Is this water loop done the right way?

[Techea]

3 minutes ago, W-L said:

The parallel will ahve the fluid split evenly since the blocks will have the same relative flow restriction, in both parallel and series you will get even cooling it's just a matter of aesthetics and what you are aiming for. There is some logic in spltting the flow between the GPU's also to not have heat get passed down the line to the second GPU in a series loop but in reality since the fluid moves so fast it has little effect once the system comes up to equilibrium. 

So series flow as in it enters and exits one GPU before going to the second like in the photo to the right? You would need to stagger the inlet and outlet pipes, if for example in the current photo (ie the left side diagram) if we remove either the left or right pipe that is between the GPU's one would not get any fluid passing through it and therefore no cooling. So in a parallel setup you need both sections between the GPU's to function. 

 

This is exactly what I'm trying to convince everyone that in parallel the flow is almost nonexistent through one of the gpu. The parallel connection impedes the coolant from flowing unrestrained which subsequently reduces the cooling of the gpu. Do you agree with that? 

[W-L]

Just now, Techea said:

This is exactly what I'm trying to convince everyone that in parallel the flow is almost nonexistent through one of the gpu. The parallel connection impedes the coolant from flowing unrestrained which subsequently reduces the cooling of the gpu. Do you agree with that? 

Well no because the flow in the left diagram will be distributed evenly as long as both GPU blocks are the same in terms of restriction, the flow will split evenly between the two and merge evenly or as close to evenly between them. I can see what you are trying to say where one block may have more flow than the other where one has little to no water going through it for cooling which can occur if you had two completely different blocks. 

 

This is one of the argument cases for not doing something like this where everything is in parallel, the CPU blocks is not the same as the GPU's and you risk having a lower or higher flow rate through it as the water will take the path of least resistance. In theory yes that can occur but as long as the restrictions between this case the 3 blocks are similar enough then you will not encounter an issue with say an extra warm GPU or CPU. 

 

For the aesthetics of doing something like this to have everything in parallel is just one side of the argument but if you are only getting a performance hit of a couple of degrees max given that the flow is nice and even between all the 3 blocks some will be willing to lose that extra bit of potential cooling for the looks.

[Techea]

Thank you for putting this picture up because on the other thread I was trying to explain that I can very well go with the piping IN through the bottom and exit through the top GPU without any problems. But here is the issue with the parallel connection. The coolant will find rapidly the fastest channel to travel and one of the gpu will have parts where the coolant will simply have nonexistent movement. 

[Techea]

1 hour ago, W-L said:

Why do you think it doesn't enter from the top right? The fluid moves in a counter clockwise position due to the inlet of the pump/res being on the right and outlet exiting the pump/res on the left towards the CPU. 

 

The reason why you want parallel flow for block is to lower resistance and balance between the GPU's, at the beginning custom watercooling didn't have dedicated pumps with good pressure so parallel setups were sometimes a good option. Not to mention if anyone out there is still sporting a quad watercooled or even a 7 slot watercooled GPU setup for some reason having that in parallel is not bad idea due to it actually having an great effect on the pump's ability to push water through.

 

For a D5 pump like they have here it's more than capable of a triple block configuration in series. At the end of the day you are talking about a few degrees of difference between a series or parallel block setup. 

https://www.ekwb.com/blog/parallel-vs-serial-loop-why-choose-one-over-the-other/

W-L I commend you for taking the time to explain your point of view which is undeniably logical and well formulated. 

[W-L]

2 minutes ago, Techea said:

Thank you for putting this picture up because on the other thread I was trying to explain that I can very well go with the piping IN through the bottom and exit through the top GPU without any problems. But here is the issue with the parallel connection. The coolant will find rapidly the fastest channel to travel and one of the gpu will have parts where the coolant will simply have nonexistent movement. 

I understand what you're saying but given that they are the exact same GPU blocks in the original photo they should have very similar flow and restriction characteristics so it may not be completely 50/50 between the two it won't have one of the GPU's with zero flow at all. That is as long as you don't have a physical clog in the block. 

 

Even in the photo with the CPU and dual GPU's all in parallel all three blocks will get flow but not as evenly since most times you will find CPU blocks with jet plates to force water but it adds extra restriction (lower flow). Still some flow but not completely zero flow. 

[Techea]

True statement, thank you.

[dxzone12]

1 hour ago, Techea said:

 

Bottom line probably it doesn't really matter what hole you chose to enter and exit the gpu block the whole point was around the double connection between the gpu's called parallel bs connection which is totally wrong because in one of the gpu's the fluid will not move.

EDIT: I see someone else posted a nice diagram and equivalent explanation before I was able to type this out so feel free to disregard this.

 

This is 100% incorrect. As many others in this thread have pointed out it is just an example of parallel flow. Each gpu will get half the flow of water.

 

Lets assume the water is coming in from the top right of the gpus though it doesn't really matter. What happens is in that inlet the water can either go down through the gpu bloc or straight through the tube into the next gpu block. As water is looking for the path of least resistance much like electricity which also has parallel circuits, half the water will flow down through the block and half will flow straight through the pipe to the inlet on the next gpu block.

 

The half that flowed straight through will then go down into the second gpu block as that inlet is capped on the other side. After going down through that second gpu block it will come out the outlet on the bottom left of the gpus.

 

Now the half that flowed down through the first gpu block will come out the outlet on that gpu block and flow through the other straight pipe connecting the gpu's. At which point it will combine with the flow coming out of the second gpu block there and flow out the same outlet in the bottom left. 

 

Here is a nice breakdown from EK waterblocks about how parallel flow works, their example uses a cpu and gpu but the effect is the exact same: https://www.ekwb.com/blog/parallel-vs-serial-loop-why-choose-one-over-the-other/#:~:text=After clarifying it like that,due to the split flow.

Here is a video from JayzTwoCents explaining the same concept and showing that it has little to no effect on cooling. 

Here is a page from koolance that suggests the setups and highlights the negligible cooling differences. They make water blocks and would therefore know what they are talking about when setting them up. https://koolance.com/help-video-block-connecting

[For Science!]

Another example of real life parallel flow. Water doesn't just gush out of the first hole and trickle out the last. Assuming the size (and thus the resistance) of each hole is about the same, an equal split flow comes out of each hole. Same goes for the GPU blocks, as they are the equivalent of the holes in the sprinkler.

 

Adding a CPU block in as well slightly complicates it as the resistance of the block is different, but there is enough power in a D5 pump to supply enough flow for the higher resistance block as well (in this case, the CPU block).

[Techea]

26 minutes ago, dxzone12 said:

This is 100% incorrect. You clearly do not understand how water flow works. As many others in this thread have pointed out it is just an example of parallel flow. Each gpu will get half the flow of water.

 

Lets assume the water is coming in from the top right of the gpus though it doesn't really matter. What happens is in that inlet the water can either go down through the gpu bloc or straight through the tube into the next gpu block. As water is looking for the path of least resistance much like electricity which also has parallel circuits, half the water will flow down through the block and half will flow straight through the pipe to the inlet on the next gpu block.

 

The half that flowed straight through will then go down into the second gpu block as that inlet is capped on the other side. After going down through that second gpu block it will come out the outlet on the bottom left of the gpus.

 

Now the half that flowed down through the first gpu block will come out the outlet on that gpu block and flow through the other straight pipe connecting the gpu's. At which point it will combine with the flow coming out of the second gpu block there and flow out the same outlet in the bottom left. 

 

Here is a nice breakdown from EK waterblocks about how parallel flow works, their example uses a cpu and gpu but the effect is the exact same: https://www.ekwb.com/blog/parallel-vs-serial-loop-why-choose-one-over-the-other/#:~:text=After clarifying it like that,due to the split flow.

Here is a video from JayzTwoCents explaining the same concept and showing that it has little to no effect on cooling. 

Here is a page from koolance that suggests the setups and highlights the negligible cooling differences. They make water blocks and would therefore know what they are talking about when setting them up. https://koolance.com/help-video-block-connecting

Why Jayz 2 cents didn’t do a comparison between the parallel and series configuration in the same video to show clearly to everybody there is no difference between those. In reality there is a huge difference and those 3 gpu’s had they have been put in series most likely they would have ran at 32 degrees instead of 50. I didn’t say there is no flowing in parallel but it’s almost half of the flowing one will get if the configuration will be series built. The same guy Jayz 2 cents will install the nuts on the cpu cooler using an electric screwdriver. I saw the video with my own eyes. He doesn’t look to me college educated. Does he look intelligent to you? Not taking any advice from him. Sorry.

[Techea]

42 minutes ago, For Science! said:

Another example of real life parallel flow. Water doesn't just gush out of the first hole and trickle out the last. Assuming the size (and thus the resistance) of each hole is about the same, an equal split flow comes out of each hole. Same goes for the GPU blocks, as they are the equivalent of the holes in the sprinkler.

 

Adding a CPU block in as well slightly complicates it as the resistance of the block is different, but there is enough power in a D5 pump to supply enough flow for the higher resistance block as well (in this case, the CPU block).

You really push me right now to spend $300 on two gpu blocks and another $400 on parts to show you there is a huge temp difference between series and parallel configuration of the gpu’s.

[For Science!]

2 minutes ago, Techea said:

You really push me right now to spend $300 on two gpu blocks and another $400 on parts to show you there is a huge temp difference between series and parallel configuration of the gpu’s.

Go ahead, but why stop at two? As somebody that routinely does parallel flow on two and four GPU configurations, I do know for a fact that they are fine and are not outperformed by serial flow in any way.

 

If you're getting confused with parallel vs serial at a two-GPU level, have a look below at my quad-GPU semi-parallel semi-serial. The bottom two cards are in parallel, then the flow converges for serial, before splitting again for parallel on the top two cards.

[Techea]

19 minutes ago, For Science! said:

Go ahead, but why stop at two? As somebody that routinely does parallel flow on two and four GPU configurations, I do know for a fact that they are fine and are not outperformed by serial flow in any way.

 

If you're getting confused with parallel vs serial at a two-GPU level, have a look below at my quad-GPU semi-parallel semi-serial. The bottom two cards are in parallel, then the flow converges for serial, before splitting again for parallel on the top two cards.

If you get the same temperatures when you mount them both in parallel and series configuration that means there is something very wrong with the design of the gpu block. 

[For Science!]

Just now, Techea said:

If you get the same temperatures when you mount them both in parallel and series configuration that means there is something very wrong with the design of the gpu block. 

I think the only thing that is very wrong is your understanding of fluid dynamics.

 

I think the only way to convince you or us, one way or another is for you to go out and do your testing, as clearly you think you know better. At least after $700, you should have a nice looking custom loop, whether it be parallel or serial.

[Techea]

In conclusion you are telling me there is no difference between the series and parallel configuration using identical gpu blocks. Temperatures are the same which means the velocity of the coolant is identical in both setups.

[For Science!]

21 minutes ago, Techea said:

In conclusion you are telling me there is no difference between the series and parallel configuration using identical gpu blocks.

Yes, not just me, pretty much everybody in the thread.

 

21 minutes ago, Techea said:

Temperatures are the same which means the velocity of the coolant is identical in both setups.

The flow rate will be lower in a parallel flow, but not low enough for there to be a meaningful difference given the high flow rate of modern pumps. The flow rate in a single parallel block will theoretically be 1/n the serial rate (where n is the number of equal components in parallel), however in serial, you pay additional price in the form of additional restriction so the overall flowrate is lower in serial. And if you are in low flow rates where this mattered, you will be paying the price that the 1st GPU would be heating up the water prior to going to the 2nd and so (in contrast to the conventional wisdom of "order not mattering") it would, and you'd rather split the flow so you don't get weird clock discrepancies due to different core temps.

 

[Blai5e]

1 hour ago, Techea said:

In conclusion you are telling me there is no difference between the series and parallel configuration using identical gpu blocks. Temperatures are the same which means the velocity of the coolant is identical in both setups.

I've run both (on the same rig - 1080TI's) and the temperature difference was negligible although the second in line GPU in a series configuration was a bit hotter than the first (assuming a perfect SLI load balancing of course). Parallel simply looked better in my opinion and I now use one of those fancy dedicated parallel terminals as shown on the first page. I go further using a Phanteks D140 distro plate that allows me to run my CPU in parallel with my GPU's (also in parallel). I'm more than comfortable with the temperatures although this might have something to do with having a 240, 360 & 420mm radiators in the loop. Lastly, in a parallel loop, I lowered by D5 pump to 40% from 50% when I ran the GPU's in series.

[Techea]

9 hours ago, For Science! said:

Yes, not just me, pretty much everybody in the thread.

 

The flow rate will be lower in a parallel flow, but not low enough for there to be a meaningful difference given the high flow rate of modern pumps. The flow rate in a single parallel block will theoretically be 1/n the serial rate (where n is the number of equal components in parallel), however in serial, you pay additional price in the form of additional restriction so the overall flowrate is lower in serial. And if you are in low flow rates where this mattered, you will be paying the price that the 1st GPU would be heating up the water prior to going to the 2nd and so (in contrast to the conventional wisdom of "order not mattering") it would, and you'd rather split the flow so you don't get weird clock discrepancies due to different core temps.

 

Would you mind sharing your idle temp and load temp and what TR processor you use? 

[For Science!]

24 minutes ago, Techea said:

Would you mind sharing your idle temp and load temp and what TR processor you use? 

this machine is in a different country now, so unfortunately not really easily. Also, I was interested in GPU utilization as that is what is primarily under heavy load for our use case. I can show you one chart of our load (temp, utilization, core frequency) plotted over a long calculation. When watercooled the GPU load temperature is below 50 degrees for all cards. Idle temperature was equivalent to the liquid temperature, which was very close to ambient temperature (i.e. high 20's - low 30's)

 

 

One card (the blue one) is anomalously hotter than the other three, we confirmed this to be position independent as during maintenance we frequently switched the card around various place, it is likely (though I cannot confirm) that the hotter card is the MSI aero card whereas the other three were the FE cards. 

 

 

The TR CPU in the watercooled picture (blue coolant) is a 2990WX, but when I ran it for the tests below it was a 1950X (clear coolant, picturered below).

 

[Techea]

8 hours ago, Blai5e said:

I've run both (on the same rig - 1080TI's) and the temperature difference was negligible although the second in line GPU in a series configuration was a bit hotter than the first (assuming a perfect SLI load balancing of course). Parallel simply looked better in my opinion and I now use one of those fancy dedicated parallel terminals as shown on the first page. I go further using a Phanteks D140 distro plate that allows me to run my CPU in parallel with my GPU's (also in parallel). I'm more than comfortable with the temperatures although this might have something to do with having a 240, 360 & 420mm radiators in the loop. Lastly, in a parallel loop, I lowered by D5 pump to 40% from 50% when I ran the GPU's in series.

So you went 10% down in pump speed when you used the series setup and still recorded temperatures lower than the parallel loop with the pump at 50%? Would you mind sharing numbers in idle and load mode? After reading all the inputs from everybody I'm inclined to believe something is definitely wrong with the gpu block design.

[Techea]

4 minutes ago, For Science! said:

this machine is in a different country now, so unfortunately not really easily. Also, I was interested in GPU utilization as that is what is primarily under heavy load for our use case. I can show you one chart of our load (temp, utilization, core frequency) plotted over a long calculation. When watercooled the GPU load temperature is below 50 degrees for all cards. Idle temperature was equivalent to the liquid temperature, which was very close to ambient temperature (i.e. high 20's - low 30's)

 

 

One card (the blue one) is anomalously hotter than the other three, we confirmed this to be position independent as during maintenance we frequently switched the card around various place, it is likely (though I cannot confirm) that the hotter card is the MSI aero card whereas the other three were the FE cards. 

 

 

The TR CPU in the watercooled picture (blue coolant) is a 2990WX, but when I ran it for the tests below it was a 1950X (clear coolant, picturered below).

 

I have to admit these are ideal temperatures in idle and load mode.

[For Science!]

1 minute ago, Techea said:

I have to admit these are ideal temperatures in idle and load mode.

There is nothing wrong with the GPU waterblock design, I genuinely don't understand why you insist you think there is an issue with parallel flow. The flow splits equally, at half the flow rate, but also half the resistance, and then rejoins on the other end. 

 

I think you need to explain more clearly exactly what you think is happening in parallel flow, and what "improvements" to GPU waterblocks you propose on making.

[Techea]

2 minutes ago, For Science! said:

There is nothing wrong with the GPU waterblock design, I genuinely don't understand why you insist you think there is an issue with parallel flow. The flow splits equally, at half the flow rate, but also half the resistance, and then rejoins on the other end. 

 

I think you need to explain more clearly exactly what you think is happening in parallel flow, and what "improvements" to GPU waterblocks you propose on making.

First I need to see a GPU block in reality, hook it up an do some tests. It really bothers me because I'm sure the velocity of the fluid changes in parallel setup and that's the only variable that moves the heat away from the card. That's why when you increase the speed of the pump the fluid travels faster and the temperatures obviously get lower and vice versa.

[Techea]

17 minutes ago, For Science! said:

There is nothing wrong with the GPU waterblock design, I genuinely don't understand why you insist you think there is an issue with parallel flow. The flow splits equally, at half the flow rate, but also half the resistance, and then rejoins on the other end. 

 

I think you need to explain more clearly exactly what you think is happening in parallel flow, and what "improvements" to GPU waterblocks you propose on making.

Of course the longer the loop and the more radiators you mount in the system the lower the temp goes, obviously.

[For Science!]

1 minute ago, Techea said:

First I need to see a GPU block in reality, hook it up an do some tests. It really bothers me because I'm sure the velocity of the fluid changes in parallel setup and that's the only variable that moves the heat away from the card. That's why when you increase the speed of the pump the fluid travels faster and the temperatures obviously get lower and vice versa.

Pump speed is a variable, but it is far from the only variable, and it is also far from being the rate limiting variable.

 

It's not really the point of this video, but see that indeed as flow rate decreases, component temperatures do increase as the delta across the inlets and outlets of each component also increases. However, even going from the fastest flow achievable with a D5 to the slowest, the increase in CPU temperature is less than 6 degrees.

 

There are many other factors in play and flow rate is not that important.

 

[Techea]

14 minutes ago, For Science! said:

Pump speed is a variable, but it is far from the only variable, and it is also far from being the rate limiting variable.

 

It's not really the point of this video, but see that indeed as flow rate decreases, component temperatures do increase as the delta across the inlets and outlets of each component also increases. However, even going from the fastest flow achievable with a D5 to the slowest, the increase in CPU temperature is less than 6 degrees.

 

There are many other factors in play and flow rate is not that important.

 

Yes, the guy in the video explains clearly what's happening with the cpu temp when you decrease or increase pump speed. Exactly what I saying myself about the gpu temp that is directly related to the velocity of the fluid in the system. If you also increase the number of loops and add more radiators to the system maintaining the same velocity then the temperatures automatically go down.