Let me start with an introduction about this build.
I apologize this post is going to be very long and probably annoying, I know
If you are brave enough to read from now till the end, you probably need a good shrink!
But since it's an insane build, I feel that I have to explain a bit the genesis of this insanity.
I'm going to be very detailed as I'd like to be every build log.
Usually after a while I have hundreds of questions... why he did that? What's the goal? How he did it?
Hope it's going to be at least a bit different than the usual this time.
Took me a lot of days and effort to write all this stuff but it has become an handy diary. I'm starting to forget stuff.
Hope you can find some useful knowledge from my mistakes if you end up adventuring with TECs.
Many things will be obvious if you have some experience already but it wasn't for me.
I recently retired my beloved i4770K; did many years of happy overvolting at 1,38V and oc at 4.4/4.5 GHz, delidded with liquid metal between the die and the heat spreader.
It was still running pretty well but a bit outdated with only 4/8 cores.
Also it doesn't run anymore at standard voltage, electromigration degradation kicked in already some years ago. It was due time.
It came back handy now cause I'm using it as a test system, it was piling dust on a shelf.
I'm planning to use it as replacement for the FX-8320 in the HTPC in the future but the cooling part could be challenging
I decided to give a chance to AMD and bought a Ryzen 3800x and a x570 AORUS Master.
It's an amazing processor but very annoying to tweak and always on the edge of thermal throttling.
Until it actually goes into it and whatever you put there it's always too early.
There are multiple "gates" were you see the performance dropping; 40, 50, 60, 65, 70, 75 degrees celsius etc...
You can tweak it to get amazing perf at 40-60 but then if it goes over 60 it's suffering, over 70 it's pulling the hand brake, over 75 starts sinking.
What could I do to improve this situation...
I really don't like maintenance so I've always preferred the air cooling solutions, except for a while using a Swiftech x220 AIO.
Was nice till it lasted (then clogged and briefly after mounting on another PC started leaking) but not giving much more than a good air cooler.
Right now I'm using a beQuiet Dark Rock Pro 4; it's a wonderful cooler, extremely quiet but also barely enough for a 3800x.
My first choice was a 3950x, then a 3900x but neither were available so I went for the 3800x.
At the end was not a bad move; I would have needed something better for cooling from start.
Now I'm waiting for the future 4000s but the cooling issue would just get worse; I want something more beefier of course.
So there it is my first requirement forming out; keep the processor at least 5-10 degrees lower and avoid huge spikes.
I started investigating the WC options but I wasn't willing to change my case, the Corsair 750D Airflow.
There is space for a 360mm rad on top and a 240mm on the bottom.
Well, the temps with these configurations I found online were better but not that much...
Plus I didn't realize my EVGA Supernova G2 PSU is so big that the 240mm rad on the bottom wouldn't fit.
So the only option left was a 360mm on top plus a 120mm on the bottom. Maybe a slim 120mm on the rear.
Even so, with the water temperature limited to ambient and the unavoidable build up, I barely would have reached my goal.
Really good wc builds requires an impressive amount of rads and fans otherwise they get quickly very noisy.
I had to break somehow the barrier of ambient temperature... whatever is the cost.
That's the point were I started again scouting for TEC builds
I never stopped dreaming about a viable cooling solutions with TEC but I remember only one commercial attempt and want's that great.
I've always kept an eye on interesting builds and have always been almost there to start mine... but never enough.
My only "build" was when I was a kid, more than 30 years ago. Something crazy and dangerous.
My father forced me to dismantle it and probably saved my life.
The custom builds I've seen so far are almost all designed for massive fixed overclocks aimed at bench, very long pre-cooling, staggering power consumption.
Extremely difficult to build and to maintain. I live in Germany now and I don't have any tools like at home in Italy (well now I have quite some...)
Milling a waterblock to fit a TEC inside and fixing it with epoxy is not what I was looking for.
I had to find an easier design which could give me good performances with an easy maintenance. More dynamic than powerful.
The TEC only used ona needed basis and the power consumption in idle between low and acceptable.
And compact, very compact. Keeping my Corsair case without any external units. An almost impossible goal.
Sourcing high power TECs has become quite easy. There's plenty now and if you live in the US they can also be quite cheap.
But the controller is always a pain point.
I've seen many attempts over the years but at the end the only good option I've seen is the Arduino PWM controller project shared by Krow.
At some point I noticed something I just didn't knew; many TEC datasheets like the Laird were stating: "DC operation".
I thought driving the TECs with a PWM controller was the best option... but it's not. Maybe many years ago.
The extremely powerful and expensive bench controllers are DC. Very expensive... guess that's the main reason to use PWM, it's cheap.
Was there something else other than a big bench controller that could be used in a PC?
Is there something industrial grade that can be easily integrated? What are they using in aerospace, avionics, medical grade equipment?
All industries were they make a constant use of peltier cells and they work, they last and they perform.
Well I did a Google search and exactly 3 seconds later the first result was Meerstetter engineering from Switzerland.
Shameful I never ended there before in the long hours scouting around...
These are relatively small amperage controllers mostly made for bio medical equipment, they go in tandem with laser controllers.
And they have tutorial videos:
Complete and detailed datasheets for everything.
Software for control with constant firmware updates.
A whole set of API to control the device, USB and serial ports, GPIO.
Auto tuning of PID controls, internal tables for scripting.
I'll talk about the features and how they can be useful in this scenario later.
Less than 30 minutes later I was done, I had to get one at all costs and play with it.
Guess at this point you understood why I named the build "MeerTEC"; it's all around the Meerstetter controllers.
They promise reliability, speed and an incredibly low power draw. I decided to check myself the claims.
I bought the TEC-1089-SV model, 10 Amps up to 21 Volts and a bunch of random stuff to make experiments and find a good design for the heat exchanger.
Didn't even consider air coolers for the TECs, the heat side dissipation is crucial and I knew that.
Not enough space for tons of fans so I went directly looking for a dual loop setup.
For the first prototyping setup I had in mind, I sourced these:
- 360mm AIO for the cold side, planning to dynamically exclude the rad by electromagnetic valves. Bought a super cheap Alphacool Eisbaer LT360 AIO, very expandable stuff but already with quick disconnect fittings
- 120mm rad for the hot side, took an Alphacool Nexxos Monsta since I had the height at the bottom of the case
- 2 x Alphacool D-RAM x4 for the heat exchanger
- Laing DDC-1T Plus PWM plus 5.25" res
- Bunch of copper plates with different thickness
- A couple of Laird UT619F14040 and a couple of Marlow RC12-9 TECs
- DC-DC converter Victor Orion 12-24/10
This was mostly to build experience on how to use the controller and the TECs, find a good design for the heat exchanger.
Here's a quick picture of the heat exchanger and the test setup:
A few days later I was already in love with the controller.
And I realized one of my main requirements was unrealistic; to keep the Corsair case.
I could have found a setup to fit everything inside but it would have been busier than Istanbul's traffic.
Did I mention that I hate maintenance right?
Well maintenance would have been a disaster to say the least.
So I had to revise my requirements and ended up with these:
- Dynamic, 5-10 degrees benefits with acceptable power consumption
- Max 40-50 degrees peak CPU temps with moderate power consumption
- Switchable, system should work with or without the TECs active
- When needed powerful enough to drop the temps down for some serious benching
- Completely self enclosed in a single case
- Relatively easy to build
- Easy maintenance
- Easy TECs replacement
- Components sourcing not too difficult
And here I'm, with an Enthoo 719 sitting next to me, ready to be packed in with already too much stuff.
Of course I fell in love with the controller and I just can't get enough.
Didn't even finished thinking about the v1 design and I already want more, addicted already.
Cause heat is a bad beast; the water in the loop can do just as much.
The CPU waterblock gets heated quickly when the load starts and builds up over time.
Cooling down the water at the inlet takes a bit of time and, even if it's fast without pre-cooling the loop, there's an unavoidable spike in temps.
Yes mostly a sudden 100% heavy load like OCCT starting. But I want to cover this too.
If you spend hundreds of watts in cooling you really want to make it worth.
What if the load after a little while goes down and you've already spent hundreds of watts to compensate?
Not very efficient.
I went for the Watercooling Heatkiller IV Pro Threadripper to handle the massive heat generated by the TECs.
It's beautiful, sturdy and full nickel-plated copper.
Obviously the best mate to pair for the Ryzen waterblock is an HK IV Pro AMD.
The only solution I could see to make the system more dynamic is to cool down in parallel the CPU waterblock with another TEC.
That's where the build went from crazy to totally insane. Cause this means buying another controller and all that comes together!
I'm going to replace the controller for the loop heat exchanger with a TEC-1090-HV which can go up to 16A at 30V.
It's even more expensive than the 560 euro I shelled out for the 1089; I'll use the 1089 with a single TEC just for the CPU wb.
In retrospect I should have spend a bit more and took that one from start.
I could have 2 interchangeable controllers. But now it's done
Apart from the staggering peak power consumption, this controller to really shine needs a higher voltage DC input.
Which is not a trivial problem, I will cover the power feeding issues later.
Scouting for a CPU wb that is flat on top where you can slap a TEC on top I realized there were none.
Even if there was hardly could beat the performances of the HK Pro. Not an acceptable compromise.
All the good blocks have the inlet/outlet ports on top.
But the HK Pro in full copper has also nice flat surfaces around the fittings...
I have bought but not yet installed the excellent Roman's Ryzen 3000 OC bracket custom mount.
The crazy idea is quite simple... hope it's going to work.
I'll explain later in detail but in short I'll use the OC bracket mount with longer screws to sandwich one HK IV Pro AMD on top of each other.
With a TEC in between if course.
With 10 amps, according to the tests I've done, should be possible to bring down the CPU wb by 1 or 2 degrees per second.
Cooling down quickly to the desired temp target the object is where this controller really shines.
And it's going to be much easier than controlling the cold side temp to adjust the water temp.
And here's the v2 design which will probably needs much more time to complete than the v1.
You should have the Visio diagrams in attach to this post.
You should carefully make your Visio before ordering all the parts; not like me.
I've been lazy and updated it after the last order and obviously found out I've missed to order some stuff...
The core of this build is obviously the software development to manage the controllers.
I'll keep it in a separate post to track the requirements, strategies, features and development status.
It's probably going to take an epic effort and a few years to complete.
Hope with this post to convince some other nuts to join me in the endeavor
This is project is really a lot of fun especially now with the human malware forcing me to stay almost always at home...
I highly recommend to do something crazy to fight it, time flies away.
Sourcing the parts and materials has been a huge pain but in the near future the situation should improve, hopefully.
Now let me talk about each topic in its own chapter.
THE ENTHOO CASE
The Enthoo 719 is a wonderful case but the metal is very thin and bending like plastic.
I can't complain too much considering the price.
In the reviews they say once everything is fitted in it's not anymore noticeable.
The manual is pretty good but I also ordered the internal 5.25" ODD bay.
I didn't consider all the constraints, it'd better if it was covered in detail.
The restriction on the front radiator mounting is pretty evident.
I had doubts about the side radiator but there's a lot of space on the back to fix a fan.
What I didn't consider is the restriction on the top radiator.
There's just 25mm clearance so even a 26mm thick 360mm radiator wouldn't fit.
I choose a 20mm thick 360mm rad with 2 fans instead of going down to a thick 240mm.
This should leave me some more room to mess around the CPU.
Sorry it's not going to be a nice and pretty build for photo-shoots.
I don't care at all about aesthetics, it's not my thing.
Performances and efficiency it's all that matters to me.
I'll have some RGBpx LED strips cause they can be programmed to show the system load with the Aquaero.
And that's something I consider functional.
I went straight ahead for the dual hot/cold loops setup.
My goal is to have the TECs kick in only when needed; Just In Time model.
I don't really have much experience in watercooling, this is my first custom loop.
In the past I did help some friends but if it's yours it's very much different.
Don't know exactly how to handle the fill and drain ports and how to handle air bubbles..
I'm going to have a reservoir on the hot loop but none in the cold loop.
Not sure if a tube filling solution is enough or not.
I have one fill port which I want to split with a Y-connector.
Since I can't have the cold loop water mixing with hot loop I've taken a couple of choke valves.
But I'm not sure if the gravity is enough to let the water without pressure goes over it.
Those will definitely go into the last section with the open points and questions.
Any help is appreciated before I make a mess
In the Visio in attach you should see more or less what I'm planning for the loops.
It's probably going to be not very accurate...
For the pumps I went full Aqua D5 PWM.
On the hot loop 2 in serial with the EK-XTOP Revo Dual plus paired with an EK-RES X3 150 Lite reservoir.
On the cold loop only 1 paired with an Aqua D5 top.
For experimenting I bought a Phobya DC12-400 pump, I had to replace it with a Laing DDC-1T Plus PWM.
As I had seen in many Amazon reviews, the Phobya did work a couple of hours then started grinding like a cross bike and the flow went up and down like a roller coaster.
I still had hopes to keep the Corsair 750D so the form factor was interesting. But the quality indeed is sub-par...
The Alphacool Eisbaer LT Solo pump will deserve its own chapter.
For the rads in the hot loop I picked:
28/04/2020: Change of rads placement.
- 480mm front mount on the inside with 3 fans, Black Ice Nemesis GTS 480, 30 mm thick
- 360mm top mount with 2 fans, XSPC TX360, 20 mm thick
- 360mm side mount on the inside, XSPC TX360, 20 mm thick
Since they delivered the 480mm rad today I took 10 minutes to test the placement.
Side position on the outside doesn't fit; it's too big in height. The 530 mm in the manual is the max for the inside...
Can't be mounted either on the inside; it's blocking the fill port. Why it's not mentioned in the manual? Damn.
Didn't thought about that. You can't put anything in the inside if you want to use the fill port, it's a very bad placement.
Cherry on top I didn't consider properly the ODD bay limitations, again.
On the inside there's only 1-1,5 cm for the fitting behind the ODD frame, the fitting would not fit...
So I moved it on the front inside position. It's not as planned so the fans will cover partially the 360mm rad on the side.
The rad top port will end up magically just above the ODD bay and seems there's a lot of clearance there.
But I have to drop another fan and this is not nice...
I'll mount the side 360mm on the inside where it should end below the ODD bay.
The fans on the outside and this should not cause troubles with the fill port.
This ODD bay is costing me a lot of sacrifices but I can't drop the Aquaero 6 XT display...
The documentation and the fill port position are cracking already my love for this case.
They are all well performers at low fan speed but aren't that bad at high speed.
I plan to dump on them the massive heat from the TECs only on special occasions...
On the cold side the Alphacool Nexxos Monsta 120mm.
This was a surprise and revelation.
I wasn't expecting that much considering the single 120mm fan.
Took it only cause it was fitting well in the Corsair case.
But it does completely blows away the terrible LT360 radiators with 3 fans which are much more noisy.
At the beginning I used it only for the hot loop.
As it was supposed to, the 360mm rad in the cold loop it's a disaster when you go down ambient.
The huge surface is fighting against the heat exchanger and heating the water toward ambient temp.
It's too many degrees lost and it nullifies the controller's high efficiency in idle.
At this point I started considering how to exclude with valves the 360 rad.
Then I started testing the cold loop without any rad in it.
At some point I decided to see how bad was with the Monsta in.
And it's not bad at all.
Yes it holds a lot of water and dissipate something but it's almost negligible.
Once the water is cool down the controller can easily idle at 5-10W.
The key is to avoid the fan to start; otherwise it will quickly dissipate the lovely cold into ambient.
I'm planning to control the Aqua to switch to a profile where the Monsta fan will start only in an almost overheating scenario.
Which would mean the TEC controller failed.
I'll use Arctic P120 Bionix PWM on the Monsta and on the front 360 rad.
For the top 360 and side 480 I'll use Corsair ML120.
THE EISBAER LT360 AIO KIT
This deserves a specific section.
I didn't really think much about it when I ordered it.
It was the cheapest and most expandable, yet with quick disconnects for expansion.
Awful, terrible, real rubbish. The first kit came with a broken pump.
The second kit pump too has issues; sometimes it doesn't start.
It's just really bad and under-performing, the radiator is awful too.
I wish I took the non LT version but I think it wasn't available at the time.
The Aquaero is the second most important component in the build.
It's simply amazing a must for any water cooling build.
Wish I had done it earlier only to use it.
Can only thanks with my whole heart Shoggy and the Aqua team for these amazing products.
And for the patience for my stupid questions of course
The logging and reporting capabilities are top-notch.
I'm setting up the profiles with the test system now.
It's nothing short of amazing. I'll post some test results I've done during these days.
For the controller software I want to control the Aquaero profile and import and export the sensors values.
I have great expectations, hope I'm going toi be rewarded by the results.
THE HUMIDITY SENSOR
I bought a Sensiron SHT31 Bluetooth eval board.
It should run for a couple of years, I guess, with a button cell lithium battery.
The humidity sensor is very precise, below 2% error; should be more than fine to calculate the dew point.
Obviously I don't trust the Bluetooth connection...
I'm planning to replace it maybe with an Adafruit Feather board with the same SHT31 sensor.
THE HEAT EXCHANGERS
I've tested some combinations, mainly to keep a small form factor, and almost all were a bad idea.
There are 2 main factors that will contribute to an early death or under-performing TEC:
- Wrong assembly
- Overheating of the heating side
Needless to say, did both and killed the Laird UT9s and one Marlow as well.
Not a big pain since I had considered them already dead on arrival...
Mixing materials in the assembly is quite challenging.
I've tried with plastic parts, mostly to decouple the hot side from the cold side.
They need to be extremely rigid otherwise the TEC will not adhere to the sinks well enough cause of the bending.
It's very hard to find the right parts if you are trying to re-use whatever you have at home.
I was worried about the massive heat under load, with 2 or 4 TECs close to each other you'll have a hotspot where they are close.
But with a good waterblock to remove the heat if you g a thick heat capacitor it can be managed.
At then end the decoupling is important but not critical; I put my hopes in a decent insulation.
With a bunch of silicone washers the thermal transfer from the screws is contained, works pretty well.
It's recommended to use a spring washer and then a fiber insulating washer, quite hard to source them at M2 size but at the end I could find them.
The screws are not stainless steel but I couldn't find stainless quality at M2 size. I'll have to keep an eye for rust...
For the cold side I went 2 Alphacool D-RAM x4 blocks; they seems perfect for the job.
Spent few hours to find them and they found out someone suggested them already in the Chilled Water Cooling build thread. Ouch...
They have a large base where you can fit easily 2 or 4 TECs 40mm.
Full black nickel plated copper plate with an acrylic top.
The acrylic top is very important otherwise it will take too much time to reach the set object temperature.
It'll make the block more efficient with more time and high loads but less dynamic.
Full copper will drain more power to reach the target and after the load more power to regain the set temperature.
As heat capacitor I used a copper plate 0.5 mm thick size of 14 cm by 14 cm; it's pretty easy to order it online.
The plates from 0.3 mm thickness and up are not usually cut in sizes below 10 cm, so take it into consideration.
I drilled the holes for the M2 screws with nuts to mount the waterblocks with a 2 mm metal bit.
For the copper sheets I use https://www.ebay.de/str/onlinebleche
, for the plates https://metallstore24.de/
They have both high quality copper, mostly precise cuts and very quick shipments.
Even in the most harsh conditions with a subpar 360mm radiator I could manage to keep the heat plate below 37 degrees with 29 ambient.
The power draw and achievable deltaT are in direct relationship with the heat side temperature, it's imperative to keep it down as much as possible.
For the heat side I'm using a Watercool HeatKiller IV Pro Threadripper.
It's an amazing block; given the right radiators afterwards it can dissipate a massive amount of heat.
In my first tests I've only used up to 7 amps but didn't look like it made a flinch.
I hope it can hold also with the v2 design with the 52mm Laird cells at 16 amps.
The layout is really simple so it wouldn't take that much to change and improve it if needed.
Being so easy also have is an obvious advantage for the maintenance and repairs.
I'm using, at least right now, quick disconnect cables so I can exclude and remove almost every component on a need basis.
The heatsink NTC thermistor is placed near as possible to the hotspot on the copper plate.
I have also another Aqua NTC thermistor on top the HK block.
The PTC sensor for the object temperature is placed on the 2nd D-RAM block which is the water outlet, near the center close to the TEC.
The two D-RAM blocks are kept at about 5 mm distance; the first one cools down the liquid, the second will freeze it.
It's pleasure to feel the inlet fitting burning hot and the outlet cold like ice
Right now there's also another Aqua NTC thermistor right next to the PTC sensor from the controller; but it should become redundant later with the controller software.
I selected almost randomly two sensors in the recommended list by Meerstetter.
An TDK NTC 10K Ohm bead PN B57871S0103F001 from Digi-key and a PTC LabFacility DM-314 from Develektro.
The NTC bead turned out to be too big while the PTC sensor too small...
Didn't want to fix them with epoxy glue, to keep it easy for maintenance and upgrade.
I'm using the Tesa Aluminum tape:
If you apply it in dry conditions sticks perfectly and for a long time (at least for now)
Now the challenge will be to replace the 2 Marlow 40mm TECs with the new 52mm Laird.
Since they don't fit below one D-RAM block, I had to be creative.
I received a copper sheet cut 9 cm by 10.5 cm; I'll place the TECs in the middle, facing the TECs cold side.
This should allow me to distribute the screws pressure among the two TECs.
It'll take a bit more time to cool down the D-RAM blocks but it should be compensated by the much higher power of the TECs.
Highly recommended to buy a torque controlled screwdriver, I bought a cheap one from Proxxon on Amazon and seems working fine:
As thermal paste at first I used the Kryonaut on both the cold and hot side.
I had read here on one thread someone suggesting the Arctic Céramique 2 for the cold side instead. Thanks!
Great advice indeed; the Kryonaut despite the specs claims it's terrible on the cold side.
I found out while experimenting that it was literally vanishing after a while, what left in dust.
The Arctic is performing much better, now I use the Kryonaut only on the heat side.
Without thermal insulation and 2 Marlow at only 5 amp each I can easily keep the water temperature a +6 from the object target.
With pre-cooling and low load it goes down to a +4 dT, I guess these are very good numbers already.
Going for the 52 mm TECs also made me reconsider the D-RAM blocks; they are a better fit for 40 mm.
A you'll see below I'm going to use a couple of HK IV Pro for the CPU.
I'm wondering if they could be a better fit also here.
I could get rid of the copper layer and probably also improve the water flow.
But they have, roughly, a 3rd less of contact surface with the water.
Maybe in the future I'll try this setup too...
The CPU waterblock
The CPU waterblock is a full copper nickel plated HK IV Pro AMD.
To cool down the CPU waterblock I'll build a copper U-shaped bracket.
I'm planning to cut with the dremel a couple of 3mm or 5mm thick pieces; roughly 2.5 cm height and 6 cm length.
Those will be the legs of the bracket; the height must be just enough the keep the base on top of the angled fittings.
I'll order a cut sheet with base 6 cm by 8 cm; the extra 2 cm are to fix the L brackets to keep them together.
I still have to take the exact measurements.
I know it's not the best but I hope it works.
Otherwise I'll have to find a good tutorial on how to solder copper on your balcony...
On top of this base I'll slap the Laird TEC.
Right above there's going to be another HK IV Pro AMD with the acrylic top; I didn't take the full copper to lower the weight.
Here I have some doubts about the mounting that I hope to clarify later.
I'm going to replace M4 screws 5,5cm included in Roman's kit with longer ones; I've ordered 80 and 100 mm lengths.
I've also ordered another OC Kit from Caseking to have the same nice spring locks.
But I don't know the torque they apply to the block. And I can't use the torque screwdriver.
I've found in a Noctua document the torque for the AM4 socket screws should be 0.6 Nm.
According to Laird the torque to apply for their TECs, calculated for the 4 M4 screws and the 52 mm size, should be between 0.4 and 0.8 Nm.
I'm not sure I'll ever get from Roman an answer about these screws torque
After purchasing a bunch of terrible stuff on Amazon I've found https://www.modulor.de/
They have everything I needed more or less.
The insulation for the heat exchanger is critical.
To bring down the temps quickly you have to go down at least to 6-8 degrees.
An this is often below the dew point.
The heat exchanger will start dropping condensation water out like an open pipe.
If you don't make a proper thermal insulation will be just like leave the fridge door open, not wise.
Done properly will get you better performances and lower power consumption.
From Amazon the only good thing I got is this giant roll of glass wool aluminium taped:
I'm going to use it to insulate the hoses.
The aluminium is conductive so I'll have to roll it over with non conductive tape.
From Modulor I got some Forex PVC, Polyurethane foam and Styrofoam sheets.
I'll made a PVC enclosure with the Polyurethane inside to cover the D-RAM blocks.
The Styrofoam is going between the copper plate and the D-RAM blocks to separate the cold and heat side.
I'll make something similar also for the CPU block but I'm not sure yet how big.
In theory I'm planning to keep it well above the dew point, at water temperature level.
I don't want to rubberize the board, even if I have the art eraser ready...
But you never know with such stuff at hand
I have it there and it's tempting...
29/04/2020_2: Dedicated Microdrive, now in the horizontal slot
If you are looking for what they can do I highly recommend to check the videos and read the documentation.
Is quite detailed and well done.
There are a few aspects very interesting for this use case.
The 1090-HV is a High Voltage controller while the 1089-SV is a Standard Voltage.
This means you can feed a variable range on DC input; the DC output to the TEC channel is going to be lower.
It's minus 4 volt on the SV and minus 6 volt on the HV.
This is very important to plan for the appropriate TECs and DC-DC or AC-DC power supply you choose.
The difference is the power draw for the controller to drive the TECs.
You can set limits; this makes pretty easy and safe to pick a limited power supply and avoid errors or destroying the TECs.
Not only you can set them but you can change them via the API and this is amazing.
Via the controller software I can set a low amp limit and a low deltaT to keep the efficiency high.
Or relax the limits and set a higher dT to compensate the temperature rise from a higher load.
There are multiple control modes but the most interesting is the "Peltier, Heat only – Cool only".
You can define a low and a high temperature; the controller will cool down to keep the set object high temperature.
Will also heat to keep the low but you don't want it
After the initial pre-cooling, that you can make very efficient setting a current limit, it's going to take only 5-10 watt with idle load.
It's incredibly efficient and the power draw will raise only with CPU load.
The thermal load from the heated water in the loop will raise the D-RAM blocks temperature.
Then the TEC will kick in to keep the set temperature and the heat will be transferred away from the D-RAM blocks.
This too can be set to maximum efficiency with a current limit or less efficient but more powerful without.
Same for the CPU waterblock; you can set a limit and take away the heat efficiently or loosen it to make it quick.
It only depends on how much power are you willing to sacrifice for the cooling or up to which max CPU temperature you want to go.
It's very flexible and can be completely controlled.
The controllers also needs a lot of cooling; they have a baseplate and over 80 degrees they'll start drop current until eventually stop.
For them I have ready an Aqua Microdrive water block; it's a dual 3,5" HDD block.
It's now planned to go on the 2nd to the left PCI slot.
I'm planning to stick them to one side with a couple of layered copper sheets.
The underneath layer will have holes big enough to seat the nuts, the upper layer a smaller hole.
There are 3 x 3.2 mm mounting holes where to set bolts.
The 2 layers will have both size HDD mounting holes to secure the layers to the Microdrive.
THE POWER FEEDING
29/04/2020_2: Change of converters and dedicated Microdrive
For the 1089-HV I'm using a Victron Energy DC-DC 12/24 10A.
It's pretty small and very cheap; I didn't expect much, the one I did select was not available and then the human malware made it impossible to source.
But it's working very well and only mildly warm at half load. It has a peak output of 20 amps for 10 seconds which is not bad.
It can be regulated by a trimmer screw and so far has been able to deliver power without issues also with the Laird TECs at 8 amps.
I'm changing this for the v2 to an XP Power QSB300; the Victron Orion doesn't fit anymore.
For the 1090-HV I'm planning a Cincon QFB400W.
Quite expensive but has a very decent form factor (116.8 x 61.0 x 12.7 mm), it can deliver 28V at 14.4 amp for 400 Watt.
It's really hard to find a high power DC-DC controller at 30V or 36V; these voltages are mainly used for medical equipment.
And if you find them they are very expensive. While the AC-DC are also either very big or not fanless.
According to the Laird data sheet my TECs should not go over 20V at 8 amps so I think it's going to be ok.
Indeed this is going to limit eventually a future replacement with a higher voltage TEC. But I don't expect it to be very soon
For the cooling I'm planning to use another Microdrive, not the same as where the controllers will be fixed as planned.
A thick copper plate where to screw both power supplies and then drill 3.5" holes to secure it to the block.
The Microdrive will be attached to the vertical PCI slots.
The DC input for these massive DC-DC converters shouldn't be an issue; my EVGA PSU has a single 105A rail.
I'm going to take almost everything it can deliver
If you build your own controller you have to size it to drive your TECs.
But if you buy a controller you have to do the opposite, pick the right TECs for the controller limitations.
And you also have to consider your use case; I have set my ideal use cases at 8 and 18 degrees object and 35 heatsink.
These would be 2 ideal profiles to keep the loop chill and to compensate high load.
For learning I picked two very different models: Laird UT619F14040 at 6A/22V and Marlow 8A/14V.
The Laird has a 80 degrees max operating voltage while the Marlow at 130.
I had read about the Laird being very delicate and indeed they are.
With overheating or a misaligned assembly they melt down immediately.
The Marlow TECs are much more robust, bulldozers. But also bad performers.
They have almost half heat transfer capacity than the Laird.
The voltage difference has a clear impact in the performances.
They both go down quickly for the first 10-15 dT then the Laird takes the lead.
Can't make more specific comparisons since the Laird TECs are gone now...
This is consistent with the statements I've read here that to go down in dT the TEC needs voltage.
And you can see it from the controller; at the same current more the dT widens more voltage is demanded by the TEC.
The Marlow's datasheet is in attach; it's pretty bad. There's no COP graph, absolutely bare.
But you can compare the same graph from the Laird; they really have done an awesome job.
Online you can set all the parameters you need and create your own datasheet.
It's terrific, much better than "contact our engineers".
Now, you'll may wonder why switching to the bigger 52 mm TECs? Just overdoing, looking for a higher dT?
Not really, it's indeed also for the sheer extra power but mainly a matter of efficieny.
The 52mm Laird I picked is of course the top model; UT15-288-F2-5252-TB-RT-W6, max Qc at 220W for 16A/33V.
A whopping 288 thermocouples topping 555W power supply. A single one is more expensive than 2 x 40mm Laird UT9, about 50% more.
At 18c/35c and 7.9 amp it's capable of 148W Qc for 133W power supply, 10 amp 177W Qc for 204W power supply.
That's why I'm not 100% sure the heat exchanger can handle them at 16 amp, 580W dumped on the heat side. Yikes.
With 4 x 40 mm TECs the mounting would have been easier.
For the 52mm the starting current needed due to the higher amount of thermocouples inside it's higher.
But it's quite pointless, at starting voltage the power consumption is neglegible.
The COP is relation between Qc, the heat transfer, and the power supply to achieve the it.
At the same heatsink temperature and dT if the Qc is similar then the power supply will determine the efficiency.
And it's nothing else than the current and voltage needed to drive the TEC.
At 18c/35c one 40 mm Laird at 1 amp needs 4.6V, 4.63W power for 8.66W Qc.
The 52 mm Laird at 2 amp needs 5.6V, 11.33W power for 17.23W Qc.
A very slim advantage over small figures.
Considering all the time and money spent for this setup a 20/30% saving for such a miserable heat transfer is not worth much attention.
Let's compare them against the same 66W Qc at 18c/35c.
One 40 mm Laird needs 2.5A at 9.5V, 24W power supply while the 52mm 3.9A at 9.3V for 36W.
These are interesting numbers; less power consumption and better thermals on the controller, less heat to dump on the heat exchanger.
At full 16A 4 x 40mm needs 232W for 198W Qc, while 2 x 52mm needs 278W for 300W Qc.
Only a 20% power supply increase for a 50% more heat transfer; these are good numbers.
I know the Custom Thermoelectric would have been cheaper but I don't live in the US and I don't want to have anything to do with the German Customs.
They are terrible and a waste of time. I can order from outside EU only with DDP shipments.
THE PARTS LIST
I'm going to do a list soon.
I'm not sure I want to know how much money I've spent on this build
Since it's not easy to understand where I want to put what here's a specific section.
As suggested by
, I'm adding a sketch with the potential assembly design. Thanks!
I had to sacrifice 3 x 3,5" HDDs that I had planned to migrate from the current rig.
I've ordered a couple of 2,5" Barracuda 2TB HDDs to sit on the side together with a Samsung SSD I already have.
The D5 Pump with the Aqua Top doesn't fit on top of the Monsta, it's a 12 cm monster. But it does fir on top just below the ODD BAY.
Just below there will be the EK RES with some good clearance; if it's enough I'll put a 140mm fan below.
I've found an Akasa PCI slot adapter for 2,5" SATA drives on Caseking which is a good fit for the Microdrive.
It's all metal, seems quite robust.
Both the converters and controllers will not fit on one Microdrive...
I will order another one; one will be secured to the PCI brackets for the vertical GPU mounting with the converters.
The other on the mainboard 2nd slot form left side with the controllers.
To make this work I have to change the DC-DC 10A converter; the Orion is too tall.
I'll also change the 14A with the Cincon which is the OEM for XP Power; they are identical but it's 50 euro less expensive.
The res on the ODD bay side doesn't fit; is going to crash with the GPU.
I've found a spot just above the Monsta radiator; I'll have to drill a couple of holes on the PSU cover.
28/04/2020: Going to replace the Marlow TECs with the 52mm Laird TECs
28/04/2020: Change of radiators placement
29/04/2020: Adding Assembly section
29/04/2020_2: new assembly design, new Visio; change of DC-DC converters, adding a Microdrive, change of res position
OPEN POINTS / QUESTIONS
- AM4 mounting screws torque; is there a spec somewhere? Is the Noctua value reliable? What's the torque from Roman's kit?
- DC-DC supply; do I need an OCP/OVP protection from the PSU DC input?
- Do I really need a reservoir for cold loop or can I make it without? Could be an issue with air bubbles?
- Fill port: is a fill tube enough for the cold loop? Can gravity make the liquid go through the check valve? Should I use a ball valve too?