Post Rad Chiller Concept

[Uncle Jimbo]

No, you can create a liquid delta
Its done but normally needs much larger systems, I've gotten cold side down a few degrees actually, seems to help temps when the rad can take the heat

If the goal is to achieve and maintain a low temp at the heat source (CPU, GPU, whatever you are cooling) using circulating water to move the heat, then in a steady state model, the water exiting the heat source will be at about the target temp. If that temp is close to ambient, a rad won't do much for you - all it can do is pull out heat inserted after the load (from the pump etc.) . The rad will also keep the temp from going below ambient - it transfers heat both ways. You probably want something like a car thermostat ahead of the rad - route water around it unless the temp is above ambient, to keep the rad from working against you.

That's where a post-rad chiller comes in. In order to maintain ambient or even below ambient at the heat source, you obviously need to have the input cooling water below the desired exit temp. If something close to ambient is desired, then even a degree or so is useful.

Martin's original calculation of 395 watt/seconds for 1.5GPM was actually mathematically correct, and as he later stated, the calculation also needs to take into account heat load. But the calculation went south from there...

If we are looking for max efficiency and are using any of the combinations of TECs I have discussed, we will see about 140W transfer. OC'd CPUs can put out more than that, so maybe a higher design center is needed. We can either sacrifice some efficiency, or go with more TECs. But if your average load is less than the TEC transfer, the water will gradually get colder. You can also maintain temp above a lower limit by running water through the rad - that is more simple than varying the TEC power. Another way to control cooling below the desired level is to use the hot side. If you are using air cooling, slow the fans, if you are using WC, slow the pump or shunt the cooling water. Either way, the TEC hot side gets hotter, the resistance rises, the differential rises, and less heat is transfered out. You can achieve very tight temp control with any of those methods.

It is also fairly easy to control TEC input power directly, for low current loads. Essentially running PCM via a big power MOSFET into a low pass filter so the TEC sees a relatively constant voltage.

If I were designing to cool the big heat loads in a state of the art OC'd system (CPU and GPU), I would expect an average heat load of around 250W and upwards of 400W under heavy load. To match cooling without burning power all the time, we could let the power run up if the system starts to get hot, which means using the PCM control.

Let's say we are using the 4 03178 TECs I described earlier. We run them off the 12V line, but with PCM control around output temp at the load, or at the chiller. At idle, PCM sets the TEC voltage at 5 or 6 volts, you pull 140W out, power use is low (65W or so in addition to the heat load), and you are only burning what you need to maintain temp. If the load dumps in more heat, the TEC ramps up as the PCM duty cycle increases, and with the 600W or so transfer available at 12V, I'm sure you will be able to maintain temp if you can supply the current. But 60A or so is serious power, and the PCM system would need to be big too. You'll be burning some serious power at that point though - around 1000W in - so maybe you clamp the upper end at the 50% point, which I think would be around 300W in and 300W transfer (something like 10V, 30A), and let the rad pick up the slack. Your component temp would go up some, but would still be lower than the rad alone.