Zds Project Log

Yesterday I bought some aluminium-capable blades for my jigsaw, a drill press and router. Unluckily I didn’t have time to pick up my aluminium sheets and rest of the week goes into long weekend snowboarding in Lapland and preparing for it.

But on the bright side, I’ve made some progress on the case shape front:

The rightmost is my vision per today, middle one from Sunday and leftmost from Saturday. Definitely getting better.

The clay model did it’s job - it helped me to see which parts worked and which did not, resulting in a sleeker model. Huzzah.

But I better get the shape fixed soon, as I do not want to cut 130€ worth aluminium into something that does not satisfy me :-].

Starting again with foamcore:

Remember kids to sand the foamcore sheet before gluing.

Here you can see the size difference between the first iteration and the current:

I “maybe” overdid “a bit” the polyurethane foam part:

But there’s few problems that cannot be solved with sharp enough blade. Now looks a lot better:

Add some artist’s clay into the mix:

I saved some money and used the kind of clay used mostly to do ceramics and not the one used by sculptors. This means it’s more sticky and coloring, so I didn’t dare to touch the camera while sculpting.

Anyway, here’s what I ended up with:

As I’m sick third week on a row, no chance to get anything physical done. Instead I have done some math and drawn some pretty pictures.

As you might recall, the plan this far was to have water flow in thin layer between TEC hot sides and backplate, from up to down, and distribute heat to backplate while flowing. Well, I started to calculate the thermal conductivity performance and heat distribution along the heatsink and quickly decided to change the plans.

Here is the new plan:

The new thing is that the heat exchanging part (where cool and warm loops flow over the TECs) is now narrower and I have added copper pipes (the red lines) that will be embedded inside the back heatsink.

Cross section showing the two water loops and TECs (from top to bottom: cool loop, TEC, warm loop):

Here is how the copper pipe will be mounted. The idea is to take 12mm OD, 10mm ID copper pipe and squeeze it to 10×10mm groove in the heatsink:

The bottom grey part is the heatsink, top grey part is the backplate. Thus copper pipe is sandwiched between backplate and heatsink. If my calculations are correct, the pipe should fit just as well as in the picture; it’s in the exact scale. By forcing the copper pipe to deform I hope to get better contact between the heatsink and the pipe.

Here are gross estimates about temperature gradient over heatsink in the old and the new model:

Each step represents 2.5°C difference in temperature when the total heat load transferred is 250W. As the heatsink will be constructed from vertical aluminium pieces, I have assumed thermal conductivity of 150W/m/°K to the horizontal direction and 200W/m/°K to the vertical direction.

This is basicly optimization between flow resistance and maximum temperature gradient: more pipes mean less flow but better thermal transfer..

What I hadn’t realize before was that while flow resistance is roughly linear in relation to the lenght of the pipe, thermal transfer increases squarely with the pipe length. If I have reasoned properly, this is due to the fact the when you add more piping, you both decrease the average distance from each point in heatsink to the closest point of hot water _and_ increase the amount of conducting surface. The latter means that there is less thermal energy per cross section area of the heatsink to transfer.

And the last image shows how the heatsink will be constructed. The top of the image is the outer side and the bottom is the inner side. Note that most of the inner surface will be flat; this depicts the part where there is also internal heatsink:

The outer fins are optimized for passive heat transfer, thus they are thick and sparse: 2mm per fin and 10.5mm between them. Inner fins, where present, are supposed to be cooled via forced airflow, and there fins are 0.5mm thick and have 4.5mm worth space between them.

Proceeded with the second scale model. I let the images do the speaking:

The backplate and its screw mounts are sturdy enough that I consider cutting it out and using in the final assembly. This way I also save the pain of trying to find fitting metric screws to replace the original ones:

This shows why this is perfect PSU for TEC use. 70% of the rated maximum power is available in a single 12V line:

It seems that I do not want to remove the original heatsinks. They are made of single piece (each one) and screwed with some dozen, half of which are behind soldered components. Here are the only four screws I _could_ remove easily:

This component it a bit of mystery to me. It was screwed to main heatsink separately. OTOH, I have not been able to find anything resembling thermal diode:

The scalp! As I will not use any standard connectors, I cut them all away:

I also cut away lot of the power cables to reduce the clutter:

What I left is 8 +5V cables (red ones), 7 +12V cables (all there was..) (yellow), few 3.3V cables (orange) and 12 ground cables (black). The thinner ones that are together separately are standby power cables:

There was some 16 ground cables in total, but as the TECs will draw power from either 5V or 12V line at any single moment, never from the both, it’s enough to have one 5V, one 12V and one ground line per TEC module. Thus the six TECs eat up 6+6+6 wires in total, leaving all the 3.3V wires, two 5V wires, one 12V wire and six ground wires for whatever other needs I will find.

I also played with idea about replacing half of the remaining ground wires with one very thick one and using it for all the TECs.. lets see.

As you saw in the blueprints, I plan to rip the standard fans off of the PSUs and make them leaner, while offering straighter and less restrictive airflow path and external heatsinks and fans.

The first victim was the Antec 480W TruBlue PSU, which will power the TECs and some of the lights and microcontrollers.

It’s a dual-fan model, 92+80mm:

To make room for the two fans, the main PCB is split into two parts, the smaller one raising almost to meet the opposite cover:

This is bad, since I want to squeeze the PSU (sans cooling) into 150×140x60mm box.

Fortunately it’s fairly straightforward to detach it and move it to the space once occupied by the rear 80mm fan. This just means I have to lenghten the five wires connecting the two PCBs:

This PSU convinced me of Antec quality. It feels robust and solid, and has more features I could have understood to ask for. For example, it had an external Molex connector covered with plastic cap:

I remains a mystery to me what this tiny PCB between main +12V, +5V and 0V lines and external molex connector does, but it looks like Antec has tried to actually make sure the external one and internal ones will not mess each other:

And naturally there are some leds. In fact, these blue leds are the only blue element in the PSU, PSU that comes with name “TruBlue”.. I have my doubts about how that brown goo used to glue them affects the effectiveness of the heatsinks:

And here they are cut off, the external connector and its PCB and the leds along with the legacy 6-pin 3.3V connector that never actually got popular:

This PSU had more and longer connectors than any PSU I have had this far. Naturally it also had two fan only molex headers:

I didn’t cut them, since if the Antec fan controller is of any use, I might still have some use for them.

The two Dynaeon Industries fans were detached and saved for future use:

The rear 80mm fan sports also RPM sensor wire that was made available externally:

As I have the overall shape and component placement now fixed, it’s time to check that my vision about the look can be implemented.

As I have 0.8mm thick aluminium sheet and 3mm thick acrylic sheet handy, I chose to make the second scale model to 1:2.5 scale. This way the effective thickness of the aluminium matches the 2mm I have planned for the final product and acrylic thickness simulates closely enough the planned 8-10mm full scale thickness.

Especially in bending to concave shapes scale is fairly important. As with the first version it proved out to be fairly difficult, I try now to tackle that risk early on.

Enough talk, here are the images:

As it’s already quite late, I decided to leave rest of the hammering, filing and dremeling to tomorrow. The finish of the edges leaves lot to be desired, but perfection of the finish is not the point of this model. The point is to verify that the materials and shapes I have planned work together so that I can produce the same shape with same materials in full scale.

I was sick this week, so not much work or modding got done. However, I refined the new plans to the point that I can not start working with full-size model. As you can see, I went for the bottom-heavy design:

As this a bit confusing picture shows, the top decoration element is gone. What is harder to see is that I scaled the case up in size some 8% per direction, resulting in 25% more colume to fill with parts. This made it easier to fit parts and still cool them properly and retain desired case size. Now the total dimensions are 800×470x330mm.

The light magenta part between drives and backplate heatsink marks air guide for back fans.

Basic idea in cooling is that all the heat is transferred to backplate heatsink and then when necessary, fans will spin to provide additional cooling. There is two water loops, one that flows in direct thermal contact with backplate and another that’s cooled down with peltier elements.

Here you can see the two waterloops:

Dark blue marks cool water loop, dark red the warm one. Light cyan marks the cool loop reservoir/TEC block and light magenta the warm ones.

At the top, in the “horns”, are the two reservoirs. Water flows through the sides of the chassis to the bottom where the two pumps reside. Peltier elements are located between the side flow areas. When flowing downwards, the cool loop water get cooled down by peltiers and the warm water loop cools hot sides of the peltiers and further transfers the heat to the backplate.

I reasoning behind this is that as the backplate is huge and made of aluminium, which is not the best thermal conductor there is, there will be substantial temperature differences between different parts of the backplate, unless I design cooling properly. The traditional approach has been to thicken the backplate (the part that’s not fins, but mounts them), but it would need to be very thick to effeciently transfer heat through the 900mm cross length of the heatsink.

So I ended up having warm water loop flow in thin and wide layer over the backplate. This should reduce the temperature gradient significantly and thus allow better cooling.

In addition to the passive convection cooling of the backplate heatsink, there is three portions of heatsinks that can be cooled actively.

First of all, the two back (up to 110CFM) fans protrude through the backplate and when spinning, move air over both external (mostly passive) and internal heatsink. The external heatsink is massive and with thick and sparse fins, to support passive cooling.

The internal heatsink is above back fans and is attached partly to backplate and partly to HDDs. It has thinner and more tightly packed fins as it’s mainly cooled with airflow.

The third heatsink is at the bottom front and gets heat from copper tubes transporting water of the warm loop. The three front fans (slow and silent) pull air through the heatsink and then over the mainboard and display adapter.

The reason why I push _warm_ air over the mainboard is that, first of all, it doesn’t really matter, as the parts producing most heat are anyway watercooled and second, I want to keep condensation temperature as low as possible. Naturally I will utilize some thermal insulation when going to sub-zero temps, but it does not hurt to have warm air heating up the outside of the insulation layer.

The two PSUs (one for the computer parts, another for the peltiers) will be bolted to the backplate, so they should stay rather cool even without any active airflow. In case they do not, the two back fans pull air through them to be evacuated from the top of the case.

Here is image illustrating the “show” parts, ie. parts that will be constructed to look good and be visible:

The rest, naturally, will be hidden.

And finally, here is image showing the base structural elements:

These plates will mostly be 2mm thick aluminium sheet attached to each other with L-shaped brackets. They should provide robust backbone to keep the pieces of the heavy case together.

I am also planning to have double middle-plates and have hinge at the back between them. This way I could basicly split the case to two halfs, should it be necessary. As case is mostly symmetric anyway and I will have to construct the heatsink from two pieces, this halving should not weaken the case too much. And it would enable me to do maintenance to the internal parts without tearing the whole case apart.

Now I need to wait for the replies from the sellers of ready heatsinks to know whether I want to use ready one or make one myself. And get my hands on 0.5mm and 2mm thick aluminium sheets.

Forgot to mention: The next thing after I can make up my mind with the design is to make full-scale model from chipboard, foamcore sheet, dummy PC parts and cardboard. Then I shall begin replacing those model pieces one by one with real ones. This way I should have some concrete perspective on how things will look in the end most of the time.

Before I can go on, I should decide between bottom-heavy and top-heavy design.

Bottom heavy:

Top heavy:

Then next thing to decide is whether I want to use ready heatsink or construct one on my own from thick aluminium sheet.

Using ready one naturally means less work and higher price, but making your own would give more versality.

To give you some idea, here is side view with ready heatsink and one with custom construction:

Orange dots mark where heat will be transferred to the heatsink from water loop(s).

As I do not have possibility to weld the pieces, custom construction would have worse conductivity between the fins, but to compensate that, external (passive) and internal (active) heatsinks could be clamped together more tightly.

If I do it on my own, here is a top view of the design:

The idea is to have copper pipe/bar run through hole in each sheet and have wider spacing and fins on the passive side and thinner fins and narrower spacing in the active side. To have the pieces tightly against each other I will lap them and then
tighten using thread bar, similar to Zalman flower heatsinks.

If I go for top-heavy, here is cross-section from the height of the bottom edge of the MB:

The blue thing in the blueprints is Laing Delphi pump. My plan is to have 1-2 pumps and reservoir(s) in the top compartment.

Here’s a bit cleaner scetch looking from the front:

And having learned from my mistakes I decided to start with scale model. CAD programs that are available freely failed to impress me, and this way I can get even better understanding about how things will or will not fit - even to the point I can test that all the plates can be bent from stiff stuff before I do 1:1 pieces from expensive materials.

The scale is 1:5.

Constructing it:

And finished:

I still like the design. Next I will model tubing, reservoir and the pump and then the front plates. The last part will be the most challenging, I expect few iterations to get it both implementable from thick acrylic and good looking..

EDIT: Forgot to mention that the green foam is used to simulate the heatsink. I just didn’t feel like making 0.4mm thick fin simulations and the thickness is about right (10mm, ie. 50mm in full scale)..

Even thou it’s been silent on actual modding front, I have been thinking about this project daily.

Now that I am more familiar with TECs and properties of acrylic, I began to think again the overall structure of the case. Earlier I had designed the case around forced air cooling but now I began to think how it would work out if build primarily around free air cooling.

Here are some scetches:

First of all, this is the first design where I both like the case top looks and where I was able to fit all the components without big compromises.

The main differences to earlier design:

1. Instead of huge radiator, there is huge passive heatsink that covers the whole back of the case
2. Instead of blowing cool air over MB and then through radiator, I am blowing half-warm air over the MB
3. Power supplies are stripped from fans and attached to the big passive heatsink and when forced air is needed, cooled with same fans as the heatsink

My goal is that when I have set target temperature to 15 degrees or higher, I could run the system (at least almost) passively. Then when I want maximum overclocks, I could set target temperature near zero and utilize fans to remove additional heat produced by the TECs and overclocked parts.

In the first mode I have 150-230W worth heat to be removed from the system and in the second mode all the way to 600W. Further I want the heatsinks to stay under 60ºC in normal mode, so I need heatsink capable of below 0.2K/W cooling at 200W thermal load. This should be perfectly achievable given that I have reserved about 0.24m² worth space (600×400mm) to up to 50mm thick passive heatsink.

If I want to go sub-zero, I will need either double the amount of TECs or feed them with some real power. As the latter case is thermally the more challenging, I have planned to case to survive it. The way to handle that load is to rise the temperature of the main heatsink and to have fans blow across internal (smaller) heatsinks. The internal heatsinks are way smaller than the main one, but OTOH, with forced cooling efficiency of those heatsinks rises five-fold.

Next task is to find fitting heatsink with less than insane price. This far closest possible one is priced at 200€ in amounts I need, but what wouldn’t I pay to get the perfect case..

Here is the currently best candidate for heatsink:

My plan would need two one meter pieces of that, costing just below 200€.