My first foray into watercooling (aka yet another desktop thread)

EDIT: For completed loop, skip right to post #36

I'm going to start by saying that all I did was a dinky CPU loop, nothing too impressive so don't get your hopes up. However if you're curious to know what building a custom loop is like for a first time watercooling noob and care for my <del>ramblings</del> thoughts, read on.

Since the entire thing is far too tl;dr to fit into one post, I'm going to break it up into several posts to make it easier to manage, and also so those of you who just want PICS NAO can skip right ahead to the last post and ignore everything else.

Quick background: desktop in sig finished around end of September. Here I have some pics of the finished build. The Enthoo Primo is an extremely watercooling friendly case, and it just didn't feel right to have a Corsair H110 CLC as the only "watercooling" component. Plus I was drooling at the aesthetics of the many watercooling builds, and it just seemed like a great way to kill some time and tinker with stuff, which I enjoy more than using the computer sometimes.

After spending god knows how many hours researching online, picking out all the technical drawings and doing a "mockup" with cardboard and measuring tape, I finally took the plunge over the Black Friday weekend. First I had to gather all the necessary materials:
<

Parts List
Alphacool Nexxos XT45 420mm radiator
Koolance COV-RP450 D5 pump/reservoir combo top + BDY-TK80/120 tube res
Koolance 380i CPU block
Koolance QD4 quick disconnects
Primochill Advanced LRT 3/8ID 5/8OD tubing
Swiftech MCP655 variable-speed pump (rebranded Laing D5 vario)
Deionized water and a secret ingredient which I will touch upon from Sigma-Aldrich (perks of being a chemist )
and god knows how many Bitspower fittings

If you care about why I picked these components check the spoiler. (tl;dr warning)

Spoiler

For my loop I decided to go the "max flow, lowest restriction" route, especially since I'll be using quite a few pairs of quick disconnects, which can add significant flow restriction if you pick the wrong kind. The XT45 is an excellent all round radiator with very low restriction that has superb performance regardless of fan speed. The Koolance 380i block might not be the lowest restriction, but offers great performance and is fairly consistent from one mount to another.

As for the pump, seems like Laing is the only player these days, with Eheim and Iwaki becoming increasingly rare. The choice was pretty much between the DDC and D5, and it was easy. A picture is worth 1000 words. As for why I went for Swiftech, wanted a D5 without a tach sensor.

When deciding between a bay or tube res, it seemed that bay res in general are harder to fill and drain, and have a harder time with bleeding the loop, so I stuck with tube. Then came the question of whether to keep separate pump and reservoir mounts or use a pump+res combo. Ultimately decided to go the combo route since it made loop planning much easier, plus none of the 3 pump mounting points in the Enthoo Primo case were ideal for my setup. The Koolance D5 pump top (COV-RP450) comes with a pump body cover, and offers very good performance for the money, so another easy choice.

Primochill's Advanced LRT tubing contains no plasticizers. You don't want plasticizers. And to wrap it up, Bitswpoer has become the time-proven gold standard when it comes to fittings. They do come with a matching price tag, but dammit if I'm doing a custom loop I'm doing it right. Besdies no point trying to save $100 on fittings only for them to leak and ruin a $3500 computer. And with that let's carry on...

Before I could install however, I had to fix a problem:


You see the case PCI bracket plate is riveted to the rear case plate, and it is just a hair too wide for the Koolance bracket so the screw holes don't align. Had to grind about 1mm off the Koolance bracket to make sure the M5 screws I bought would thread through.

 

The Installation
Started out by mounting the reservoir since this would set the tone for the overall layout of the loop. Since Koolance was too cheap to supply their own screws, I bought some M5 strip-resistant T25 Torx screws from Amazon and some wing nuts from Home Depot:


Those wing nuts were a real lifesaver, not sure how I would've tightened the ridiculously hard to reach screws on the other side otherwise.

With the reservoir done, I proceeded to install the radiator. What I thought would be the simplest part of the loop installation, a mere 30 minute job, turned out to be a 2 hour nightmare. It turns out you can look at blueprints and do mockups all you want, but there's never a subsitute for the real thing. I ran into my first snag right off the bat here:


See those 2 port plugs? Yeah so turns out Alphacool's "cool feature" of including 6 ports wasn't so cool after all. The plug bumps right up against the mounting bracket if I try to slide the radiator underneath, so this is as far back as the radiator would go. Thank god I didn't opt for the 480 radiator because I would've been mad beyond rad. No way it would've fit, look at how much of a close call the 420 is already.


Once this was settled I ran into another problem -- damn thing is too thick with fans pre-mounted to thread through the case, so unless I wanted to tear everything down IT WAS NOT GOING INTO THE CASE WITH THE FANS ALREADY MOUNTED PERIOD. Fine, so I mounted the rad, and then began a 90 minute journey to hell. MY GOD was installing the fan closest to the front panel an absolutely miserable experience. I may as well have been blind because I couldn't see a godamn thing, and one screw stubbornly refused to meet its destiny until I finally figured out the fan wires were pushing the fan just a fraction of a millimeter off-center, and so I had to undo the other 3 screws and remount this fornicator of moms. Finally after much sweat and enough swearing to make a sailor blush, I finished the rad installation. Didn't help I was doing this at 2am so I had to keep my swearing on the down low.


Moving on, the CPU bracket was a breeze compared to the radiator, although I did have to remove the retention mechanism (secured with T20 Torx screws) to instead the threading posts for the CPU block:

 

Finishing the Loop & Leak Testing

I must apologize for only having a picture of the finished loop here. I did try to take some pictures along the way but I had to change my loop quite a few times due to space issues or ridiculous tight bends or just because "it didn't look/feel right", and it becamse too time consuming to snap some shots every step along the way. Anyhow...


This will probably be very different from what you're used to seeing, mainly due to the inclusion of 3 QD4 quick disconnects. But hear me out:

From the start I planned to make my loop as "modular" as possible, so each component could be checked and replaced individually without having to drain and bleed the loop everytime, and so I went a bit nuts with the quick disconnects. The basic idea is that each major component (CPU, GPU, radiator) will be isolated with 2 quick disconnects, so each could be removed individually without affecting anything else. Since I definitely plan on watercooling the GPUs as well, it made sense to me to have 2 quick disconnects on the radiator side, since the top 420 rad will be connected to both the CPU and future GPU blocks. Also, the radiator's weight is much better supported by the mounting bracket, so an extra 260 grams (130 gram for each quick disconnect) wouldn't make a difference. But adding 130 grams of weight to a poorly supported GPU block just didn't seem like a great idea.

And before you say it, I know mating the CPU block outlet with the rear port of the radiator would've made more sense, but unfortunately those Dom Plats are just too damn tall, and without that silver those sticks look ugly, so I routed the outlet to the front port of the radiator.

Now that the loop is finished, it's time to fire 'er up for leak testing. But let me digress for a quick minute. Remember the secret ingredient I mentioned in the first post? It was fluorescein:


I'd be lying if I said I didn't add fluorescein for some added bling, but it's also extremely valuable in helping to detect leaks because the chemical is just so ridiculously fluorescent. Seriously, this is what fluorescein-doped water looks like under ambient lighting:


And UV (blacklight) completely lights it up:


With fluorescent deioinzed water in hand, I proceeded to top up the res and flick the switch on an HDD power adapter I bought a long time ago that came with a Molex good for 24W (yay for not having to jump PSU).


I started the leak test at about 8:40pm last night. For the observant among you however you may have noticed the res size got smaller. Indeed just before I filled the res I switched out the 120mm/237mL res for a shorter 80mm/157mL res, mainly because I didn't feel comfortable adding too much weight to what seemed to be relatively fragile Koolance mounting brackets. This however turned out to be a mistake which I will touch upon a bit later. For now I'm just going to say that I put back in the larger res at about 1am, and when I woke up this morning around 8 no leaks have sprung (knock on wood). If you count the 4 hours in between switching out the reservoirs, that makes it 11 hours. So far so good.

Well boys and girls, this is why you leak test for at least 24 hours if not longer. When I got back, everything looked peachy and towels around every single joint was perfectly dry. Just as I was about to call it a day I noticed some moisture on the paper towel I wrapped around the reservoir base. Turns out where the reservoir meets the base it was leaking. A very, very, very slow leak. So slow in fact it took more than 12 hours to show up. Will have to drain the loop and reseat the reservoir it seems, or might need new O-rings... At least I have some motivation to change out the bottom fans, which is now long overdue.

to be continued...

 

Nice... I'll be following I'll be glad to lend and mental help I can if you might have any question I can help with. I do t k ow everything but did learn a lot from my current build.

 

Cool! (Pun intended.)

I'll be following as well.

 

Chemist buys watercooling loop that looks like mad scientist chemical vat. #MakesSenseBro XD

 

Man make sure you triple clean and then triple clean those alpha rads. Trty are about the most dirtiest rads from a factory.

 

Oh yeah the rad was flushed with hot (~50C) tap water for 30 minutes in each direction (1 hour total). Saw a lot of junk accumulate on the port plugs which I promptly cleaned. Then did the radiator dance wih deionized water to flush out the tap water. Water was sparkingly clean by this point and so I stopped after another rinse.

 

Following

 

Second 24h leak test done, and thankfully not a drop spilled this time. Promptly fired up Prime95 to test temps and they're pretty good. Running small FFT, 3 cores run around 65C, 2 cores at 68C, and a single core at 72C. This is with 4930K @ 4.5GHz/1.39V. Pretty confident it's not a mount issue since the single hot core always did run hotter even when I had the H110 AIO. IIRC this represents about a 10-12C reduction in temps.

The computer is now so quiet, that the pump itself has become the loudest component. The actual pump itself isn't loud, but because I didn't pay too much attention to decoupling the pump, the case now emits a humming sound from pump vibrations. More work is still needed.

 

Time for another update. After trying various decoupling materials, I finally found that a combination of egg carton and hard foam (the kind that's not squishy and difficult to compress) worked the best in suppressing the very annoying high pitched ring that used to be present.



While the high pitched ring disappeared there was still a faint, droning hum. Interestingly, it seems there's an "acoustic sweetspot" for the pump. Turning down the D5 from 5 to 1 (lowest setting) actually resulted in WORSE humming. Somewhere between 4 and 5 the humming on my D5 seems to reach a minimum, and that's where I have it currently. Additionally, ramping up the fans just enough to mask the hum was far more fruitful than me trying to pad the thing with additional foam. Noise-proofing the case wasn't an option since the Enthoo Primo is so damn porous to allow for good airflow. Guess it's a double-edged sword if you're after absolute silence.

2 more shots of the finished CPU loop:


I also remounted the CPU block and improved temps by 2C. Prime95 small FFT doesn't go over 70C on the hottest core (other cores are 66C or below). After playing The Crew for 4 hours the hottest core goes up to 60C. Might not sound very impressive, but keep in mind the two GTX 970s are dumping all their 350W worth of heat into the case, since the only exhaust point is the radiator (all other fans are setup as intakes). The inside of the case is so hot that the water in the reservoir actually measured 35C! And no I know it isn't the CPU heating it up because even when running Prime95 for an hour the water barely goes past 26C. Moral of the story? GET THOSE GPUS UNDER WATER RIGHT NOW!

So it seems my Christmas shopping list will be two EK full cover blocks and matching backplates. EK is the only one making a full cover block for the Gigabyte 970 so despite my reservations about EK it seems I have no choice.

Stay tuned for a part two...

 

Egg... carton... and hard foam... how did you even attempt to begin to come up with this?

Also, is it just me, or is the posting method changed? It looks weird and quoting automatically takes me away from the thread page into a full editor.

 

Got the egg carton idea from Martin.

Tried the spongy soft foam first since I figured it would be best at dampening vibrations. Nope it was too compressible so even with an inch of that stuff the mounting bracket was pretty much right up against the case and transmuted most of the vibrations. That's when I realized hard foam that doesn't compress as much would work better.

 

Where people come up with this stuff I will never know. Ever.

 

Perhaps not the most pragmatic comment but it does look very pretty. Full black and blue. I do have to say that one tube of green looks a bit strange though...

 

Fits the theme perfectly

 

From using random house materials to see how they perform of course. The random things I did with my desktops with blutak and cable ties

 

I've heard legendary stories of people using butter as TIM.

 

I think you're talking about mayonnaise, and yes it's true--it actually works surprisingly well in a pinch, better than some of the no-name pastes on the market.

 

No, I literally mean "butter". I've been to places on the internet.... places I should never go back to... places where people compared butter, chocolate, margarine, mustard, ketchup, etc to various thermal pastes.

I've seen.... things....

 

You sound traumatized/scarred for life.

 

Stuff that you wouldn't normally use as thermal paste
Mayonnaise
Toothpaste + Denture Adhesive

On the other hand, this silver paste looks intriguing...

 

When Deep Cool Z or whatever it is performs worse than toothpaste, you would be too. Hell, they're charging MONEY for that stuff.

 

Almost there, so... close... yet so far away

Spoiler

Finished plumbing after a week of work, was pretty confident things were fine but decided to leak test because it's the proper thing to do. Hit the switch to the pump, and my heart sank when I saw the paper towel around the slave GPU block turn wet almost instantly. Then had a heart attack when I saw water BACKING UP ONTO THE GPU from the block. Promptly cursed, shut off the pump, drained the loop as fast as I could, and took out the GPU assembly before any more harm could be done. Took the slave GPU block off and gave the GPU a thorough wipe, then left to air dry.

Meanwhile I inspected the slave GPU block and found that I literally screwed up by cranking one of the screws in too tight which cracked the acrylic on the GPU block. Welp that's $125 down the drain. But thankfully due to the leak test I saved myself from a $3500 disaster.

 

Well GG.
10char

 

Water should not have done any harm to a GPU that is off so don't worry about that, leak tests are really the best way to go for sure.

 

You b*****d, just had to post that, eh !

Oh well, ordered a bottle ... for science!!!

Though 25g = 2.5ml; 10 cores. So I'll split it up 5-ways, if anyone's interested in testing that nano-stuff; PM me. Damn shop was ex-VAT, so total cost is €245/5 = €49,- (or $57) (inc. ww shipping, in envelope). Have some (vintage) glass syringes laying around somewhere, will add these for €8.70 (or $10) more. Otherwise it's a 1ml plastic syringe. And "no", will not post results before all four are sold .

Spoiler: 5ml glass syringe, Czechoslovakia - 1

Spoiler: 5ml glass syringe, Czechoslovakia - 2

Oh, right ... and good luck with that hydro-rig, looks awesome . And if there's threading with some of those water-circuit-components; might consider gas pipe tape for that, works perfect for sealing connectors. Hmmm ... now considering an 2m-tall-aquarium-pc-tower, with weeds+fishes and everything ... one more retirement project .

 

^Are you actually serious about the silver paste...

 

Sure, what the hell; I like to tinker/experiment. Bought a 1500W/33,000 lumen 'lightbulb' recently, just because it exists ... a high power incandescent would be 100W/~800lm, so ... probably need to wear welding helmet before switching it on (and a 1500W-capable switch).

 

Guys sorry for the OT... just a question... i'm looking for a fully modular PSU for my rig around 800/850W (it is a WIP project yet), what i'm looking are fully slaved cables (i don't like the flat ribbon ones that i have on my CM psu..). Which one?

 

@t456: I was being facetious when I linked the silver paste lol. It's essentially silver powder suspended in a filler medium (terpineol), and I really doubt it would work better than a true liquid metal formulation like Liquid Ultra.

I'll link you to OCN's PSU master list. Make sure you pay attention to the actual OEM and not the brand. Apart from Enermax, Seasonic and Super Flower who sell PSUs under their own brand, none of the other companies actually make the PSUs. Instead they likely buy the bare units and repackage them before selling. Be sure to also check the reviews where available. JohnnyGuru does some good testing but can get a bit too enthusiastic I find. Also it's rumored that he is affiliated with Corsair so for Corsair units make sure to check out other reviews.

From memory Antec, Corsair, and EVGA PSUs come with sleeved cables instead of flat ribbons, but check the reviews for other units. If you want best bang for buck, then the EVGA 850 G2 @ $109.99 after rebate simply cannot be beat.

 

Yes, I know , it's a lot like AS, except the particles are smaller by a factor two. Doesn't appear I'll be receiving that stuff anyway; they'll only sell to professional labs.

Found something else though; List of thermal conductivities > Graphene thermal paste > Graphene Thermal Paste 2nd Generation:

If true, that would beat Liquid by 3°C, will test that instead; out of Gelid anyway.

 

That came up when I was looking for a better paste than Gelid, but I'll just say unless it's a pure graphene paste (which it obviously isn't), I highly doubt it will be as effective as liquid metal.

Also I wouldn't compare temps across different websites, especially one on the manufacturer's own site. Just too many factors that can influence the results.

 

Liquid Ultra beats ICD and Arctic Ceramique 2 (if applied correctly) by a solid 10+ degrees, and both ICD and AC2 beat AS5 by a solid amount.

People seem to forget to list ambient temps and humidity in their testing. The most perfect test needs to be done in a controlled environment room with an air conditioner set to the same value all the time. This is often not the case and they won't say it. I've seen people claim MX-4 is better than ICD according to such tests; but obviously with my near-consistent room I have long proven otherwise.

 

Sure, and a proper test setup should ideally do without cpu/hardware altogether; heat is heat, regardless the source. Take a ceramic plate and a copper block, place them atop each other (weight=pressure, so copper block with sufficient/added weight) and heat the ceramic. A lot less hassle too ...

Hmmm ... if anyone wants to trade 1.5ml graphene paste for 1.5ml Liquid (don't have that); send PM. Can do our own test then.

While you're waiting for the plexiglass components; research conductivity and thermal capacity of the cooling liquid itself? There's lots of pro's and con's with all the nuclear reactor designs, so ... would bet there's some possibilities here as well.

 

So as it turns out, the other GPU block also had to be junked. It didn't have any cracks, but I could tell it was leaking since water got in places it wasn't supposed to go, and hooking the block to a tap confirmed just that. And just to be safe, I tossed the GPU bridge as well, so it worked to be a $300 mistake. :cry: Oh well. With that out of the way, time for some happy news.

I finally finished the damn loop last night and leaked tested for 16 hours. No leaks, but slave GPU still has some air that might take a while (a week?) to bleed. Other than that things are looking good. I'm going to go light on the pics for now, but eventually I hope to do a detailed build log when I get some more time. But for now enjoy these:




Preliminary results are nothing short of AWESOME!!!. Looped Unigine Heaven at 1080p Extreme for 1 hour, master GPU topped out at 36C, while slave GPU reached 41C. HWInfo showed CPU reached a max of 55C but while the loop was running the temp fluctuated between 43-47C. Best part? Fans on the radiator stayed around 900 rpm, so I couldn't even hear them. If anything the pump was louder than everything else LOL. See attached HWInfo screenshots. (chassis2 = bottom 360 rad fans, CPU + CPU fan OPT = top 420 rad fans)



Ambient temp was around 20C (68F), so hopefully I shouldn't more than 5-6C increase in temps during summer. I was initially concerned whether I had enough radiator space but seems like I have way more than enough. At this rate my loop should be able to handle 2x 300W GPUs, especially since fans are only running at 50% rated rpm to achieve the current temps.

 

Looking good I look forward to some benchmarks of it fully unleashed.

 

I did not read if you used acrylic or soft tubing. The air bubbles with work their way out eventually. Usually lean the PC in different direction while running and or speed up and slow down the pump. Normal use with the pump speeding up and down (if its PWM) will work it out. If you are manually controlling the pump then just alternate the speed over a few days and itll work out.

 

Sorry for the slight necro, posting preliminary benches from another thread.

Here are my results just for fun. All runs done at 1600/7800 with my own bios that clamps P0 to a fixed voltage. For 3DMark11 runs I was stable at 1.25V, but Firestrike needed 1.26V to be stable.


http://www.3dmark.com/3dm11/9408472


http://www.3dmark.com/3dm11/9408431


http://www.3dmark.com/fs/4023124


http://www.3dmark.com/fs/4023157


http://www.3dmark.com/fs/4023199

Have a look at this as well:


Notice how both of my 970s reached 90% of their power limit? This was mostly during test 1 in both 3DMark11 X as well as Firestrike. Gigabyte was generous enough to set the power limit to 250W in the stock bios so I didn't have to mess with that. But if you do the math, 90% of 250W means each card was pulling 225W!!! Might as well be running 780s at that point lol

Btw I have the highest 3DMark11 X score for 970 SLI, and in the top 20 for Firestrike Ultra as well. Not bad for an amateur bencher running Win7 I dare say.