It started with a simple question. What's the most single-threaded performance I can get in a practical 24/7 stable setup? It was clear that the answer was taking a binned + delidded i7 8700K to the highest overclock possible.
But what is the highest overclock possible for 24/7? That depends on cooling, which is when I started looking around this forum for ideas. In particular the builds done by OCAddict and foxrena stood out as excellent examples of what I want to do. I decided to go with the water chiller approach because it doesn't require making a custom CPU block.
The next thing I needed to figure out was how cold I would be able to run the loop. Without really good insulation the limit would be just above the dew point. In this case I'm fortunate that I live in Colorado where the dew point is below freezing for much of the year. The plan is to only insulate the parts of the loop that are not visible. The only visible parts will be a short tubing run to the CPU block and back. The end result is that I'll be getting coolant temps between 1-9 C, which is 20-25 C better than high end watercooling. That should make enough of a difference to be worth it.
At this point it was time to build a prototype. I decided to start with 2x CTE 19911-5P31-15CQ (40mm, 24v, 225W) sandwiched between some EK RAM blocks. I will be running them at 12V from 6-pin PCI-E power connectors. I also decided to use an Arduino to control everything. This has the advantage of being capable of easy software temperature monitoring and control. For the initial tests I just used some electromechanical relays stuck to the back of the control board.
This is the test setup
The CPU is an i7 3930K 4GHz @ 1.19v.
Running prime95 AVX hits about 35C initially, but slowly creeps up because the test setup is unable to maintain the cold loop temperature at full load.
Based on this testing, I decided to use 4x CTE 19911-5M31-28CZ, which is a 24.8V, 62x62mm, 400W TEC.
I will have the TECs sandwiched between a pair of waterblocks similar to this:
http://www.shop.customthermoelectric.com/Water-Block-575-x-575-x-115-WBA-575-115-AL-01.htm
I was able to custom order a pair of the above blocks with G1/4" threading, and holes to mount 2 of them together. The material was also changed from aluminum to copper for compatibility with PC watercooling parts. The blocks ended up being $595 each, which isn't cheap but I will be able to re-use them in future builds for a long time.
If I'm reading the datasheets correctly, this setup should be more than enough to maintain 5C in the cold loop at full load.
Now, on to the hard part: controlling the TECs.
The current test setup is not ideal. I don't want to deal with clicking relays all the time. I plan to use a PWM signal from the Arduino to drive MOSFETs. Generating the PWM signal is simple enough, but I don't have very much experience with designing circuits. As far as I know, the following setup should work:
Datasheet for that MOSFET:
https://www.infineon.com/dgdl/irlb3034pbf.pdf?fileId=5546d462533600a40153566027b22585
The plan is to replicate that circuit 4 times, once for each TEC. I know that using PWM isn't as efficient as voltage control, but I'm not too concerned with that since I'm already running the TECs at around half their max voltage. I don't see any way to make this compact without getting a custom PCB fabricated. Solderless breadboard probably can't handle the ~12.5A current that each TEC is going to pull. I don't mind the cost of getting a custom PCB, but I have no idea how to design one.
So my questions are:
1. Is my choice of TECs/waterblocks enough to cool a i7 8700K at around 1.45-1.475v?
2. Will that PWM/MOSFET circuit work?
3. What should I do to make the PWM/MOSFET setup compact/clean?
4. Is there a better option for the controller? It needs to run off the PC's PSU.
But what is the highest overclock possible for 24/7? That depends on cooling, which is when I started looking around this forum for ideas. In particular the builds done by OCAddict and foxrena stood out as excellent examples of what I want to do. I decided to go with the water chiller approach because it doesn't require making a custom CPU block.
The next thing I needed to figure out was how cold I would be able to run the loop. Without really good insulation the limit would be just above the dew point. In this case I'm fortunate that I live in Colorado where the dew point is below freezing for much of the year. The plan is to only insulate the parts of the loop that are not visible. The only visible parts will be a short tubing run to the CPU block and back. The end result is that I'll be getting coolant temps between 1-9 C, which is 20-25 C better than high end watercooling. That should make enough of a difference to be worth it.
At this point it was time to build a prototype. I decided to start with 2x CTE 19911-5P31-15CQ (40mm, 24v, 225W) sandwiched between some EK RAM blocks. I will be running them at 12V from 6-pin PCI-E power connectors. I also decided to use an Arduino to control everything. This has the advantage of being capable of easy software temperature monitoring and control. For the initial tests I just used some electromechanical relays stuck to the back of the control board.
This is the test setup
The CPU is an i7 3930K 4GHz @ 1.19v.
Running prime95 AVX hits about 35C initially, but slowly creeps up because the test setup is unable to maintain the cold loop temperature at full load.
Based on this testing, I decided to use 4x CTE 19911-5M31-28CZ, which is a 24.8V, 62x62mm, 400W TEC.
I will have the TECs sandwiched between a pair of waterblocks similar to this:
http://www.shop.customthermoelectric.com/Water-Block-575-x-575-x-115-WBA-575-115-AL-01.htm
I was able to custom order a pair of the above blocks with G1/4" threading, and holes to mount 2 of them together. The material was also changed from aluminum to copper for compatibility with PC watercooling parts. The blocks ended up being $595 each, which isn't cheap but I will be able to re-use them in future builds for a long time.
If I'm reading the datasheets correctly, this setup should be more than enough to maintain 5C in the cold loop at full load.
Now, on to the hard part: controlling the TECs.
The current test setup is not ideal. I don't want to deal with clicking relays all the time. I plan to use a PWM signal from the Arduino to drive MOSFETs. Generating the PWM signal is simple enough, but I don't have very much experience with designing circuits. As far as I know, the following setup should work:
Datasheet for that MOSFET:
https://www.infineon.com/dgdl/irlb3034pbf.pdf?fileId=5546d462533600a40153566027b22585
The plan is to replicate that circuit 4 times, once for each TEC. I know that using PWM isn't as efficient as voltage control, but I'm not too concerned with that since I'm already running the TECs at around half their max voltage. I don't see any way to make this compact without getting a custom PCB fabricated. Solderless breadboard probably can't handle the ~12.5A current that each TEC is going to pull. I don't mind the cost of getting a custom PCB, but I have no idea how to design one.
So my questions are:
1. Is my choice of TECs/waterblocks enough to cool a i7 8700K at around 1.45-1.475v?
2. Will that PWM/MOSFET circuit work?
3. What should I do to make the PWM/MOSFET setup compact/clean?
4. Is there a better option for the controller? It needs to run off the PC's PSU.