Further reflections on watercooling

dankblog! 2022-05-01, 0349 EDT, at the danktower

I got some good feedback regarding my earlier "reflections on watercooling" piece, and I thought of a few things I'd missed, so here's some more dankwisdom. Take it for whatever it's worth. Please feel encouraged to send in further data for these tables, assuming you got the info directly from the manufacturer, or rigorously acquired it yourself (don't send me data you read off some web forum). I can be reached at nickblack@linux.com.

• Tubing (soft tubing anyway—if you'll recall, I have never used hard tubing, and don't intend to) exists in your loop for three reasons:
  • it is cheaper per unit length than metallic fittings,
  • it can be cut to arbitrary lengths, and
  • it is flexible.

Don't be tempted to directly connect components with metallic fittings. Unlike soft tubing, fittings are completely rigid, not admitting even sub-millimeter deviations. You're effectively working with hard tubing at this point. You might manage to get things hooked up, but check the base of your connections: you'l likely find that the last threads are poorly seated. This is a recipe for slow leaks. This means components must be, at a minimum, two compression fittings' lengths apart.

• Some general tips regarding soft tubing:
  • Check your compression fittings' o-rings for dirt, washing them if necessary. Damaged o-rings must be replaced. Without functioning o-rings, the fitting may not be waterproof.
  • Cut generously, then trim as necessary. You can't add length back to your tubing.
  • With compression fittings, I recommend first completing a compression mating on one end of the tube, then screwing this complete mating into the source, then screwing a fitting base into the target, then performing a cut, then putting the fitting ring onto the end of the tube, then putting the tube onto the (fixed) fitting base, then finally tightening the fitting ring. Don't try to complete both compression matings outside of the loop; screwing in one side will unscrew the other side, and if you attempt to hold the other side, it'll twist the tube.
  • If you're having difficulty getting the fitting ring onto the tube, run some warm water over the tube's end. It'll get in there eventually.
  • To tighten the fitting ring onto a fitting base in the last step, use your secondary hand to hold the tube fixed (it ought not move during the process), and your primary hand to rotate the fitting to the right while pushing it constantly towards the base.

• Knowing the volume of your components is useful for working the heat equations, for knowing how much coolant you'll need (particularly relevant when using concentrates), and also for determining completion when filling or draining. Unfortunately, this information isn't generally published. I've measured some components using a graduated cylinder at 22℃. The cylinder was labeled at 2mL increments, so no more than 2mL of precision can be assumed. This is unfortunate, as most fittings hold not much more than that. I would measure them with a more accurate pipette, except that it really doesn't matter; their volumes will be dwarfed by your reservoirs/radiators, and probably even by your tubing.

Component	Volume (mL)
Hardware Labs XFLOW 240 radiator	90
Hardware Labs GTR 360 radiator	320
Hardware Labs GTS 360 radiator	130
EKWB Aorus Master monoblock	45
EKWB EK-Vector waterblock	50
EKWB Quantum Kinetic FLT 240 reservoir+pump	265
DiyHZ sensor	4
Monsoon Series Two reservoir+pump	300
EKWB XTOP Dual D5 pump	35
aquacomputer highflow NEXT sensor	8
EKWB Quantum Torque 45° rotary adapter	2

For those of you who dropped out before the first grade, the volume of your (cylindrical) tubing can be calculated by multiplying the length of the tubing by pi by half the inner diameter squared. Area of a circle is πr², r is a radius (half of your inner diameter), and the volume is then area times length. If it's not obvious, this works no matter how you might curve or bend your tube.

Compression fittings negligibly effect the total volume. If you wanted to model their impact, each tube ought be considered shorter by twice the length of the barb (the extrusion inserted into your tube), indicating the volume occupied by the barbs. The barbs' volume in these regions is then equal to that length times, once again, the area of the barbs' interior. Using calipers, I measured the inner diameters of some compression fittings intended for 0.5in (12.7mm) inner diameter tubing:

Barb model	Inner diameter (mm)	Length (mm)	Volume (mL)	Δ (mL)
EKWB EK-AH	10.9	10.0	0.933	0.334
EKWB Quantum Torque	8.75	10.0	0.601	0.666

The non-linearity of the r² term is obvious here: the Quantum Torque inner diameter is 80.3% of the EK-AH's, but the total volume is only 64.4%. Either way, 10mm of the tubing is 1.267mL, so you're talking ¾ or ½ of that in the fitting. Assuming half-inch tubing, you can thus subtract 0.666mL for every Quantum Torque compression fitting, or 0.334 for every EK-AH. A 200mL reservoir would represent almost 600 times this latter value. You can use this same reasoning to calculate the (very small) volume loss at each fitting juncture—just use the inner diameter of the threading.

• When completely idled, a modern processor ought be using only a handful of watts. This is insubstantial heating at the surface area of an integrated heat spreader / waterblock. Ergo, an idled processor ought report temperatures not much greater than the coolant's temperature; this certainly ought be a constant. If an idle processor reports temperatures more than ten degrees or so over the coolant's temperature (modulo any systematic error in your sensors), you have most likely fucked up application of the thermal paste. If they don't, you most likely haven't; congratulations.

• Sensors are notoriously inaccurate. You can explore their relative systematic inaccuracy by placing them immediately after one another in series (this can of course be changed later). Sensors like to be level relative to gravity.

• Quality radiators will generally be designed such that the screw holes don't sit atop any water-carrying channels, or that a plate sits between the screw holes and any such channels. It is otherwise possible, however, that screwing in too far will pierce a channel. Use screws of the proper length for your radiator and whatever you're binding to it.

• Your loop is never a vacuum. This is good, as nature would otherwise abhor it. Assuming you're not actively asphixiating, you're working in at atmosphere of air, and air fills the regions of your loop not filled by water. In particular, when you've first hooked everything up and before you've begun to fill the loop with water, your loop is filled with the equal volume of air. Water cannot exist where air already does, and thus air must leave your system for water to enter. There are three ways to accomplish this:
  • Use a powerful vacuum pump to evacuate the system. Contact your local HVAC technician to explore this route.
  • Ensure air can leave from the same place you're adding coolant. This can be accomplished via filling with a graduated cylinder, a filling bottle, or anything else with less area than the inlet, and ought purge any air below the inlet. Notably, a funnel attached to a fitting, screwed into the inlet, will generally block the entire area and not allow air to leave.
  • Ensure air can leave from somewhere else, some outlet. This ought purge any air below the outlet. Once coolant reaches the level of the outlet (or the inlet), further filling will expel coolant, rather than air. This is also known as "making a mess", and is undesirable.

It is hopefully obvious from this that your filling inlet ought be as high in the loop as possible, and essentially must be in order to quickly remove air. This ought be considered when designing your loop. If it's impossible to get it there, rotate your case while filling.

• It is natural that pockets of air will persist in your system after filling it. Microbubbles will eventually diffuse naturally out of the system, but larger bubbles need your help. Air, being lighter than water, naturally moves to the local top of a section of the loop. It is thus advisable that your reservoir (and filling point) be as high in the loop as possible. Rotate the machine while the pump is running, attempting to coax the bubble to your reservoir (or any point where you can allow air to leave). Eventually, the air will leave (often with an audible burp), and your coolant level will fall. Top off the coolant (this is why you wanted the air to leave via your refill point), and live happy.

If you calculated your total volume accurately, and track how much fluid you've added, you can know to your satisfaction when you've eliminated all air. This can be useful when you have opaque volumes (like most radiators—I've never seen a transparent radiator, don't know why, it would be cool. I guess you can't get highly thermally conductive clear materials?) at the top of your loop. With that said, it's pretty difficult to get your total volume calculation accurate to more than a few mL, which can be a significant amount of air if it's in e.g. a waterblock.

• Pascal's principle states that a pressure change at any point in a confined incompressible fluid is transmitted through the fluid such that the change acts everywhere. This has various effects:
  • Opening a horizontal port lower than other fluid in the loop will see fluid leave that port until the heights are equalized, once again making a big mess.
  • You can't fill above the filling point without rotation or pumping.
  • You want your draining point at the bottom of your loop.
• You'll sometimes read that "each fitting reduces your flow rate", presumably relative to the same length of tubing. This is only true for angled fittings, just as it is true for kinked tubes. Similarly, the resistance to flow is greater as the angle incident to the direction of travel. Consider a tube, or a linear fitting: travel is 180° and unobstructed. Using a 90° adapter, flow is affected more than a 45° adapter, and less than a theoretical 135° adapter (I have never seen such a thing). This is similarly true for tubing, but exacerbated by the area lost to kinking. A tube traveling an effective 90° path but free of kinks (perhaps due to anti-kinking coils) ought obstruct flow no more than a 90° fitting. Of course, the fitting will likely introduce less volume, since it performs the turn in less length.
• Flow rate is affected by some interesting non-linearities. This article from EKWB is useful. I've combined their data below (rather infuriatingly, they didn't supply measurements with zero fittings):

PWM level	Flow (1)	Flow (3)	Flow (5)	Flow (7)	Flow (9)
25%	249	220	210	200	190
50%	510	460	420	410	390
75%	940	870	810	770	720
100%	1470	1340	1260	1180	1120

Here's the same data normalized against 1500:

PWM level	Flow (1)	Flow (3)	Flow (5)	Flow (7)	Flow (9)
25%	0.166	0.147	0.14	0.133	0.127
50%	0.34	0.306	0.28	0.273	0.26
75%	0.626	0.58	0.54	0.513	0.48
100%	0.98	0.893	0.84	0.786	0.746

Some properties are immediately apparent. I'm surprised EKWB didn't point these out in their writeup, but hey, that's why I'm here:

• Adding even nine 90° adapters is, at all reduced power levels, more than overcome by a an increase in power to the next 25% level.
• Across all four measured power levels, adding eight adapters (nine total) cuts a little over 23% of the flow relative to a single adapter. This is remarkably consistent: 76.1%, 76.7%, 76.5%, and 76.5%. Of course, these are absolutely larger as flow increases: the 23.7% reduction at 100% represents a loss of 350L/h, whereas the 23.9% reduction at 25% represents a loss of only 59L/h.
• As more fittings are added, the reduction per fitting decreases. This correlates perfectly with our statement above: the obstruction to flow presented by any unit results in a greater loss of flow for greater flow rates.
• Adding more power increases flow rates more as the total power level increases. I.e. going to 50% from 25% PWM adds less flow than going from 75% to 100%. This is contrary to my intuition.

And one final, probably controversial point, from my own experiments: quality waterblocks are very effective even at low flow rates. According to my two sensors, I have never managed to exceed 1L/min in my loop, even when using two D5 pumps. My loop is quite large, and climbs over a meter through the height of my very large case, but this still seems a lower flow rate than one might expect from two D5 pumps (admittedly at low power settings). Common hearsay is that one wants at least 0.5G/min (1.9L/min); I rarely manage even half of this. Regardless, my temperatures are excellent, both when measured absolutely at the heat generating sources and when the coolant is measured relative to ambient. Optimizing for flow seems a foolish errand, and ought be of only secondary concern when designing a loop.

previously: "reflections on watercooling" 2022-04-17