Two GPUs, three radiators

Hi all,

newbie here, please bear with me, this is my first computer build and also first custom water cooling. I tried to read through some threads, but I couldn't find anything helpful for my problem. If there is, please just refer to it, sorry and thanks in advance!

I am building a computer for some physics simulations, not for gaming, but I need quite the same performance. So far I have built in two RTX 2080 Ti Rog Strix, a Threadripper 2950X on a Rog Strix Zenith Extreme. The computer has to be quiet. Therefore I decided to go with the custom water cooling with two XR7 360 radiators and one XR5 360 with three LL120 fans each. All the rest is also Hydro X stuff for GPU and CPU, Pump etc.

Now the problem: How do I connect all three radiators in the most efficient way with just the one pump/reservoir? Everything I tried so far failed.

On the picture you can see the actual build, which is working okish without connecting the XR5 radiator (around 90% CPU and 90% GPU usage lead to 80c and 70c after some time (1-2h)).

But the wheel is not spinning in the upper GPU. So, I guess the build is not totally correct. At least for the GPUs I have to blockade in each case the opposite line I would say, so that the coolant is forced to run through the Cooling block?! I thought it could also work this way with a higher coolant flow.. :)

But I rebuilt everything a few times now, so I decided to wait for your input.

I am really looking forward to your suggestions. Thanks a lot in advance!

Best, J

https://imgur.com/wyWN0tR

PS: I am working on Ubuntu, so I cannot use the iCUE. Nevertheless everything is connected to two Commander Pros, which are connected by USB. The pump lies on 6 and the temperature sensor is also plugged in.

With such a build i would leave the pump at full speed to start with. If you still have one GPU that doesn't spin, i would try putting the CPU in parallel too. D5 pumps have a fairly high flow rate and you use no 90° fittings so it should work well.

The order is not too shabby. The GPUS are what will heat up water the most so putting the CPU before these will save you a couple of degrees on CPU temp (if you go full parallel, forget the comment :p).

Regarding loop order, as usual, it doesn't matter at all. but a good practice is to have water pass through all the rads before going back to the reservoirs, so the pump sees the coolest water possible. but again, we're talking like a couple of °C at best.

Personally, I would try a full parallel loop to take advantage of the high flow of this pump. And it makes tubing runs cleaner too.

Thanks for your comment. Full parallel sounds good. But with parallel you mean using a y-splitter at the output of the reservoir? So one loop including the CPU and a XR7, and one loop including the GPUs and both XR5, XR7 using in total two inlet ports of the reservoir? And the double/parallel string in the GPUs is fine apart from that? I have the feeling I should block one line each.

"With such a build i would leave the pump at full speed to start with" - I don't have access to the performance without iCUE, do I? But it is running full speed anyway?

Thanks! Best J

OK, the easy part - GPU spinner. It doesn't always spin. The water is absolutely still flowing. It takes a good bit of speed to make that thing go. It's not necessarily a bad thing if it stays still at low speed. You might hear it during quiet desktop work. You'll know if the flow is too slow. You will be able to see your coolant and component temps creep up in unison. It would be odd and noticeable.

Now the harder part. Right now you are set up as total exhaust. Every single fan in the system is set for that and the only inlet is the rear slots at the bottom. The fans are having a hard time. Now the 2 second fix would be to remove the rear fan and then the entire back end becomes passive intake. However, I recognize that case immediately and I have been testing in it for the past 6 weeks. To my surprise, the result I got with a three radiator exhaust, passive rear intake sucked - for lack of better word. This was considerably worse than all the other arrangements I tried, including with only 2 radiators in any configuration. This was surprising to me because I had run a "no intake" 2 radiator exhaust set-up in an 011D and it was fairly efficient and quiet. For whatever reason, the fans fight each other, create a cyclone in the middle, something... but when I popped the glass off the coolant temp dropped 4C in 60 seconds. The radiator fans could not get air. The above said, I would flip that rear fan around to intake as the "right now fix" band-aid.

In terms of 3 radiator set-ups, I have now done all the possible combinations, and there is only one that was a viable. Bottom/side intake radiators, top exhaust radiator. With that comes a caveat. Ideally, you want the heat load to go to the top rad first, then to the other side intake radiators. The problem with a three rad set-up is some of the radiators are going to dump heat into one of the others. That really limits its cooling effectiveness. By going top first, you make the top radiator the warmest, then the exhaust from the other two radiators should be 2-3C cooler and you will still get some cooling effect out of the top. In that configuration, I still get about 1C coolant temp reduction out of my top 360x30mm with QL fans. That's about all you would get anyway. Then the real workhorse rads are the the bottom and side. I get a 2-2.5C drop from the XR7 in the side intake with 120x38mm fans. And that's about the sum total of things. 1C from the top. You hope for 2C from the side and bottom and more reasonably a 3-4C coolant temp reduction for the trip through the loop. Fortunately, it looks like you would be able to run this top down set-up without major modification.

What's in the side radiator position? Is that a 30 or 54mm radiator?

The bottom and side radiators are both XR7 360 54mm. The XR5 in the top is 30mm.

So, in summary I'll turn the fans of the bottom and side radiators as well as the rear fan in order to have two main intakes and only one exhaust in the top. And take the top radiator as first input. Thanks for sharing this!

But the main question is still open: If I connect everything serial as follows: reservoir output - bottom radiator - GPUs - top radiator - CPU - side radiator - reservoir input (like in the picture, but the GPU output goes to the top radiator instead), then the pump is not able to pull the coolant through the system. It stucks after the GPU output fighting against gravity. How do I fix this? Parallel sounds reasonable to me, but how do I do that best (y-splitter at reservoir output e.g.?)? Or does it already help, when I go top radiator first and GPU to bottom radiator to reservoir input last, to have less work against gravity in the end? But I am mainly interested how to do the parallel build, since that should work best I guess @LeDoyen?

So you actually have low coolant flow? That wasn't just based on the spinner behavior? If your parallel GPU set-up is the slow down, you should see both GPU diode temps heating up past their normal offset over coolant temp (typically +15-20C for 2080 ti). I am not sure you need to run the GPUs in parallel. Each one will add in about 2C as the coolant passes through. That makes GPU #2 on the bottom 2C warmer than it would be if it was first. That is likely to be a lot less concerning then if you change pump speed and it results in unexpected pressure changes in the system and stalls flow. However, I typically don't run parallel multi-GPU blocks and will leave that open to someone who does. Mathematically, I don't see an immediate benefit to parallel with 2 cards.

If you flip fans to make bottom and side intake, then the rear is going to be exhaust (either passive or with the fan). I played around with this thinking I could force some some cooler air from the back to help increase my top of case radiator intake air temp. Instead it appeared to cause a build of other waste heat in the top and my sensor inputs on the other 2/3 of the radiator went up. Noticeable effect on RAM temps. I had never hit 50C or anywhere close to that before.

Regarding parallel loop, you'd need like one Y on your CPU block. One side taking the pump outlet, and the other feeding the GPUs. Or alternatively, just feed from the bottom GPU upto the first and to the CPU, then connect CPU outlet to GPU1 outlet to GPU2 outlet, and out to the rads.

Depends if you want to feed from the top/exhaust at the bottom or whatever.

Each waterblock will limit flow a lot more than any other component of the loop.

As it is now, if your CPU block restricts flow to say half what the pump can do, well, your GPUs have to work with that limited flow (and the pump is a but noisier at full speed because back pressure).

In parallel, if Corsair did their homework correctly, the restriction inherent to the blocks should distribute the flow evenly across all 3 and they will all be fed water at the same temperature.

It's easy enough to test serial/parallel with soft tubing so you can try and see what topology works best for you.

Personally, with 3 blocks and a D5, i'd go parallel.

It's a shame Corsair doesn't do crossflow rads, it makes 3rads setups in the O11D way neater tho

As for the pump speed, do you have any reading on Linux? how do you setup fans and pump speed?

@c-attack: Yes, the main problem is slow/no coolant flow. As I mentioned before, when I include the top radiator in the cycle (such that there are in total three rads), the coolant doesn't make it back to the pump. I think I'll change the GPUs to serial.

@LeDoyen: As far as I know, there isn't any reading for linux.

It would be nice, if you could confirm/correct my next try: reservoir output -> CPU input y-splitter -> (y-splitter output to GPU#1 input -> GPU#2 output to top radiator -> top radiator to side radiator -> side radiator to reservoir input #1) -> CPU output to bottom radiator -> bottom radiator to reservoir input #2? Thanks!

I had to draw your loop to make sure i didn't miss anything hehe

It's not necessary to split radiators for CPU/GPU. it adds more tubing runs and has the risk of having one side more restrictive than the other (the GPU one) and have most of the flow go through the CPU loop.

Here's an example of what you could do to simplify and avoid that risk i mentionned.

From the picture you sent, keeping some existing runs :

Pump outlet to bottom rad

Bottom rad to GPU2 inlet - GPU1 inlet - CPU inlet (might need a 90° fitting to connect from below)

GPU2 outlet - GPU1 outlet - CPU outlet with a Y conector

CPU Y to top rad

top rad to side rad

side rad to reservoir

This way, the split is only through the waterblocks, but the circuit then goes through the 3 rads in series.

Regarding control of the commander pro, whatever fixed speed you set a PWM output on windows, it will stay at when running standalone (without iCUE).

I suppose if you set fan curves controlled by a thermistor connected to the commander pro, it should also keep in memory fan curves you defined..

So you could maybe do a free windows install on a spare disk you have loungin' about (or do a dual boot if you fancy messin your linux install) and just use it to program the commander pro.

Before trying that you may first try to unplug the PWM cable coming from the pump, to make sure it runs at full speed. By default, the commander pro runs on silent profile which may have the XD5 pump run at like 20% PWM.

Unplugging the cable makes the pump run at its defaut 100% speed.