Try Running RoG Real bench Open CL test and watch your voltages..... if you are under Adaptive control, I expect you will exceed 1.5 volts.
Deja Vu all over again
Again, your reasoning is perfectly logical and well thought out ..... but there are different ways of looking at the issue and under the different scenarios, the best solution may vary. This is mainly due to the following:
-High end GPUs produce more than twice the heat of CPUs so in a twin GFX system, you are looking at a 4.5 to 1.0 ratio of heat produced. They can in no way therefore be considered equal contributors to the heat issue.
-In a WC loop, it's rare to see a GPU hit 50C .... 30C below where we start being concerned. OTOH, CPUs oft hit 75, 80C or more, well into the area which we get concerned.
1. First off, PWM is not relevant to the issue of temp control / fan speeds. Here's why .... if instead of using your CPU or CPU_OPT header which are PWM, you used the 3 or 4 pin MoBo Headers (These are variable voltage control and respond to the same CPU Temp criteria), how would things change with respect to the fans ? Nothing changes with regard to speed control. If you stop using PWM but your fans are still responding to CPU via voltage control signal, eliminating PWM won't change anything. There are some fans that have a particular noise to them but a) the easy solution is just avoiding them, b) you can control 3 pin voltage controlled fans in response to CPU temps with PWM header and Phanteks PCB eliminating any PWM noise as the PWM signal never reaches the fans or c) ya get no noise using VV fans and MoBo CHA headers.
2. As for the CPU versus GPU issue, you can connect to the fan header signal from the GPU if you want to control the Rad fans downstream of the GPU separately.
3. As we discussed before, even with synthetic GFX benchies have you found any of them where the GPUs only heat up ? In any demanding current game, my CPU temps are mid 60s whereas the GPU temps are in mid to hi 30s. While in theory, the logic is totally sound, in practice the problem never materializes. Even in Furmark which I have running right now, w/ CPU running at a mere 40 watts, the CPU Core temps hit 53C (hottest core)..... (51C on Valley)..... while the GPU temps stabilize at 44C w/ 850 rpm fan speed limit .... If there is something you can run on a PC that gets the GPUs hotter than the CPUs on a WC build, I haven't seen it. So while you have a laid out a perfectly logical solution to the problem of the GPU being hotter than the CPU, I don't think that is a problem that anyone has ever actually experienced.
4. As for the starting / ending early thing .... again, you have a perfectly logical solution. But again, is there a "problem". Due to the thermal mass of the loop, how long it takes ya system to stabilize will depend on how much coolant is in the loop and to a lesser extent the thermal mass of everything in the loop. But if ya fans go from 550 rpm to 850 rpm, yes the geek in me says that I have made my system more efficient ..... but 3 minutes after P95 shutdown, my fans are still at 710 rpm ...at 6 they are at 617 rpm..... the FanXpert2 Software also provides for a 3.5 minute spin up / spin down time and even lets you program in a fan stop point.
5. With CPU only loads, my coolant temps don't really change that much .... Under P95, with an ambient temperature of 20.5, my coolant temp is 25.7 going into the rad downstream of the CPU and 25.2 coming out of the rad upstream of the CPU ....a Delta T across the rads of only 0.5C....my worse core temp peak was 84C (AVX loads and Adaptive). If I was controlling based upon water temps, I'd really be concerned about having more pump and fan speeds when CPU temps are that high than when delta T = 5C.
In summary, everything you said (except for the PWM part), makes perfect logical sense ...... in theory. But in a practical sense, I don't think the problems which the solution addresses arise all that often (if ever) and, if so, alternatives exist which provide equivalent functionality.
a) PWM and VV are just two different means of accomplishing the same thing. To control ya system via CPU temperature with PWM, connect what you want to control to one of the two PWM headers on the MoBo (CPU or CPU_OPT). To control ya system via CPU temperature with variable voltage (VV), connect what you want to control to one of the many 3 or 4 pin headers on the MoBo (CHA or OPT).
b) If you wanna control fan speeds on different rads separately, you can use the fan headers on the GPUs instead of the MoBo. Not something I'd recommend
c) I have yet to find any program that I can run that drives my WC'd GPUs to a higher temp than my WC'd CPU. As my GPU temps have never risen above 44C with fans running, having the fans respond to any change of temps is of no consequence. Running at fixed minimum speed still keeps them > 30C below any temp I'd be concerned about.
d) Ramping fans up and down is provided for via the pump curves and setting ramp up and ramp down speeds. However, if you ignore them, what is the impact on the user experience if the fans did not start and end early ? Saving 4.5 watts for 3 minutes say 6 times a day ?
From a power cost perspective .... 4.5 watts x 3 min / 60 min per hr x 6 times 1 kw / 1000 watts x $0.10 per kw/hr = 5 cents a year
From a noise perspective ..... in my case there really is 0 impact as my fans are dead silent from 900 rpm down. You you don't hear anything at any of those speeds and since nothing has as yet made me exceed 850 when fan curves are controlling, there is no impact. However, if you went light on rads and made up for that with high speed fans..... you would be able to observe a difference.... that being a 3 hour gaming session would only have perhaps 2 hours and 57 minutes of observable noise.
I have easily spend 50 - 60 hours running tests, tweaking fan curves and the like but must honestly say, other than loads of geeky enjoyment, it has not impacted the system or the user experience in any observable way. If set the fan curves to a fixed 850 rpm, the only effect is that they would be running faster than they need to.... there is no observable noise impact and the power savings is barely measurable.....so, one could argue, what's the point ?
Many MoBos today come with OPT thermal sensors. Asus for example has 3 OPT thermistor inputs on the M6F and you can read the temps in the BIOS and in the AISuite software. They also provide 3 OPT fan headers. I do think at some point in the very near future we will see those OPT thermistor inputs and OPT fan headers linked via the FanXpert software.
e) If I was controlling fan speed based upon Delta T and my worse case was 10C .... say:
Delta T = 1C = 25% fan speed
Delta T = 2C = 25% fan speed
Delta T = 3C = 30% fan speed
Delta T = 4C = 40% fan speed
Delta T = 5C = 50% fan speed
Delta T = 9C = 90% fan speed
Delta T = 10C = 100% fan speed
What I would have is my fans running very close to full speed (90%) and in the audible range at 9C under huge GPU load ..... do I really care whether temps are 39C or 44C if I gotta pit up with fan noise ?
OTOH, my fans running at 50% speed (Delta T of 5) with a max CPU core temp of 84 under a heavy CPU load and I'd be very much want it to be in the 90%+ range at that temp. At 84C, unlike the GPUs 40ish temp, I am starting to get concerned about temps. Instead of 0.5C across the rads and a max coolant temp of 25.7 at 50% speed.... I dropped 8C to 76C on CPU temps and max coolant temp was 24.6 at 90% speed
To summarize, I don't think there is a "best way" to control fan speeds in all circumstances..... there is a best way for each of several conditions and circumstances.
-Control based upon GPU is rather meaningless as under any conceivable circumstance, your GPUs are in no danger of overheating.
-Control based upon CPU does make sure that the only component in danger is always of primary concern..... CPU testing does result in hi CPU temps and meaningless GPU temps and GPU testing, even with GPU benchies still produces significant CPU temps. Under any circumstance I have experienced with a water cooled system, GPU's never exceeded CPU temps.
-Control based upon water temps will work well in high GPU loads and in mixed load situations but will be less than ideal under CPU only loads because the CPUs thermal outputs is typically only 15% or so of the thermal load a twin GFX card system is designed for. the control system will therefore be indifferent to hi CPU temps because while CPU temps will be bear max, the system coolant temps will be scarcely be above idle....since stopping P95, I went from 24.6 max coolant temp @ 90% fan rpm (25.7 @ 50%) to just 24.3
Its that last scenario that is troubling .... with fan speeds based upon coolant temps, how do I get my fans to 90% at 24 - 26C (CPU temps at 80) under a CPU test ..... and then at an inaudible speed when gaming with the coolant temps at 30C, CPU temps are 65C and GPU temps are at 40C ?