[foxrena]

I start this project because I really want to go sub-zero on the cold plate. The 4 TEC chiller I used before was rough (no water channel/pins), requires two loops and difficult to insulate. So I decided to try the direct contact TEC block route. I was about to purchase a dual TEC block from Ultrasonic2 (I like it a lot), but sadly learnt that he doesn't make it anymore.
After some testing I figured out a way to make deep water channels with my drill press ( with x-y table) and I made the block!

I figured I should create a thread here to share my lessens and mistakes.

It will be divided into five parts:

1. Block design and making
2. TEC controller and modding a Lab grade variable DC power supply
3. Assembly and initial testing
4. Insulation and installation over 3930K
5. Real world test and 24/7 run

This is just my fist trial in making a TEC block. Any input and comments will be appreciated. I will post A LOT of pictures, I promise

 

[foxrena]

Part 1.1. Block design

This block will hold two 50x50 mm TECs. The copper plate I use is 1/2'' thick for both water block and cold plate. The hot side of TEC is water cooled and cold side is clamped with cold plate and then mount onto CPU.
For the hot side water block I go with the simple design of straight water channels (1/16'' wide, 5/16'' deep), hoping to maximize the flow rate. On both ends there is a small chamber to help distribute the flow evenly inside channels.

TEC Hot side water block, made by 1/2 inch copper alloy 110




Cold plate:




Cover plate, made by 3/8 inch acrylic:

 

[Samurai Batgirl]

Looks very interesting so far
The idea and the block do anyway

 

[Liighthead]

looking good from here

subbing in to see how it goes

 

[eskamobob1]

definately sub'd

 

[foxrena]

Part 1.2 Cover


It's acrylic, so it's easy to mill using my drill press. It's hard to get the right depth for the o-ring groove though. If it's too deep, I need to grind the surface to make the depth correct. Fortunately for this one it is just right. I like the look of counter sinks.

 

[george_orm]

So so sweet

 

[Mindchi|l]

Quote:

I was about to purchase a dual TEC block from Ultrasonic2 (I like it a lot), but sadly learnt that he doesn't make it anymore.

There is a reason he doesn't make them any more. They never really worked well for large heat output processors.

What tecs are you going to run and at what voltages if you don't mind me asking.

 

[foxrena]

Quote:

Originally Posted by Mindchi|l

There is a reason he doesn't make them any more. They never really worked well for large heat output processors.
What tecs are you going to run and at what voltages if you don't mind me asking.

Yeah, they have limitations. From what I read, on quad-cores (i7 900 series) they are very nice though. I guess it's also not profitable for US2 to make it anymore.

I am using two CTE 50mm TEC (Qmax of 331W, and Vmax of 31.5V). I have a 30V 30A DC power supply and the voltage will be controlled by my controller. It will allow a maximum of 25V while benching. On idle it should use very little power in constant temperature mode. Good for 24/7

 

[Mindchi|l]

Quote:

Originally Posted by foxrena

Yeah, they have limitations. From what I read, on quad-cores (i7 900 series) they are very nice though. I guess it's also not profitable for US2 to make it anymore.
I am using two CTE 50mm TEC (Qmax of 331W, and Vmax of 31.5V). I have a 30V 30A DC power supply and the voltage will be controlled by my controller. It will allow a maximum of 25V while benching. On idle it should use very little power in constant temperature mode. Good for 24/7

Only person I ever saw get any kind of semi descent results on a I7 900 series chip was elloquin and he had his voltage so low the chip was only putting out about 160 watts.

 

[Elloquin]

Quote:

Originally Posted by Mindchi|l

Only person I ever saw get any kind of semi descent results on a I7 900 series chip was elloquin and he had his voltage so low the chip was only putting out about 160 watts.

LOL

 

[foxrena]

Quote:

Originally Posted by Mindchi|l

Only person I ever saw get any kind of semi descent results on a I7 900 series chip was elloquin and he had his voltage so low the chip was only putting out about 160 watts.

Quote:

Originally Posted by Elloquin

LOL

Yes, I refered to Elloquin's results.

In those tests the cpu did have low voltages except for overclock >4.5GHz.
I guess we have to define the satisfactory standards first. I will consider it decent if the block holds cold plate at a dT of -20C with respect to water temperature, under full load.
Unfortunately Elloquin only posted core-temps (please correct me if I'm wrong), and I have to estimate the cold plate temp.

I choose the dual tec block scheme because it is relatively easy for me to make. I would go with 4 if I had a mill (like the "safe" you made

). Maybe that will be my next project.
BTW, are you planning to sell any large direct block except for the single TEC one?

 

[chmodlabs]

That's some really clean engineering.
Great work.
- chmodlabs

 

[Mindchi|l]

Quote:

Originally Posted by foxrena

Yes, I refered to Elloquin's results.

In those tests the cpu did have low voltages except for overclock >4.5GHz.
I guess we have to define the satisfactory standards first. I will consider it decent if the block holds cold plate at a dT of -20C with respect to water temperature, under full load.
Unfortunately Elloquin only posted core-temps (please correct me if I'm wrong), and I have to estimate the cold plate temp.
I choose the dual tec block scheme because it is relatively easy for me to make. I would go with 4 if I had a mill (like the "safe" you made

). Maybe that will be my next project.
BTW, are you planning to sell any large direct block except for the single TEC one?

I've never been happy with any direct die block I've designed for a cpu because heat only seems to move in a ~ 40-50mm circumference around the heat source (cores). I'm sure this is a main reason waterblock designers engineer blocks that concentrate on this area. As you move away from that circumference the block moves less and less heat ouside that area.

I've had dual tec blocks that read -10c on the ends and 20c in the middle. I've found a dual tec chiller will move more heat because heat transfer is dristributed equally across the whole tec.

This is the main reason I just don't make dual tec blocks. I think the only way to actually get a dual tec block to work properly is using heat pipes to move the heat away from the cores.

 

[foxrena]

Quote:

Originally Posted by Mindchi|l

I've never been happy with any direct die block I've designed for a cpu because heat only seems to move in a ~ 40-50mm circumference around the heat source (cores). I'm sure this is a main reason waterblock designers engineer blocks that concentrate on this area. As you move away from that circumference the block moves less and less heat ouside that area.
I've had dual tec blocks that read -10c on the ends and 20c in the middle. I've found a dual tec chiller will move more heat because heat transfer is dristributed equally across the whole tec.
This is the main reason I just don't make dual tec blocks. I think the only way to actually get a dual tec block to work properly is using heat pipes to move the heat away from the cores.

I totally understand your concern. Actually I have done some rough simulations to calculate the optimal cold plate thickness. You readings of -10c at end and 20c in the middle is close to my simulation result.

I have a question, What size (thickness) of cold plate did you use in that test?

Here is the simulated temperature distribution across the center line on the top of 10cm by 5cm cold plates (400W cpu power). Note that all temperature is relative, so only take the dT reading.

4mm thickness:


10mm thickness:


Graph for 10mm:


As the cold plate gets thicker, the temperature at the cold side of TEC gets more uniform. But thicker means larger value of C/W value from CPU to TEC, too. The optimal thickness is between between 8-12mm. Certainly this aspect affects the performance of the block. But overall it's not too bad because in the center the hotter area compensates the lost heat transferring across TECs at the two ends, as Qmax is a semi-linear function of temperature.

Yeah, heat pipes will help. I bet you have though about a sandwiched version of direct die block, which the water block is in the middle and on the top and bottom there are two TECs with cold plates linked by heatpipes. I like chillers, with their ever expanding Qmax. Adding more TECs easily fixes the efficiency problem. My only concern is the complexity of two loops and insulation issues.

 

[foxrena]

Part 1.3. The water block

The water channels are cut with a 5/32 inch end mill bit on my drill press. My drill press is terrible at milling copper, but I found an interesting way to cut channels which is cut download (plunge cutting) and have overlap on each cut.


In this way I can cut 8mm deep channels easily. It is just very time consuming to do this manually. It took me 8 hours to cut this water block. The bit is still sharp after I finished the block though. No breaking bit

Enjoy the pictures:

 

[Rognin]

I love copper!!!!

 

[foxrena]

Part 1.4. Cold plate

I though the cold plate would be easy, using the same method I cut the channels of water block. But it turns out it's much more difficult to do plunge cutting on a straight edge (not in a channel). I almost broke all my end mills

In the end I did it with the normal milling style. It took me much longer to finish the cold plate. I desperately hope I had a cnc mill.

They look quite good after some polishing. I don't have fly cutter, so I know the surface is not flat for the TECs. I will live with that first.

 

[rafety58]

This looks amazing man, can't wait to see more

 

[Coolwaters]

cant you hand sand that down to make it flat? i seen a lot of people lap CPUs pretty deep.