Log SMOL 4.2L - Fully watercooled workstation with RTX 3070 and 5900X

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85


This has been a long going design process. I began working on this years ago but haven’t come around to do it.
This has had some benefits, some parts have become more SFF friendly, and by revisiting my design I have shrunk
the case from 9.7 L down to 4.2 L, without losing any performance. (In theory at least).

I mainly use my rig for work, which includes heavy simulations, rendering and multitasking.
I like to game sometime, but when choosing between a larger case and a 3070, I chose the latter.

To get this to work I need to do a couple of things that has to do with temperatures.
I started out with making the assumption that most people overestimate the amount of cooling needed when water cooling.
This usually makes sense, bigger radiators isn’t that much more expensive and more fans does probably help to keep the noise down if done correctly.
At the same time, LTT showed us that while not satisfactory, it is possible to cool a 3090 on a 120mm radiator (66 degrees with full turbo).
This got me thinking. The 3090 has an average power consumption of 350W. The 3070 has an amazing performance at only 200W average.
That would potentially leave room for a high-end CPU.
The only thing I could not change size of was the motherboard, which became the starting point for about 20 iterations of different placements with different hardware.




When looking through cooling options I first opted out of water cooling because of the extra parts that would take up space.
I began looking at it again after realizing the thickness of the GPU is critical as it is hard to place somewhere where it fits.
After opting in to water cooling again, I began looking for radiators which is as large as the motherboard, but not larger. A 140mm radiator would be suitable here.

When this was figured out I had to find a suitable GPU. The 3070 FE was really the only alternative.
No other latest-gen GPU has a PCB below 170mm. Finding a 3070 FE is hard enough (I am yet to get my hands on one).
Finding a block for it is even harder. The blocks for the 3080/3090 FE versions are rather large and I would assume the 3070 blocks would be of equal size.
I decided I must make my own GPU-block to make this viable.

Specs:
- X570 of some sort, the flatter the better.
- 5900X / 5950X (Same TDP, can't really defend the extra cost for 5950X but would be cool for performance)
- 3070 FE
- 64 GB, 3600MHz RAM of some kind, probably Vengeance low-profile.
- 500-700W Flex ATX - PSU (been looking at a few options, but most are on sites I do not know the language of).
- CPU block/Pump/res combo, been looking at the Alphacool LT Solo, might be too thick.
- Some slim fan, probably a NF 12x15.
- A 140mm radiator.
- Case, SMOL, by me.

ToDo:
Order components for practical measurements
Tubing size test with 10mm ID and 6mm ID for differences in tempertaure and flow.
Fan test for radiator with regards to temperature
Get hold of a 3070 FE...
Design waterblock for 3070 FE.

Challanges:
Make sure 140mm radiator that is heavily covered by CPU block will be enough for cooling high-end CPU + GPU
Fitting it all in 4.2L and below, this number might rise or fall depending on any last minute design progress.
Making a block that will be slim and cool sufficiently with zero air flow.

I would love your input, whether it is ideas or something I've missed. I'm also aware of that 4.2L might be overconfident, but I need something to drive innovative thinking.
 

harveyy_sh

Chassis Packer
Nov 14, 2020
14
7
The G Unique, I am not sure which one that is. It is the brick one?
I’m not sure but I think he has brickless psu’s that go over 600w.


I wish there was a 700w flex atx from reputable brand.
 
  • Like
Reactions: SFFManiac07

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
I’m not sure but I think he has brickless psu’s that go over 600w.


I wish there was a 700w flex atx from reputable brand.

Exactly. I have no doubts these guys who make custom PSUs know what they are doing. But I also know that testing generally is a lot stricter in the industry.
 

paulesko

Master of Cramming
Jul 31, 2019
415
322
FSP have something that can fit your case

it´s only 500, but I guess it could be probably enough
or this one

if not, try with one of these server 1u psu. they are longer, but very very slim... you´ll probably will need to mod some things, but these are top quality

You also has the enhance version with 600w, and the variants, which have some sellers here in this forum (If I remember correctly). Enhance is a very reputable brand. I believe they manufacture silverstone power supplies
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Dimensions?

Right now the outer dimensions are 216*173*112 = 4.185L . They might change depending on my thermal tests.
If my thermals will be alright I might be able to sink the fan and pump into the radiator, making it 216*173*107 = sub 4L
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Hardware just showed up at my doorstep!

First of all we have the motherboard, Gigabyte X570 I Aorus Pro WiFi



5900X also arrived, I went for the 5900X in the end. The 5950X would be cool for performance per liter, but not for performance per cost.



Mounted!



I also got hold of a Corsair XR5 140 Hydro Series Radiator. Hopefully this will fit my needs.
Below it is on top of the corrent cooler height. I am not sure which distance I need from the block to radiator, but I think a few mm will do.



Right now I am trying to get hold of a 3070FE for further measurements.
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
So this showed up! It has been a hassle to get hold of this one.



I feel bad about removing this cooler to watercool it...


Gorgeous



Also had the RAM show up. 2x32GB sticks for this one. Finally I will be able to keep all my tabs browsing the internet...



I will test this card out ASAP and hopefully get some good results of undervolting it!
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Been sick for the past days, taken a covid-test, but I have had some strength to chill out with building computer things.

Before I can take the GPU apart I need to make sure to know how flow is needed in terms of cooling. The pump is very weak, but it is the best I can fit. So I had to set up some thermal testing.



The Pump/CPU block will be the Alphacool Eisbaer LT,




To save some space I ordered a Noctua NF-A12x15,





As you can see this shaved off 15mm of the chassi height! About 0.55L saved!




I started making some base-line benchmarks using Cinebench R23 with a normal BeQuiet cooler. Went up to 74 degrees, with CPU power consumption reported at 115W. I tried using AMD master software for a quick undervolting which brought it down to 94W and a temperature of 65*C with sustained results.



I then put the loop together, using the 16/10mm tubing. The flow is horrible, which was not even enough to use the flow-meter. I measured by using a glass and I calculated the flow to be around 30L/h at most. This doesn't matter much as long as the temperatures are alright though. I fired up Cinebench and did the same testing. With stock settings it performed similarly to the BeQuiet cooler, at 72 degrees. When undervolting, and cranking the fan speed up to levels I was okay with in terms of noise, I did manage to bring it down to 61 degrees with sustained results







I do wonder how the pipe diameter might change the flow though. I was interested in using 8/6 mm tubing, to make it fit easier with push in fittings. I will do some more testing of this later this weekend. Below is an image for comparison.







And the push in fitting







Ordinary fluid dynamic formulas show that 6mm dia tubing should have 13% of the flow in a 10mm tube. However the flow might already be restricted by the blocks so much that the tubing size does not really matter.



Next update will be when I have been able to test the small tubing!
 

harveyy_sh

Chassis Packer
Nov 14, 2020
14
7
Been sick for the past days, taken a covid-test, but I have had some strength to chill out with building computer things.

Before I can take the GPU apart I need to make sure to know how flow is needed in terms of cooling. The pump is very weak, but it is the best I can fit. So I had to set up some thermal testing.



The Pump/CPU block will be the Alphacool Eisbaer LT,




To save some space I ordered a Noctua NF-A12x15,





As you can see this shaved off 15mm of the chassi height! About 0.55L saved!




I started making some base-line benchmarks using Cinebench R23 with a normal BeQuiet cooler. Went up to 74 degrees, with CPU power consumption reported at 115W. I tried using AMD master software for a quick undervolting which brought it down to 94W and a temperature of 65*C with sustained results.



I then put the loop together, using the 16/10mm tubing. The flow is horrible, which was not even enough to use the flow-meter. I measured by using a glass and I calculated the flow to be around 30L/h at most. This doesn't matter much as long as the temperatures are alright though. I fired up Cinebench and did the same testing. With stock settings it performed similarly to the BeQuiet cooler, at 72 degrees. When undervolting, and cranking the fan speed up to levels I was okay with in terms of noise, I did manage to bring it down to 61 degrees with sustained results







I do wonder how the pipe diameter might change the flow though. I was interested in using 8/6 mm tubing, to make it fit easier with push in fittings. I will do some more testing of this later this weekend. Below is an image for comparison.







And the push in fitting







Ordinary fluid dynamic formulas show that 6mm dia tubing should have 13% of the flow in a 10mm tube. However the flow might already be restricted by the blocks so much that the tubing size does not really matter.



Next update will be when I have been able to test the small tubing!
What push in fittings are those??

Hope you feel better soon!
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Challenging project, good luck with it! :thumb:

Thanks!


What push in fittings are those??

Hope you feel better soon!

Those were just the first I could find, just took some 8/6 push-in fittings I saw on aquatuning. They work wonders! I was sceptical at first, but I can now show that fitting/tube size does not matter even at 6mm thickness. I just did the same tests as earlier today and I got the same results! 72 degrees in stock and 62 degrees with undervolt.

Test-setup

 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Hi! Tested positive for covid last week and has been incapable of getting anything done the past week. I have been able to pull the 3070 apart though for some waterblock designing! This PCB is amazing for SFF builds.



Next I measured every part individually to be able to model it in scale in CAD. That will help with block planning a lot. This also mean I can make simulations easily and check various parameters as flow, temperatures, restriction and pressure drops within the block. Below is my proposed water flow. I have marked in yellow the parts which Nvidia have pads on which I will assume needs cooling. As I have no fans blowing air on the card, I will probably put some pads on extra parts. If you have any input, please tell me! The block design is one of the most critical in this build. Especially if I want to try to aim for a SUB 4L.





Thanks for cheering me on! Next time I will show you some simulations and tests of the waterblock.
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
4.2L is small enough. I have faith, but sub 4L will take a miracle, or some sort of mad genius. Rooting for you.

Thank you! We will see about that 4L.... Right now it seems a bit off the limits, all depends on how thin I can make the GPU-block.



Block-design is kind of ready for production, I ran some simulations today to make sure I got no deadspots and a good flow. I will 3D-print a test-block later to make sure everything fit as it should. I went with a droplet low resistance high-flow design to cope with the low pressure of the Alphacool pump. This ensures a low pressure drop in the block while maintaining a large area for heat transfer. Water temperature should rise about 7.5 degrees given the flow and heat output. On the simulation I estimated VRAM power at 30W, VRM at 20W and core at 200W. I'm not really sure about the VRM and VRAM real output, however as I've been able to push my power consumption down to 160W from 216W I have a lot of play. Under no circumstance should the heat output to the water be more than 160W. Brining the CPU down from 105W -> 80W also make sure that I will have a full output of around 240W, which would heat the water 15-30 degrees at most.


Below are my benchmarks with peak power consumption at wall. With stock settings in blue and undervolted (but maintained performance) in orange.
I'm really satisfied with bringing down the power consumption this much. Will really improve my thermals given the small radiator size. It also give me confidence
on the 500W PSU from Silverstone.

 

Runamok81

Runner of Moks
Jul 27, 2015
445
621
troywitthoeft.com
Ah! So... when you say in the original post you do a lot of "heavy simulations," does that include thermodynamic water simulations? Or, is this new territory? I ask mostly because I understand that thermodynamic simulations are a real PITA, so much appreciation.
 

Hiraeth

Caliper Novice
Original poster
Nov 20, 2020
27
85
Ah! So... when you say in the original post you do a lot of "heavy simulations," does that include thermodynamic water simulations? Or, is this new territory? I ask mostly because I understand that thermodynamic simulations are a real PITA, so much appreciation.

Thermodynamic simulations are new territory (this is the first real simulation I've done!), however I do more and more FEM-analysis and structure analysis specifically in school and work. Thermodynamics sims are really PITA yes, I still haven't figured out a way to apply heat further away and conduct it through the copper, in the simulations above I had to apply the heat directly to the water (which is still a good approximation, but if I could apply the heat to the copper instead I could literally measure the die-temperature of the GPU which would be cool. I also struggle with setting correct boundary conditions. Knowing input pressure is not something you easily calculate. In my case I have set the output pressure to 0 and adjusted the input pressure so that the energy dissipated matches the theoretical values for heat transfered to the liquid. But in my case I don't think the boundary conditions matter that much. I know that the heat dissipation works well, I just need to make sure there are no dead spots and flow works in the way it should.

I'm also trying to learn more about AI and machine learning, in which a good GPU can take you a long way from my understanding. I feel like I could use the computing power in many fields to come!
 
Last edited:
  • Like
Reactions: SashaLag

BaK

King of Cable Management
Bronze Supporter
May 17, 2016
930
931
Thermodynamic simulations are new territory (this is the first real simulation I've done!), however I do more and more FEM-analysis and structure analysis specifically in school and work.
I guess you are using a specific software for these simulations, can you share a little more about it?