Zds Project Log

I have never really been satisfied with my first try at MB tray. Then I recently ran into this:

Sure I have known a lot about Zalman TNN-500, but this detail had escaped me: backside of the MB is thermally connected to the side of the case.

I began to think what should I change to have thermally conductive MB tray, and in the end I began to like the idea a lot more than the previous one. After all, no one sees the MB tray so it does not has to be transparent or fancy.

I further concluded that I might also get rid of some unnecessary resistance in my water loop if I could attach some of the less heating components to the MB tray plate and then watercool it.

Like this:

This is pictured from the MB IO-panel side. My plan uses this aluminium profile meant for watercooling (click the flags in upper right corner to get some human-understandable language):

In the picture above is a design with three pieces of that liquid cooling element and then pieces of acrylic between them to act as rails. Each hole accepts G1/4″ fittings, but I might not want to connect all of them.

Since the regulators and DRAM modules do not need any extreme cooling, I plan to attach copper pieces to them from another end and to MB tray from the another. This means the heat has to travel some distance, but as said, they do not produce so much heat so if the MB tray is cooled below 20°C, there should be no problems for them to stay cool enough.

The advantage of this is that I can cool the regulators from both sides and I need less tight bends and fittings and thus can run with less unnecessary restrictions in the loop.

Maybe :-). It remains to be seen if my plan works.

This beast arrived today:

It’s second-hand Antec True Blue 480W and it will power my TECs. I have some time been looking for another PSU, either to replace my current Nexus NX-4090 as main PSU and move Nexus to power TECs, or to complement the Nexus by powering the TECs. Now that I found exactly matching PSU from local online auctioning system, this decision is made.

Most of the <500W PSUs (and many 550-580W PSUs) are not really that powerful, when you look at the specs. The capability of the 12V lines are nowadays almost all that matters but still it was hard to find a PSU that would be able to give 21A or more juice from the 12V line.

Another obstacle was that for PC usage dual, triple, or even quad lines are better than single powerful line, but for operating TECs I wanted to have maximum flexibility while exact voltage levels are of secondary importance.

This Antec fulfills this need: it can give 28A from 12V line, and this is not a peak value. It means I can cool my primary water look 3 degrees below freezing poing with this and 6 Kryotherm DRIFT-0.8 TECs. Not bad at all.

It’s heavy as hell, which is a good sign. In PSUs heavier is usually better.

I found very handy TEC information and simulation tool from Kryotherm website. Now I am beginning to discover how peculiar devices TECs really are.

After finding out how much CoP (coefficient of performance, amount of heat pumped divided by amount of power supplied) is affected by voltage and temperature gradient I decided to go for the most powerful (in terms of heat pumping capability) Kryotherm model, DRIFT-0.8.

Here its specifications:
29,90 €
Qmax: 172W
Umax: 24,6V
Imax: 11,3A
DTmax: 69°C
Size: 40 x 40 x 3,2mm
Weight: 20g

Using the Kryotherm simulator I searched for the best combination in terms of cost, cooling capability and power usage.

Below are two graphs showing how number of DRIFT-0.8 modules and radiator/waterloop efficiency affects the TEC power draw (in Watts). Temperature of the cold side of the elements was kept at constant 16°C. This temperature was chosen since it’s my best guess for lowest condensation-safe temperature in finnish climate. Ambient temperature was assumed to be 22°C.

In this simulation TECs were connected parallel. For each data point I adjusted the TEC voltage at 0.1K incements until I found the lowest voltage that still kept cold side of the TEC at 16 degrees or cooler. Power draw of TECs was then written down, and is shown in the graph below.

In the Y axis is power consumed by the TECs (W). In X axis is the number of DRIFT-0.8 elements used. Each line represents certain K/W (or C/W, if you prefer) value at 0.1 intervals.

Here the thermal load is assumed to be 200W (estimated maximum power draw for high power AMD system with Radeon X800XT):

Here the thermal load is assumed to be 120W (estimated maximum power draw for high power AMD system GPU stress):

Looking at these, the optimal point for my purposes seems to be between 4 and 6 modules (inclusive). That would be 120 to 180 euros.. Cheaper than phase-changing system, but still a lot of money.

The reason why I added four different K/W levels is that I want the machine to run as silent as possible. This means that under normal load (no heavy gaming) I could run the fans near the minimum (estimated to be 0.3-0.4 K/W) and under heavy load (gaming) the fans would be allowed to run faster (0.1-0.2 K/W).

I know 0.1K/W is fairly low even for watercooled system, but remember that I am using a car radiator and six 120mm fans.

I have to say I am pretty amazed to what you can do with peltiers. They cost a pretty penny, but 60W extra power to keep your CPU and GPU cooler at constant 16 degrees when under maximum load isn’t bad at all. And under “just” full CPU load it drops to 26W.

What was even more surprising was that if I accept CPU temperatures a bit over 30 degrees Celsius, TECs will draw just one or two Watts of power. This means I can propably have two separate loops. And 30 degrees is not bad at all, if you just want to surf web at stock speed.

And for the record: if the 16 degrees sounds to be a bit on the warm side, those TECs can do a lot more, if asked to. With six modules and 12V of power (the highest you can get from regular PC power supply) you could keep the coolant liquid 6 degrees below freezing point, at full 200W load. This of course comes with the cost, in form of 380W of power draw for the peltiers alone.

I have been planning and re-planning waterloops, tubing and TEC (Thermoelectric Cooling module) placing quite a lot lately. My plan is to have secondary PC PSU to run TECs and have PIC microcontroller running on its standby power and starting it up and shutting it down when needed - and of course controlling TECs.

While searching for information on what’s a best way to set up TECs, I ran into an article in Electronics Cooling about how to select TEC configuration that gives best results per watt.

Some of the charts were eye-opening, like this:

As you can see, contrary to popular belief, it is possible to get much better than 0.4-0.7 for CoP (coefficient of performance; amount of heat (W) pumped per power (W) fed to TEC). If you run your TECs at low enough voltage and design your system to have low ΔT, you can far beyond CoP of 1, even close to compressor-cooler territory (CoP of 2-9).

In laymans terms: if you throw in more TECs and run them at lower voltages, you can pump a lot more heat with same amount of electricity.

I knew it was not recommended to run TECs at more than 80% of their nominal voltage, but if the graph above holds true to Kryotherm modules, it means I might want to run my TECs as low as 20-25% of the nominal voltage.

This means that with bigger up-front cost (buying more TEC modules) you can reduce your systems electricity consumption, or get better cooling with consumption remaining equal.

I am now considering buying 4-6 172W modules (DRIFT-0,8) instead of 4-6 80W ones.. Costs a pretty penny, but hey, what wouldn’t I do to have a case that re-defines overkill?

This goes out to prove that sometimes you do not need lot of time or budget to silence something:

I got bored to Titan fan howling on the home server heatsink and swapped in thermally controlled Acoustifan. In addition to swapping the fan I also added four tiny pieces of 1mm rubber sheet and attached the fan with cloth rubberband. Very cheap, very easy, very effective. And my ears thank me.