Zitat von »Diago«

Zitat von »mattlach«

Zitat von »mattlach«

Diago,

As an aside, I just realized what may have been a contributing factor in our disagreement earlier.

I don't think we were talking about the same Delta T.

In retrospect I think you were calculating the temperature of the coolant before and after the block. Is that correct?

The Delta T I was talking about was the temperature difference between the coolant temperature, and the actual silicon core temperature as measured on die.

This is much too complex of a relationship (IMHO) to accurately calculate in theory, as it depends on too many unknowns. Some things we know (coefficient of heat conductance of copper, for instance) but then there are other aspects as well. How flat is the block? How flat is the top of the heatspreader of the chip? What thermal paste is being used? How thick is it in its final mount? What does the chip assembly look like underneath the heat spreader? what is the exact heat dissipation, etc. etc. etc.

I tend to think of water cooling loops from the perspective of two distinct Delta T's.


Ambient Air <---- DT1 ----> Coolant Temp <---- DT2 ----> Die temp.

DT1 depends on a combination of heat load, flow rate, radiator capacity/efficiency and fan speed.

DT2 depends almost exclusively on heat load, block efficiency and flow rate. Since I know I was running the same silicon at the same settings and voltages, I knew that the heat load was the same as before, and the block was the same as before, thus the only variable here is the flow rate, and I feel fairly confident that I can tell the difference in flow rate up based on this delta T value.

This is also why I am a proponent of high flow rates. If I can minimize DT2, I can allow my coolant temp to be higher, and thus lower my fan speeds making everything quieter. a difference of 1C can mean the difference between deafening fans and relative quiet, especially since DT1 heat transfer is more efficient as the DT increases.

In my ideal water loop, the flow rate is high enough where you hvae a perfect steady state, and cannot measure a difference in coolant temp between any two points in the loop. The coolant is passing through so fast that you hvae a complete steady state.

No we were talking about the same Delta T, but i wasn´t specifically talking about the delta T because its just a relative number. For PC Cooling the absolute number should be more important - thats the CPU/GPU Temp.

If your water has 40°C and your delta T is 5K your CPU has 45°C... if your water has only 30°C and your delta T is 10K your CPU has 40°C. That said it is more important to get your water temp down than the delta T.

It´s just a different view point - i would never use 2 pumps to get such a high flow, first because its expensive and uses more power. Second, and that is a simple truth: If your system won´t run when its 3K hotter your cooling solution is not
sufficent at all and you need to make fundamental changes to it.

To get a steady state you don´t need such a high flow. i have about 1 liter (little less) in my system that means it will be circulating about 60 times per hour.. that is pretty fast already.

btw. my delta T (CPU) is 42K right now (125W Ryzen 2700x) - GPU its 12K both at 100% Water temp is 33°C. Flow 53,7L/h

I agree, the absolute core temp of the chip is the most important, but these deltas are key to how you design your system to get there.

The way I have my system set up, the pumps sit at their lowest speeds at idle resulting in very low flow. This allows for a pretty large coolant temperature delta across each block.

I have Calitemp sensors before and after each block. Pump speed is determined using a PID setpoint controller set at 31.7C based on the max of the outlet temperatures of the two blocks. The theory being that if your flow is low, and your coolant is hot after your blocks, but cool coming in, the fastest way to get cooler coolant in there is to move the coolant faster, and get fresh cool coolant in there fast. My setpoint controller targets

The fan speeds are governed by another setpoint controller, at 32C, based on the max of the two block inlet temperature sensors. This way, if the water coming in to the block is too hot, we know the problem is that it isn't being dispelled in the radiators fast enough so the fans need to spin faster.

At heavy load the pumps provide very high flow, but sitting idle at the desktop the flow is low, and they are completely inaudible.

In the idle state there are large Delta T's on both the CPU and GPU, and temperature in the loop differs widely based on where you measure it, with the hottest coolant coming right after the blocks, and the coolest coming right after the radiators. At high load - however, the flow is sufficiently high that the coolant temps are identical throughout the loop, just like I like them to minimize the core temp deltas.

The pump setpoint is slightly below the fan set point for tuning purposes. I have been e3xperimenting with PID settings to try to get a faster response. With both set the same I had two problems. Firstly, the flow didn't kick up fast enough when starting a heavy load, allowing the temperature to spike higher than ideal before settling in at its steady state. Secondly, I was having issues due to my excessive radiator capacity where the fans would start to spin up, keeping loop temps low enough that the flow wouldn't get going. Eventually I'd wind up with the fans on max, and the flow still low.

Setting the pump setpoint slightly below the fans seems to address this, getting the pumps going first, and then slowly raising the fan speed as needed.


I control my tolerances tightly to get the most out of the system, desiring low delta T's at full load. This does not mean my system fails if it gets warmer. I have found that core temps in both the CPU and GPU can be almost perfectly predicted based on a combination of load, loop temp and flow, with other noise factors being minimal. With a set target of 32C for the loop temp, if it gets hotter, the worst that will happen is that my fans will run faster, and I have plenty of spare capacity there.

I don't ever use the on board or on die sensors to control my fans or pumps, for two reasons. Firstly, I spend most of my time in Linux, and the Aquaero can't read those sensors under linux. I like my fan control to be OS independent.

Secondly though, I learned a long time ago that fan control based on core temps in a water loop is not a good idea. During idle states, the fans spin down very slowly, allowing heat to build up in the water in the loop. Then when a heavy load starts and the fans spin up, they cant remove that heat from the coolant fast enough, so you wind up with a temperature spike before the fans eventually bring the loop temp under control and it settles in to a steady state.

This is why I always control everything on loop temp and on loop temp only. I usually start by doing worst case load testing with everything (fans and pumps) at 100% speed. Once it has reached a steady state I note the delta T between the chips and the coolant. Then I select my ideal target temp for the chips, and set my loop setpoints to equal that target temperature minus the observed delta during the stress test. Sometimes some added tweaking is necessary to get the pumps to spin up fast enough, or when I have misjudged the worst case load and I have to tweak things a little, but usually once I get it dialed in, this is a very reliable configuration that results in the GPU and CPU max temps being completely independent of the room temperature (within reason, as long as I have the radiator and fan capacity to keep up, which with my overkill radiators I do.)