briefly pondered about the XSPC Acrylic Reservoir for Laing DDC and the Watercool DDC-Tank LT. The XSPC Reservoir performs slightly better than the EK but I eventually found out it didn’t fit into the case. The Watercool Tank is nice and tiny, which is great for a mini-ITX build, but I eventually decided against it because it has less inlet and outlet options than the EK-XRES and the small size might also prove a little troublesome while trying to bleed out the air bubbles.
At first a was a bit worried that, since I had to use one of the side inlets (there is no clearance to use the top inlet), that performance might be impacted as was evident in this test. I even briefly thought about building my own reservoir to circumvent this issue. However, since the side inlet feeds into the reservoir first, which then feeds into the top of the pump, there shouldn’t be a problem.
Last but not least, I decided to go with the slightly more expensive Elite version which features a heatsink, since a DDC 3.2 can get pretty hot without one.
For the tubing I used Primochill PrimoFlex Advanced LRT 10/13 mm. 10/13 mm is a little on the small side, but since tubing size doesn’t have a huge impact on the flow rate, it’s better to go for a smaller tube size so it doesn’t take up too much space in an already cramped build. I went with white tubing since I think it’s a nice accent color. Even though you can’t look inisde the case for now, I’m planning on potentially adding a small window in the future. For fittings I just went with EK fittings since I was already ordering a bunch of EK parts anyway.
I used a 90° elbow at the second side inlet of the pump to act as a draining point and also used a Y-splitter right before the front radiator to act as a second draining point. A T-splitter with an XSPC temperature probe right before the graphics card block is used to measure the water temperature and a second temperature probe (supplied with the motherboard) is placed right before the front fans to measure the temperature of the incoming air.
I was planning on just using some distilled water with a silver coil as coolant. But I eventually went with Mayhems X1 Clear for two reasons: Distilled water doesn’t seem to be as readily available here in Belgium and I got a little worried about possible corrosion.
Normally speaking, corrosion shouldn’t be a worry in my build since I’m only mixing nickel, brass and copper. However, EK has had some problems with its nickel plating in the past (which should be fixed now, but still), and I didn’t want to take any risks. Mayhems seemed to be a popular and reliable brand when it comes to coolant, so I went their X1 series, which has all the protections against corrosion and algae you want. I didn’t use any colored coolant as to avoid any problems with coloring additives. The only thing that is transparent in the whole loop is the reservoir anyway, and that’s a part of the build that I wouldn’t want to show with a window.
I first assembled the build cooling everything on air to test if all the components worked correctly and to perform some base benchmarking tests for future comparison. I used the stock cooler for the GPU, the stock Intel cooler for the CPU, and the fans supplied with the case set to high (12V) with the case’s fan controller. I didn’t really cable manage anything very well since I had to take it apart again fairly quickly anyway. I had some pictures of my building process, but these unfortunately got lost somewhere along the way.
Before I could start assembling the watercooled build, I had to clean out all the watercooling components. I just flushed most parts a few times with demineralized water, but for the radiators I also used the “shake method”:
- Boil demineralized water
- Let it cool down for a while
- Fill radiator with it
- Let it cool down a little bit more
- Poor out 1/3 of the water
- Vigorously shake the radiator
- Pour out the water
I did this two times for each radiator. To my surprise, the insides of the radiators (or any of the components for that matter) weren’t really dirty at all, as opposed to what a lot of the guides described. Following the cleaning of the components I took apart the build on air. Nothing too difficult there, but I was very thankful that I hadn’t cable managed it, since that would’ve made disassembly a lot more difficult.
After all that, I could finally begin building. I started with installing the motherboard monoblock. I removed the Intel stock cooler (I actually had some trouble turning the tabs 90° because there was a lot of stuff in the way, but eventually it worked out) and removed the thermal paste with some Artic Silver ArctiClean.
I removed the stock heatsink for the VRM module, which was easy enough, and also removed the RAM sticks since they were a bit in the way. I applied the EK supplied thermal pads to the VRM, put thermal paste on the CPU (I used Thermal Kryonaut for both the CPU and GPU, one of the best non-liquid metal thermal pastes according to this test) and installed the monoblock.
I had some trouble aligning all the screw holes, which wasn’t made easier by the fact that I had to hold the monoblock in place and screw everything in at the same time only using my two hands. Eventually I got everything aligned though, and the monoblock was nicely in place.
I then installed the motherboard into the case, which was easy enough. I also installed the PSU using the SFX-ATX bracket. Then came the rigorous task of cable managing. I routed all the front panel cables (USB, audio, etc.) to the motherboard and plugged them all in.
I managed to tuck most of them away between the power supply and the motherboard. I also plugged in all the PSU cables I was going to need (SATA, Molex, 24-pin, CPU 8-Pin and PCIe 8-Pin) and routed them to the right places. I DID NOT plug them into the components yet, however, since the loop would have to be tested later on without powering any of the components.
I used the honeycomb intake-pattern that was left in front of the PSU to tie the cables down and to tuck them away. I spend quite a bit of time making sure I used the space as efficiently as possible (which was VERY important in this build), but when I was finally done I was quite happy about the results. During this time I also installed the front fans + radiator and routed the fan cables to the right spot using the holes in the SFX-ATX bracket.
Installing the GPU waterblock was probably the one thing I was somewhat fearful of, to be totally honest. I had never done it before and was really afraid I might break something. I followed EK’s instructions to remove the cooler and backplate, which went well, as went removing the thermal pads. To remove the thermal compound I used coffee filters and Artic Silver ArctiClean.
I was very careful during the process as to not break anything. After that it was simply a matter of putting on thermal paste and pads, installing the block and backplate and screwing it down. All in all the whole process went smoother and was easier than expected, which was a whole relief for me.
When all that was done, I installed the GPU into the case, which took some twisting and turning. Then I installed the back fan and radiator. It’s very important to install the back radiator AFTER you install the GPU, since it’s basically impossible to do it the other way around.
The last component that had to be installed before the tubing runs was the pump. I attached the necessary fittings and elbows and cut down the feet of the pump a little so that it would fit in the case. Then I glued the pump to the top of the PSU with some heavy duty glue. Now came the time to run the tubing. I cut down the tubing to approximately the right length, cutting it a little long rather than short. I attached one side to a fitting, cut some extra tubing off if necessary and attached the other side to the other fitting.
To easily slide the tubing over the fittings, I heated the ends up a little with hot water. When everything was tightly attached, I screwed on the rings of the fittings to hold the tubing in place, using pliers where my fingers didn’t really fit. I kept most runs as short as possible to save space, but I left some slack on the tubing between the two radiators so I can easily turn the Y-splitter for when I want to drain the loop later on.
After all this was installed, it was time for the moment of truth: leak-testing the loop. I connected the pump to a molex connector for power and connected the PWM connector to SATA via an adapter. I also bridged the 24-pin connector with a plug so that I didn’t have to power any components besides the pump. I basically filled the whole case with paper towels so that if there were to be a leak it would hopefully be contained by the towels.
To fill the loop I filled the reservoir about 2/3 of the way and turned on the power supply until most of the water was pumped away, but making absolutely sure the pump never ran dry. Then I filled the pump about 2/3 again and repeated the process until the whole loop was filled. I leak-tested the loop for about 20 hours, during which I encountered no problems at all. Bleeding went pretty quick too, considering the fairly small reservoir.
The final parts I needed to install where the two drives and the fan controller that came with the motherboard. I screwed them all into a simple bracket for 2.5” and 3.5” drives that I found online. I then attached the bracket to the bar that goes into top middle of the case and normally supports the back part of the drive cages. I plugged in the last cables, did some more cable management, and it was finally done. The only thing left was pressing the power button and hope for it to turn on, which it did! Hooray!
I did some minor overclocking on the RAM, CPU and GPU. For the RAM I simply enabled the XMP profile, which sets it to 3000 MHz. For the CPU I only overclocked the Core, since I didn’t really want to put any time in Uncore overclocking. I started at 1.3V and 4.5 GHz and pushed it further in increments, upping the voltage when needed. I tested stability using some heavy games, synthetic benchmarks and AIDA 64. I was able to push it to 4.7 GHz at 1.34V, but temperatures were getting a little high under the stress test so I dialed it back to 4.6 GHz at 1.32V, which I am more than happy with.
For the GPU, I used EVGA Precision X. I maxed out the voltage, power and temperature limits and started with setting the GPU Clock to about +100 MHz, which came to about 2036 MHz total. I then increased the clock in increments of 25 Mhz, all the while testing stability with games and synthetic benchmarks. At +200 Mhz (2125 GHz), there were some very minor glitches in The Witcher 3 and the Time Spy benchmark, so I dialed it back to +175 Mhz (2101 MHz), which was perfectly stable. After I overlocked the GPU Clock I also overclocked the Memory Clock. I was able to reach +500 MHz (11 GHz total) without any problems.
I ran a large suite of benchmarks on the air cooled, stock water cooled and overclocked water cooled setups to see what the effect of water cooling and overclocking was on the performance and temperatures of my build. The benchmarks that I used are:
|Games||Synthetic Benchmarks||Stress Tests|
|Heroes of the Storm||3DMark Firestrike||AIDA 64|
|XCOM 2||3DMark Time Spy|
|GTA V||Unigine Heaven|
|Total War: Atilla|
The Heroes of the Storm and Bioshock benchmarks show some weird behavior, not adhering to the trend that can be seen in the other benchmarks. The anomalies in the Heroes of the Storm benchmarks could be explained by some graphical updates that were made to the game, but I’m not exactly sure what causes the Bioshock benchmark to benefit so much from the watercooling and the overclocking.
HOTS and Bioshock excluded, all the benchmarks show a similar trend. There is some very minor improvement (1% average) because of the watercooling, which keeps the clocks a little bit more stable (see later). Overclocking gives an average improvement of 11%. One thing that has to be noted is the fact that the CPU scores for both the 3DMark benchmarks went down a little for stock watercooling compared to stock air cooling. Since there were no problems with throttling, I’m not exactly sure what caused this.
I ran an AIDA 64 stress test (CPU and GPU combined) for half an hour for all three configurations. In the air cooled setup, the stock fans were used, set to high (12V). This comes down to 1300 RPM for the 2 92mm fans and 1000 RPM for the 120mm fan. The CPU fan was plugged into the CPU fan connector and used the standard profile that was set in the BIOS. It almost reached a whopping 2300 RPM, which was clearly audible.
The GPU fans also used their stock fan profile and went up to about 1900 RPM. The CPU temperature maxed out at 80°C, while the GPU maxed out at 66°C. Not bad for such a small build, but that doesn’t leave much room for overclocking and it did require very high fan speeds. Also, while the CPU clocks and GPU Memory Clock were stable, the GPU Clock quicly dropped down to 1885 MHz due to temperatures.
With water-cooling in place, everything got a lot better. Instead of the total 7 fans in the air-cooled build, the water-cooled build only has a total of 3 fans, spinning at a maximum of 1550 RPM. This is higher than the case fans in the air-cooled build, but a lot lower than the CPU and GPU fans at the same time. For the two 92mm fans, I used a custom fan profile based on the water temperature. For the 120mm fan, I used a custom fan profile based on the CPU temperature. (I would’ve based this fan on the water temperature too if not for the fact that a fan connected to the CPU fan connector can only be controlled based on CPU temperatures, and there was no way to reconnect the fan elsewhere now everything was in place.)
Temperatures also got a lot better, with a maximum of 63°C for the CPU and a maximum of 51°C for the GPU. Last but not least, the throttling observed in the air-cooled setup was no longer present due to the lower temperatures.
And finally there’s the overclocked setup. Fan speeds went up a little bit compared to the non-overclocked build, with a maximum of about 1670 RPM (+120 RPM). GPU temperatures also went up a little bit to 55°C (+4°C). CPU temperatures, on the other hand, increased significantly to a maximum of 85°C. The CPU temperatures also fluctuated a lot more compared to the other two setups. Clock speeds were again very stable. Only the GPU Clock displayed a minor dip (from 2101 MHz to 2088 MHz) as soon as the GPU temperature reaches 52°C.
All-in all I’m very happy with the end result. It’s not perfect, and there are a few things I’d like to fix/replace during the first maintenance round. The front audio doesn’t seem properly connected for example, something I only noticed at the very end and can’t solve now the case is cramped full with parts. I can’t use Sleep mode either, since whenever the PC wakes out of sleep the pump spins at full RPM and can’t be controlled anymore. Luckily it’s not very bothersome to live with these problems, and I can easily wait until I fix them in the future. In the end I have something that is (semi-)unique and something that I’m very proud of. And, perhaps most importantly, I can finally do some heavy gaming again.