24/7 Sub-Ambient Build

[caliusoptimus]

This has been on my todo list for about 10 years and the project is finally in progress. Basically, I've run out of other things to blow money on

I don't have a serious goal or high expectations here. Just building it all for fun!

 

SO. Here is the plan.

 

Phase one:

Water cool CPU and GPU. (done and working)

Find and experiment with a chiller unit. (chiller ordered, mods needed)

Connect and test the system above the dew point.

 

Phase Two:

Build an air tight case so that the internal atmosphere can be dried to minimal humidity.

Test case for static discharge problems and solve any issues.

(optional) upgrade chiller or add second stage for lower temperatures

 

Phase Three:

Explain electricity bill to wife.

 

 

At the moment I have a water cooled setup on my PC that should be sufficient for sub zero temps. I chose all fittings and components so that I won't have any compatibility issues with glycerol, ethylene glycol or propylene glycol. The GPU block and pump/res contain acrylic, and from what I can tell this should be fine with glycol. The fittings contain nickel plated brass. nylon, and Buna N o-rings and the tubing is PE. Subzero temps may cause issues with the acrylic or the PE, but I'll cross that bridge when I come to it.

 

I ordered a chiller today. Link to chiller. Elkay model ECH8. I chose this chiller for several reasons. The internal, insulated, 0.4gal reservoir is the main factor. Another bonus is that it has 1/4" copper lines for input/output rather than plastic hose barbs (see: aquarium chiller). Also, I got a great price! This chiller doesn't provide a BTU or tonnage specification but judging by the FLA and power consumption it looks to be a 1/4hp motor. Comparing to other units a 1/4hp system is usually rated around 2500-3000btu. The chiller comes with a mechanical thermostatic control which will be promptly replaced with a PID controller and SSR.

 

My current PC eats about 700 watts (prime95 avx, plus furmark, with a modest overclock for both) at the wall, with an 80+ bronze PSU. Based on this I estimate the whole system is pumping out about 600w of heat, PSU excluded. I'm pretty much a pleb when it comes to thermodynamics, but from what I can tell, this chiller should at least maintain ambient at full load. If not, I'll put this chiller under my sink and get a bigger one!

 

Anyway, photos and more details as the build continues.

 

 

 

[SolarNova]

@caliusoptimus

 

Ok so i've done loads of research and planning for a similar build.

 

here's what i know.

 

That chiller is expensive for what it looks like it is. at only 1/4 hp.

 

You really should find out what refrigerant it uses. if its 134a like alot nowadays, dont expect great subzero results even if it can go that low without bypassing the thermocople control. 134a has a relativly high boiling point for a refrigerant at roughly -26c, as your coolant temps get closer to the refrigerant boiling point, the efficiency of the chiller goes down, so dont expect to ever hit -26c, infact ur likely only ever gong to get down to at most -10c in ideal conditions (very cold air going into the chilelr unit, e.g durign winter).

 

Your case, best situation is to replace the air with an inert dry gas like nitrogen. This is what im going to do. Its cheap and slightly lighter than air so u dont have to get the case completly 100% nitrogen, just put the input nossle at the top fo the case, and exhaust nossle at the bottom and fill it a couple times till u get mostly nitrogen coming out the bottom. Once it settles you should only have a small layer of normal air at the bottom of the case. Design it so the PC components are above this area.

 

As a precaution, still apply some condensation protection to critical areas like aroudn the CPU. Dielectric grease in the socket, conformal coating around it, and some neoprene ontop. Build a Motherboard cover to make it look nice if ur into asthetics, or buy a motherboard that has an 'armor' already.

 

GPU is harder to do as u have very little space to work with between the PCB and waterblock, i sugest a spray on confromal coating and some thin neoprene or similar material.

 

As you go subzero, be extra cautious of O-rings, both fittings and blocks.

 

Acrylic will develop spider cracks around tapped screw holes ..be carefull. Idealy u should use full metal blocks or acetal topped ones.

 

I would sugest trying out glass tubing if u want the clear asthetic, or going copper for safety.

 

Build a large res so your chiller can be set to keep the res chilled all the time, even when the PC is of, this will allow the PC to run with instant chilled liquid. it wil lalso help reduce temperature cycling which can weaken the o-rings and tubing. Additionaly, not going straight from the chiller to the PC will lower the amount of time the chiller is on, as it will have a buffer.

 

Coolant. if your going subzero, consider Mayhems XT-1 Nuke concentrate. they do 5 litre bottles. it can be used down to as low as -50c, u wont get clsoe to that so u can probably run a 30/70 mix of coolant to distilled water for around -15c protection.

 

 

Feel free to ask more, i've done years worth of research into this kind of project. been put on hold atm though due to other life responsibilities.

 

Look forward to seeing how ur build goes

[caliusoptimus]

Thanks for the input! Yes, the chiller is r134a. I would have preferred a more powerful unit but I had to consider power requirements for both my PC and the chiller. My setup can draw around 1000w as-is (pc, monitors, desk lamp, speakers etc) so I have to be mindful of the 15A breakers in my house. I got a smoking deal on the unit, so if it turns out to be a waste, the chiller will find a perfect new home under my sink next to a RO system.

 

Instead of using an inert gas I plan to use a strong desiccant like calcium chloride to dry the air. That should keep the whole setup fairly simple, and should I spring an air leak the CaCl2 will have a color indicator to show if it needs to be changed.

 

It's gonna take a while to finish building the case for all this so in the mean time I'll do some testing with the chiller and see where I stand.

[caliusoptimus]

Here's the setup I intend to use with the chiller. I'll get it all hooked up and run it above the dew point for a while, until I find the time to build a sealed case.

 

prime x299 deluxe

7820x delid w/ conductonaut

evga 1080ti ftw3

 

I've currently got it stable at 4.9ghz 1.25v all cores, avx at 4.5. Temps range around 75c on prime 26.6. With prime avx 8k fft's the temps are in the low 90s.

GPU is stable at 127% power limit, 100mhz OC and never goes above 60c.

 

With 70f room temp the loop saturates up to around 35c, running furmark (no AA), and prime avx 8k fft's. All fans are on the same linear curve based on the loop temp, min speed at 27c, max speed at 32c.

 

If the chiller can keep the loop temp down around 20c I think 5ghz w/ 4.6avx should be no problem.

 

I chose 10mm push-to-connect fittings for the entire build. Great fittings, super easy to install and lots of different options for 10mm tubing. I'm surprised that so few people use them, considering the price is about the same as any other G1/4 fitting. I went with 10mm PE tubing, 8mm ID.

 

Right now the loop is in this order: pump > temp sensor > tee fitting (drain line) > cpu > 120x120 radadiator > gpu > 120x240 radiator > reservoir

 

Once the chiller is ready to go I'll remove the radiators and the pump/res combo from the case. I ordered a dual/series D5 pump mount to increase pressure in the system. The chiller uses 1/4" lines which are a bit small, so the series pumps should ensure a decent flow rate.

 

The chiller loop will be in this order: series D5 pumps > chiller > temp sensor > Y fitting > individual lines to cpu and gpu > both lines converge into a Y fitting > temp sensor > back to pumps

 

The chiller has it's own 0.4gal reservoir so no need to put another in the loop.

 

So close to testing... getting excited!

[SolarNova]

15 hours ago, caliusoptimus said:

~SNIP~

 

The chiller loop will be in this order: series D5 pumps > chiller > temp sensor > Y fitting > individual lines to cpu and gpu > both lines converge into a Y fitting > temp sensor > back to pumps

 

The chiller has it's own 0.4gal reservoir so no need to put another in the loop.

 

So close to testing... getting excited!

Its best to have a res befor the pump to ensure it never runs dry, especialy when filling. As i mentioned in my last post, its a good idea to have a large res anyway, to reduce Chiller on time. While the chiller has its own res, that res is where the chilling occur's and likely doesnt have a bottom port but rather 2 top ports, making primeing the system a PITA if u set the pump after the chiller.

 

My personal project plan, similar to urs . has a dual loop setup so that i can turn of the PC side and keep the chiller side on.

 

Loop 1

 

Large res* > Pump** > Chiller > back to res.

 

Loop 2

 

Large res* > Pump > PC loop > back to res.

 

* the same res

 

**pond pump e.g Eheim 1262 aka Eheim Universal 3400, or Eheim compact 5000)

 

Doing it this way also allows u to run the chiller side pump at the speed sugested for the chillers best operation, they usualy have much higher flow rates specifcations than what is possible inside a PC loop.

 

For example the Hailea HC2000 r-22 2hp chiller  (the chiller I once considered for my project, im now planning on building my own) has a flow rate spec of 3000 to 6000 L/PH . Good luck getting that in a PC loop considering most dont even reach 1 G/PM (~227 L/PH)

[caliusoptimus]

3 hours ago, SolarNova said:

Its best to have a res befor the pump to ensure it never runs dry, especialy when filling.

.....

Doing it this way also allows u to run the chiller side pump at the speed sugested for the chillers best operation, they usualy have much higher flow rates specifcations than what is possible inside a PC loop.

My thinking is that pumps produce heat, and I want the cpu and gpu to receive the coldest part of the loop possible. Probably wouldn't make a noticeable difference though, so I'll give it a try both ways and see what works best.

 

The "minimum flow rate" problem only applies to chillers with a certain type of evaporator. The chiller I have chosen has the evaporator and temp sensor directly inside the reservoir, allowing it to run properly with zero flow. If you do a deep dive I'm sure you could find one like this in the 2hp range.

[asand1]

Why not skip the water altogether and make the copper blocks part of the refrigerant loop?

[SolarNova]

31 minutes ago, asand1 said:

Why not skip the water altogether and make the copper blocks part of the refrigerant loop?

Your then looking at a pure phase change system which uses a different set of hardware, also at that point going single stage is a waste and you would likely just go fully into a 2 stage or higher system that can go down around -100c.

That kind of system isnt really usable 24/7, but rather a bechmark system akin to a Ln2 /DICE potted system.

[caliusoptimus]

Success!

 

Chiller arrived today. Did a quick swap, removing the mechanical thermostat in favor of a PID controller.

 

 

 

[caliusoptimus]

Got it hooked up and running! So far working great.

 

The maximum load I can get on the computer is running prime95 avx 8k FFTs and furmark (no AA) at the same time. This pulls about 700w at the wall with an 80+ bronze psu. The chiller was able to maintain 12c at full load, so that's good news. I didn't set it any lower as the dew point here is 10c at the moment.

 

With a more realistic load this chiller should be able to pull much colder temps!

 

 

The system was a pain to prime, I definitely need to add a reservoir to help capture the air bubbles.

 

The chiller is also noisier than I had hoped. The sheet metal enclosure amplifies the vibration from the compressor and the fan. The fat is also quite large and noisy, considering the small amount of air flowing over the condenser. I will probably repurpose my ml120 rad fans to go on the condenser. I'm also thinking about adding some foam to the inside of the enclosure to damp the vibration.

 

Here's a run with prime26.6. Temps for prime avx maxed out at 75c. Before the chiller these temps were about 15c higher.

 

 

To be continued...

[SolarNova]

Looks good, will be interested to see howwell it holds colder temperatures, especialy subzero.

 

Buy some 1mm , or the thinest u can get, sound dampening sheets and dissasemble the metal case. Reassemble with the foam sandwiged between all the joins, this will stop vibration between each joint. Normal thickness sheets on the inner walls will help reduce noise even more so. Also looks like you could unmount the compressor and remount it on some foam pads if u so choose.

 

Replacing the fans is a good idea but ensure your not acutauly downgrading, i doubt it will be noted on the fan itself but check to see if it has the CFM and pressure specs on it. It looks to be a pritty beafy thick fan, so replacing it with ML120 fans will likely be a downgrade. maybe consider just adding those fans as intake fans to aid the original fan.

 

Consider a igloo, or similar brand, cooler box (Those boxes used for keeping your beers cold, and for picnics), for the res. They are pre insulated and come in various sizes from , i think, 5 litres to over 25 litres , and can be easily modded for use as a chilled reservoir.

[caliusoptimus]

On 1/10/2019 at 9:07 AM, SolarNova said:

....Buy some 1mm , or the thinest u can get, sound dampening sheets and dissasemble the metal case. Reassemble with the foam sandwiged between all the joins, this will stop vibration between each joint. Normal thickness sheets on the inner walls will help reduce noise even more so. Also looks like you could unmount the compressor and remount it on some foam pads if u so choose....

I picked up some thin foam strips to wedge between the joints. Definitely helped, unfortunately the fan is still quite out of balance and is causing 75% of the noise. I'll get around to trying some different fans once I have time to make a duct.

 

 

On another note, I think 5.1... even 5.2 is within the realm of possibility without pushing below the dew point.

 

Prime 26.6

 

 

Prime 29.4 (AVX)

 

 

 

[caliusoptimus]

Lots of changes since my last update! Turns out this chiller was super cheap for a reason... the reservoir inside the chiller has been leaking. I didn't notice because it was a small drip soaking into the carpet. No damage, I found it and cleaned up promptly.

 

After some experimentation I've noticed that the flow rates through this chiller are very low. Even with three D5 pumps in series, running full speed, the 1/4" copper lines inside the chiller are just too small. SO... since the reservoir was leaking I was able to gut ALL of the 1/4" lines. The evaporator is made by soldering two 5/16" copper coils together, one coil for refrigerant and the other for water. The water side had 1/4" lines brazed into the 5/16" lines, and I was able to cut all of the 1/4" out which drastically improved the flow rate.

 

Since the reservoir was toast I made a new one. I picked up a 1 gallon vacuum jacketed water bottle and added some bulkhead fittings to the lid. The lid connection is pretty decent. The plastic is thick and all of the seals work quite well. Not sure what pressure my system is running, but it holds no problem. The res was extremely helpful in buffering any spike and dips in thermal load. With a 1gal res I can now regulate the water temp within 1degC without cycling the chiller too quickly.

 

 

I solved another issue by installing a flow sensor in the chiller. The original temp sensor was attached directly to the evaporator, which kept the system from freezing up. The problem with this location is that it turns the compressor on and off very often, and does not provide an accurate reading of the water temp. I moved the sensor into the PC case, right before the CPU which is the first component to receive chilled water. This way the chiller is regulating the temperature at the most important part of the loop. With the flow sensor I can keep the system from freezing up by shutting off the chiller when the pumps stop.

 

 

Overall it's working extremely well now. Last thing I'd like to fix is the fan in the chiller. I've got a plan, just gotta find some time to make the parts.

 

[caliusoptimus]

Woohoo! Fan problem solved.

 

I threw together some sheet metal and acrylic to make a shroud for some 120mm fans and so far so good. I measured the high side temp before and after to make sure the fans were adequate. With the stock fan high side was around 95F. With two ML120's the high side was around 100F, and with four it was back to 95F. I didn't take pic but I slapped a 12v meanwell supply inside the case and managed the fan wiring.

 

The noise level is SO much better. I am definitely satisfied. Next step is to make a sealed case and dip down below the dew point. TBC...

 

[SolarNova]

Looking good.

 

Shame about the quality considering its price, i wouldnt call that cheap for what it is but none the less, good job on the mods.

 

Look forward to seeing some sub ambent and eventual sub zero results.

[caliusoptimus]

On 1/27/2019 at 6:39 AM, SolarNova said:

Shame about the quality considering its price, i wouldnt call that cheap for what it is but none the less, good job on the mods.

 

For the chiller I'm only up to around $250-$300. The chiller itself was 200 "new" (but leaking?!). Everything else I already had laying about, accept for the fittings and flow sensor. The 1gal res cost about 80 bucks fittings included, but I would have needed that with any other chiller in the <$1k range. As far as cooling capacity, functionality and noise level... I feel like I'm ahead of the curve for <$400 all in!

[AnonymousGuy]

I've already been down this road and back.  Long story short: you don't get any tangible benefit to running chilled water, as you'll more likely than not be silicon limited before you're thermally limited vs. plain ambient watercooling.  You can't really get water cold enough to start changing the silicon's conducting properties like you can do with LN2.

 

It does end up costing $3 a day in electricity though.  

[SolarNova]

4 minutes ago, caliusoptimus said:

 

For the chiller I'm only up to around $250-$300. The chiller itself was 200 "new" (but leaking?!). Everything else I already had laying about, accept for the fittings and flow sensor. The 1gal res cost about 80 bucks fittings included, but I would have needed that with any other chiller in the <$1k range. As far as cooling capacity, functionality and noise level... I feel like I'm ahead of the curve for <$400 all in!

Ah , appologies, I must have been thinking of a different chiller >.<

 

Yes indeed that seem like a good price for a chiller. A 1/4 hp chiller from Hailea usualy costs around $500, how they would compare is a different story since the Hailea is most certainly a high flow rate chiller compared to ur lower flow rate one.

[caliusoptimus]

14 minutes ago, AnonymousGuy said:

I've already been down this road and back.  Long story short: you don't get any tangible benefit to running chilled water, as you'll more likely than not be silicon limited before you're thermally limited vs. plain ambient watercooling.

I came into this project without expectations, and so far it's been fun and interesting, which is good because that was my only goal!

 

That said, do you have a build log or something to show your assertions? Seems like there are a lot of folks who have gotten significant performance boosts from running a chiller. Even running above the dew point I have been able to stabilize my system at a higher clock than I ever could with radiators alone.

[AnonymousGuy]

I did the most ideal conditions possible of double water chillers, direct die CPU cooling.  Basically I couldn't get more than about 5.4Ghz truly stable, and you could do 5.2Ghz on air.  5.4Ghz required psychotic voltages that would kill the chip eventually.

 

Temperatures  were never an issue, it was simply "I can't make this silicon work any faster".

[caliusoptimus]

6 minutes ago, AnonymousGuy said:

I did the most ideal conditions possible of double water chillers, direct die CPU cooling.  Basically I couldn't get more than about 5.4Ghz truly stable, and you could do 5.2Ghz on air.  5.4Ghz required psychotic voltages that would kill the chip eventually.

 

Temperatures  were never an issue, it was simply "I can't make this silicon work any faster".

 

Nice job!! That was a good read, haven't found many threads going into such detail. I've tried some clocks over 5ghz on my current setup with the water temp steady at 12c. I was able to get 5.1 stable but the voltages were off the charts, so I understand your point. Did you ever push below 7c?

 

My plan is to build a sealed and dehumidified case so that I can push temps below 0c. I'm hoping this will enable some higher clocks without the voltage... but we shall see.

[AnonymousGuy]

25 minutes ago, caliusoptimus said:

 

Nice job!! That was a good read, haven't found many threads going into such detail. I've tried some clocks over 5ghz on my current setup with the water temp steady at 12c. I was able to get 5.1 stable but the voltages were off the charts, so I understand your point. Did you ever push below 7c?

 

My plan is to build a sealed and dehumidified case so that I can push temps below 0c. I'm hoping this will enable some higher clocks without the voltage... but we shall see.

Yes I went for 1C where I was getting condensation on everything (but I had <20% humidity so it wasn't dripping on everything) but it really didn't help much.  There's a 1:1 correlation between water temperature and max CPU temperature, so I would only get a peak of 60C or whatever at 1.55V.

 

I think you would have to go substantially further below 0C to get more clocks, since you're changing the silicon properties at super low temperatures.  I don't know if it's really possible to find a fluid that can handle that low of temps and still work with watercooling components.  At some point you might as well just do a phase change system.

 

I too was contemplating a purge box to not worry about condensation, but when I saw that 1C wasn't helping me really I abandoned the project and just settled for a giant external ambient watercooling loop that I run today.  Direct die watercooling by itself pretty much enables you to eliminate thermal bottlenecking, but it's a pain in the ass of modifying your motherboard socket, supporting the CPU substrate, making a custom waterblock mounting kit, etc.

[SolarNova]

Mayhems XT-1 Nuke can be mixed to work down to -50c.

 

[caliusoptimus]

4 minutes ago, SolarNova said:

Mayhems XT-1 Nuke can be mixed to work down to -50c.

 

Probably a PG blend.

 

Both ethylene glycol and propylene glycol can be used down to -50 or lower. Glycerol is good until -40ish, but it is the more viscous of the three and pumping it is a challenge even at room temp.

 

Propylene glycol is my choice for being non-toxic and also being thinner than glycerol.

[caliusoptimus]

Mini Update.

 

I've got a plan laid out for the case. I decided to (more or less) build it around an open benchtable. I ordered the OBT but thanks to Chinese newyear it's probably 3 weeks away.

 

Here's a rough draft of the case. Still waiting for the OBT to arrive before I can work on any of the finer details.

 

The case will be built from 1" square tubing. I'm debating between aluminum and stainless... aluminum is cheap, but not as pretty or easy to weld as stainless. Stainless is more expensive but probably a nicer end result. There will be sealed acrylic sheets on 5 sides (probably 3/16"), with a metal panel on the bottom.

 

All of the cables will be fed through the wall with various sealed connectors. The PSU will be mounted outside the case, to reduce the heat load on the chiller. There will be a small radiator, either 120 or 240, inside the case to cool the internal air. There will also be a tray in the bottom to hold some sort of desiccant like CaCl2.

 

I've been debating on how to control the internal air temp. One option is to put a ball valve on the radiator, so that I can adjust the flow manually and keep the case from condensing water on the outside. The other option is to use a temp controller to operate a solenoid valve to the radiator. This would probably be best as I'm sure the heat load will be variable... just adds another step of complexity.

 

I'll probably add a separate 12v supply to control the pumps so that I can chill the water before starting the PC.