XD5 pump acting strange (?) after radiator swap.

[eracer1111]

I just finished upgrading my radiator (liquid-cooled GPU, air-cooled CPU.)

I have a flow detector inline, and the pump ran as expected when I disconnected the CPU power and GPU power cables, and had the motherboard jumper connector attached to the 24-pin cable. (This removes power from the CPU, GPU, MOBO fans, RAM, etc. but still allows the pump to run.) Useful for flushing a loop.

I just put everything back together and powered the system on. I don’t see any flow at all. I know the pump is getting power, because if I disconnect the 24-pin cable, and reattach the jumper, I see flow.

There’s a temp sensor that runs through iCue that controls the pump speed. I verified that the pump fan is connected to the correct jumper on the MOBO.
I know there’s liquid in the water block and the rest of the loop, but how long before that heats up and damages my GPU without flow?  Long enough for the BIOS to load and the pump speed control to start?

I can’t say that I’ve paid much attention to the flow detector while the computer starts up. But it troubles me that I could leave the system running for a minute or more normally and not see anything.

I’m really puzzled as to why the pump isn’t running immediately, but does when I use the 24-pin jumper. I mean, it’s connected directly to the power supply. Shouldn’t that power be applied as soon as I turn the computer on?

[LeDoyen]

you have the pump PWM connector on a motherboard fan header, so the pump will run at whatever speed that channel is set to. when you used the jumper, the PWM header wasn't working so the pump was running at its default which is full speed.

Now you have to set a speed setting on that fan header to set your pump speed (or connect it to a corsair controller if you use one to control it from iCUE)

[eracer1111]

9 hours ago, LeDoyen said:

you have the pump PWM connector on a motherboard fan header, so the pump will run at whatever speed that channel is set to. when you used the jumper, the PWM header wasn't working so the pump was running at its default which is full speed.

Now you have to set a speed setting on that fan header to set your pump speed (or connect it to a corsair controller if you use one to control it from iCUE)

Thanks for the reply.  I've been running this loop for a year with a cusom curve.  It's only after the radiator upgrade that I noticed zero flow when I powered the computer on normally.

I guess I need to just let it boot into BIOS.  I don't think I have the Smart Fan curve set to 0 RPM at any point, but maybe I do.
I hope the GPU isn't under enough load to cause a problem if there's no flow at all during that process.  And if there is a problem with the header, that there's enough of a heatsink in the water block to prevent any damage while it boots up.

Edited April 28 by eracer1111

[c-attack]

Are you sure it’s really zero flow?  Don’t use the spinner as a guide. Those things have a specific inertia to overcome and it’s probably a bit higher than most people expect. True zero flow states are pretty easy to spot. Your gpu temp will start skipping upward at a consistent rate. 45..50..55..60..etc. If your flow rate is a bit too slow, you’ll see only a small 5C or do float at idle.  
 

You can check voltage vs flow by quickly increasing the pump speed up to 3000+. If you see the gpu temp instantly drop 5-10C, then you water was lingering in the block a little too long. Each loop has a unique minimum speed based on the amount of vertical travel and overall flow resistance from components and other minor sources. You can use the test above to find the minimum pump speed that does not cause an unloaded gpu to creep up in temperature. 
 

10 hours ago, eracer1111 said:

know there’s liquid in the water block and the rest of the loop, but how long before that heats up and damages my GPU without flow?  Long enough for the BIOS to load and the pump speed control to start?

Almost zero chance. The gpu is not loaded at boot. You need to create a sustained gpu load to see a rapid temp escalation.  Sitting on the desktop, your gpu temp should essentially match the coolant temp measurement in CUE. The liquid temp is the minimum gpu temp with zero load. If your gpu temp is sitting 5C above the liquid temp on the desktop, your flow rate is likely too slow and water is lingering in the gpu block. 

[eracer1111]

Sounds good.  I'll boot the system up normally, go into the BIOS, and check the pump speed.  Then I'll let it load to Windows and run the Open Hardware Monitor gadget as usual.

I think the key is what you said about not using the spinner to determine zero flow.

[eracer1111]

Forgive the long writeup.

All is well.  I booted into the BIOS.  The pump was running at a lazy 850 RPM (PWM Smart fan curve) but the flow meter still wasn't turning.  No problem, it's due I'm sure to the intertia c-attack spoke of.  I launched Windows, and opened HWMonitor to watch GPU load and temp, and also the pump speed.  I ran a 4K full screen Kombuster stress test for 30 minutes and temps were about 5°C lower than before the radiator upgrade.  So the radiator with the new fans made a difference.

Max pump speed was still a bit low at 1500 RPM.  I ran MSFS2020 in 4K with DLSS DLAA enabled, which really stresses this 3090 Ti (constant 99% load, 440W) and while core GPU temp had gone down from 78°C to 73°C I wanted a little more cooling. I could have played around with the pump curve in BIOS, but decided to move it from the motherboard header to iCue, trying all the different profiles.  With the 'Hydro-X XD5 Pump' profile, and Extreme profile on both rad fans I found that with MSFS running I got core temp down to 68°C and a hotspot temp max 83°C.   Much better than the 78°/90° before the swap.  Pump runs at a pretty constant 2300 RPM with that profile, and the fans a pretty constant 1300 RPM each.  The system is a bit noisier, but I can live with that. 

Kind of a side note:  When I swapped the radiator I also moved the spinny thing from the radiator outlet to the waterblock inlet side.  I would have thought that alone would make the flowmeter spin faster, since it was directly inline with the pump outlet.  It was the pump speed curve in BIOS that was fooling me.  I did however notice yesterday that the vane on the flowmeter seems backwards.  I must have flipped the in and out ports around when I moved it.  I wondered if maybe that's contributing to the way the wheel turning less vigorously, so I looked up the manual for the EKWB flowmeter.  It said that any port can be used as an inlet or an outlet.  I thought about draining the loop and flipping it over, but no way I'm wasting $20 worth of coolant if the temps are good now.  Maybe I'll flip it around when I flush/fill in a year.

Thanks for the help.

Edited April 30 by eracer1111

[c-attack]

2 hours ago, eracer1111 said:

The pump was running at a lazy 850 RPM (PWM Smart fan curve) but the flow meter still wasn't turning. 

That is what I was expecting and 800 rpm is a typical baseline speed with a minimal PWM signal.  That's too slow for almost any loop and will come with a temp penalty.  You have a one block, one radiator loop and the GPU block is much less restrictive than a CPU block.  You likely won't see any improvements past a specific middle pump speed.  2300 rpm seem likely, but obviously you can test by running something with a steady GPU load and then moving the pump speed around in 500 rpm increments.  Th Hydro X profile is a preset like any other.  No special sensing technology and it's a list of data points.  It might be perfect in one loop or inappropriate in another.  Don't feel tied to it and you can create your own curve or use a fixed speed that is a good blend of performance and noise.  

[eracer1111]

4 hours ago, c-attack said:

Th Hydro X profile is a preset like any other.  No special sensing technology and it's a list of data points.  It might be perfect in one loop or inappropriate in another.  Don't feel tied to it and you can create your own curve or use a fixed speed that is a good blend of performance and noise.  

I can create a custom curve in iCue?  Or are referring to playing around with the temp/speed curve associated with the header on the motherboard?

I know I can do it in BIOS, but with mine I can't tie speed to the temp sensor on the GPU itself, unless the option they call 'PCI-E' refers to that.

In iCue I tried Quiet, Balanced. Extreme, and Hydro-X XD5 Pump.  It seems I get the best balance of thermal control and with the latter.  Like I said, it's louder, but not too bad.  This is a flight sim computer, and a little extra wind noise is hardly noticable. 😄

Edited April 30 by eracer1111

[c-attack]

You can create a custom pump curve in CUE if you have it connected to a Commander Pro/XT. Go to Cooling for the Commander and click on the yellow + bar. That will reveal a graph below. 
 

GPU temp should be there as NVIDIA xxx Temp or AMD xxx temp. However, that’s a poor control variable. You don’t want or need the pump to drop speed when you open a map in game or spike back up when you resume. Coolant temp is the proper control variable but don’t get too caught up in the minutia. You either are fine with a fixed speed and can leave it alone or a small speed shift to differentiate between idle and sustained load coolant temps.