S4MAX: Brickless S4M w/ 3090 FE and R9 5950x - 800W, 5l, water cooled

Ezzoud

Caliper Novice
Mar 25, 2018
28
7
Hello, first time poster, long time viewer here. Sorry if this sounds stupid, but ive allways wondered if you could watercool a psu. Seeing your craftsman skill i think making a small waterblock for that halfU psu seems doable. Again, im not an expert, heck i dont even know what is it in the psu that heats up, just thought id ask if its even possible. Youre gonna be running a loop anyways, so youll allways have a very small chance of a leak, that could end up frying your parts. You would also lose the psu fan entirely , and considering the cooling capability of that 40mm fan your radiator wouldnt take a massive hit.
 
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
well in theory that's all possible! That said, water and AC power make for a bad marriage, you'd want to make sure things are well separated. This particular PSU is a difficult candidate though as water cooling relies on contact between the hot parts and the cold plate of a cooler; an air cooled layout such as this spaces those parts well and keeps them away from the PSUs case to maximise exposure to the air stream. There are PSUs built for conductive cooling (such as the MeanWell UHPs - see link to my previous build), where the heat generating parts are immersed in a heat conducing resin - these can indeed be quite efficiently attached to a water block.

For this specific build I like to keep as many heat sources as possible air cooled and away from the cooling loop as it will quite likely become the performance bottle neck - the less load on the loop, the higher i can set the throttle....
 
Last edited:

Ezzoud

Caliper Novice
Mar 25, 2018
28
7
When you get to the testing point of the loop, could you run 200mm fans vs 4x100mm or 4x92mm fans to see if pressure makes alot of diffrence.
 
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
Good news: I managed to spoof the fan rpm sensor! Yay!

Bad news: I fried the PSUs logic board along the way!


Got lazy and didn't use a resistor when pulling the PSU's PS-ON low...


worked for a few minutes and then the smell of burnt pcb announced the certain death of a vital component.
Lesson learned I guess...
New brick is in the post (quite some delivery lead in), and I may unsolder the logic board when I find a minute to see whether I can identify what has blown up before the new part arrives.

Anyway - what you see in the image above is step 1 of my spoofing operation: Measuring the fan speed of the beast that comes with the PSU. Detaching the PWM wire gets it going at full blast (note that I have to hold it in place - it actually starts to fly off when not holding it). I get a tacho frequency of 535 hz, translating to an impressive 16,000 rpm. That would explain the noise.


In standby/ POST mode it reads at 153hz or 4,600 rpm - that's near the Noctuas maximum rpm of 5000 which explains why the PSU would only switch on when the Noctua spins at full tilt. Slowing down the fan with my finger tip I get the PSU to work up to about 100hz, meaning the minimum standby fan speed required by the PSU is around 3000rpm.


Time for spoofing - enter the TP354:

It's a tiny module with a NE555 signal generator and a potentiometer to produces a square wave adjustable between 50Hz and 6kHz.


It conveniently fits onto a standard fan connector cable.


Adjusting the potentiometer for the module to produce a 155Hz signal is the easy bit - this should simulate a 4650 rpm fan. Assuming the PSU is alife...

More interesting is the question which resistor to use on the output side. I have chosen 510 Ohm to not exceed the NE555's 200mA rating - it should return a 23mA signal. That said, an intel PWM signal spec I found says 5mA - that means the resistor should be more like 2.4 kOhm. Thinking of it, I might well have burned the PSUs logic board with the PWM signal... Guess I'll need to invest some more time into forensics, and once I have a new PSU I'll have to work my way back from a 100kOhm resistor and see when it starts to pick up.

Did anyone successfully do this?

As cooling parts start to trickle in, I can keep myself busy fitting hardware into the case whilst waiting for a new brick to burn through, so next up will be some more tangible stuff!
 
Last edited:

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
Time to tackle the cooling!

Pretty much all I need apart from some custom parts that I plan to get machined has arrived by now:

We look at pump, reservoir, various fittings, 140 mm rad and fan, cpu block, some rotary fittings for the GPU and a bunch of preliminary heat sinks for VRM and Chipset (a generic north bridge heat sink I intent to marry with a small 40x10 noctua) as I had to discard the stock cooling solution that came with the ASRock itx/tb3. End game will be to build a passive cooling solution for VRM and Chipset, probably using heatpipes, to connect them to the case, but for now i’ll use tried and tested means and focus on the water loop.

Speaking of which, some more items from my last round of shopping include some EK Cryofuel, a handy filling bottle...




...and Festo PUN-H 10mm OD industrial pneumatic tubing with liquid rating and height thermal resistance to go along with the Festo NPQH push in fittings i’ll use for this build. Primary reason for choosing this system is that it comes with a huge range of different tubing and threading diameters, allowing me to build a hybrid G1/8 and G1/4 loop which is required to satisfy my clearance constraints. Also, the tubing allows for down to 52mm bending radius without flow rate impact and down to 28mm assuming added resistance.

Key acquisition, however, is this awesome little gem here, that I could only get my hands on thanks to the great help of @confusis (thanks again!) who shipped it to London all the way from New Zealand:

It is what would appear to be pretty much the last available original EK Annihilator on the entire planet. EK stopped making them in 2015, and after contacting them just to find out that they have none in stock anymore, I went through their entire global dealer directory, finding a store in NZ that still had a single one on the shelf - with a neat discount on top, and in an almost bizarre co-incidence, apparently around the corner of @confusis...

What makes this block special is a unique combination of design decisions leading to an inredibly low profile of about 20mm. It has been designed as a 1u server block, so the fittings are parallel to the board...

...and as you see above, it’s a square module that comes with an LGA1151 adapter and backplate (it’s the same coming with the original EK Supremacy block) making it a perfect match for the LGA1151 form factor of the ASRock x570 itx/tb3 I’m using. All later models of the EK Annihilator were designed for larger server sockets and would have required massive modifications to fit.

The other key feature is that it uses G1/8 fittings instead of the commonly used G1/4, saving some more precious height (shown here with the default narrow ILM bracket):


To install the LGA1151 bracket and swap out the fittings to suit my configuration I need to more or less fully take it apart, giving a good view to the copper block inside:


I wil be replacing the original barbed fittings with these Festo NPQH G1/8 to 10mm OD push in rotary elbow fittings...


...with flow from port P2 to P4.


Swapping of the bracket is pretty straightforward...


...and after transferring the sprung thumb screws from the narrow ILM bracket I look at something that‘s ready to be fit to my board!


Time to get these two bad boys out of the garage:



...and also to take a first and close look at the 3950x: Packing contains about 98% foam...

...and 2% processor, owed to AMD not even attemping to ship an air cooler with it. But those 2% promise to be rather exciting.


The rear is what always makes me a bit nervous when looking at AM4s (particularly the expensive variety)- a whole bunch of pins you definitely do not want to mess with. The LGA1151‘s layout with the fragile bits on the board side is definitely more robust...

According to the manual, the EK Annihilator wants the LGA1151 backplate to be replaced with it’s own (EK Superior) backplate:


When trying to place the protective film on the rear of the board, however, it becomes apparent that this is a bat ides as the ASRock ITX/TB3 doesn’t have a compatible keep out zone and would have the backplate crushing some vital components...


...meaning that the original backplate has to come back on again to work as a spacer, and for the rather robust one coming with the Annihilator to go on top:

This provides (just) enough clearance over the components on the rear of the board. At this point, however, there are two concerns:
1) Is the EK bracket projecting too high? The original plan was to shorten the motherboard stand-offs for the NVME drive on the rear of the board (in the background) to touch the aluminum outer case. This is necessary as a) there is no space for a heat sink (PCIE4 NVMEs are said to run hot), and the case is supposed to work as such, and b) it is impossible to fit the GPU on top if I cannot gain at least 3mm clearance by lowering the main board in the case.

2) will the finger screws holding the water block be long enough to fit?

To verify the latter, its time to put the CPU in place...

...and marry it to the cooler for a physical fit, with no thermal grease applied at this time as I might still have to re-arrange things a few times in the course of the build.

And: screws are long enough - just about!

The next check is the most relevant one: The NVME projects 5.1mm above the board’s rear side and has to be the highest point for the cooling plan to work.


And to my big relief, the sandwiched brackets only build up to about 4.9mm - that was close!

This also means that by padding the bracket with .2mm of thermal tape, I can establish full contact with the case and activate it as an additional heat sink to the CPU; given the predictable thermal constraints this build will face, this is a welcome opportunity.

With dimensions confirmed, I can now trim the standoffs of the case by 3mm...

...and go for a test fit of the board:

And things appear to add up, with NVME drive and CPU sitting flush to the bottom of the case.


There is a small clash between the rear I/O cover holder and the case that will require some trimming in the metal workshop along the lines sketched here - to not make a complete mess of the case I’ll need to do this with proper tools.

Final task for today is to confirm clearances together with the GPU:
Whilst not being able to place it in it’s final position (the case notch for the PCIE connector is another one for the metal workshop) it gives me a good enough indication on whether the vertical stack pans out:

In this alignment it is fitting more or less watertight with the case. A bit too tight for comfort as I need to avoid contact between GPU PCB and case, and also still need to accomodate for the thickness of the PCIE riser cable connector (went for the HDPLEX ultra thing riser, should arrive soon).


Plan is to cut a slot into the plexi cover of the GPU water block to accomodate for the wifi module (the cover is sheer cosmetics at that location, the actual water block starts further towards the card’s rear).
This will allow to lower the gpu down to the top face of the CPU block’s finger screws and release some 5mm of vertical clearance...

...and for the GPU to sit on this level shown here - plus about 1mm for the riser cable that needs to go in-between CPU and GPU



Next up: Pump and radiator test fit!
 
Last edited:

nightmr

Chassis Packer
Apr 8, 2017
19
8
my suggestion is just to disable the pcie 4.0. For the time being, you cannot see the actual benefit in experience rather than pcie 3.0. But disabling it can not only save a lot of space for cooling, but also run a lot safer. There is a an sepc of alienware aurora which has no heat sink on the chip and runs perfectly on pcie 3.0.
 

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
my suggestion is just to disable the pcie 4.0. For the time being, you cannot see the actual benefit in experience rather than pcie 3.0. But disabling it can not only save a lot of space for cooling, but also run a lot safer. There is a an sepc of alienware aurora which has no heat sink on the chip and runs perfectly on pcie 3.0.
Interesting point... will run a few benchmarks once I can power things up... would expect to see some difference in m.2 performance (have a PCIE 4 drive) but yes for the GPU that shouldn’t matter...
 

AlexTSG

Master of Cramming
Jun 17, 2018
599
590
www.youtube.com
I found the thread for this project this morning, and two cups of tea later I've read through it all.

If you do come across more shiny things that are tough to photograph, you can try angling a large piece of white board/foamcore above the components and shining the light onto the board so they're lit indirectly. Of course, I'm sure everyone here would rather you spend any free time you have tweaking the build and not the photos.

Can't wait to see more!
 

Curiosity

Too busy figuring out if I can to think if I shoul
Platinum Supporter
Bronze Supporter
M...M...M...M...Multi-Tier...Subscriber...
Apr 30, 2016
708
827
Finally got the images to load now that the website is more stable!

That must have been mighty satisfying to see how perfect the fit for the backplate and nvme drive were!

As always, I'm watching with anticipation for how this continues to go!
 
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
Today I have some more plumbing coming up, starting with what will make or break this build: Alphacool’s NexXxos ST30 V.2 140mm, the single cooler for both CPU and GPU.


Inside the box is a well-packed radiator, a manual and a small box with accessories...


...containing three sets of mounting screws, a matching Allen key, and a tool to undo the two low profile G1/4 screw plugs coming with the radiator.


Further, some Koolance NZL-LXG2 low profile rotary fittings, and a non-rotary Alphacool elbow fitting, form part of the plan. The non-rotary is a few mm shorter, key for this build as the height of the cooling package is severely constrained. Hypothesis is that the orientation of the radiator fitting thread may allow rotating a non-rotary fitting into the right direction, using the leeway of the o-ring, to save a little height.


And indeed, it looks like I get a sufficiently tight fit using the simple elbow on the side constrained for height - on the other side I should have enough clearance to use the rotary Koolance fittings.


With a double rotary Bitspower elbow added, the radiator outlet should be in the right place to connect to the pump/reservoir combo...


...the quite recently#released Alphacool Eisstation 40 DC-LT...


...coming with pump mounting screws, matching Allen key, a straight rotary fitting for radiator attachment and yet another screw plug tool.


It will go along with an Alphacool DC-LT 2600 pump (that's the silent one - there is still the beefier 3600 out there in case things require more throughput)...


...that's coming with an additional set of (long) mounting screws and a mounting ring to better distribute pressure on the pump's o-ring.


Together this makes for an incredibly compact package that should just fit behind the radiator!


Mounted pump-down and the reservoir facing up, the central scew plug port makes for a good fill/ vent port.


Next come the Festo NPQH push-in fittings to hook the pump/res/radiator unit up to the 10mm OD cooling loop:


Fitting nicely and where they are supposed to go!


In this arrangement, the cooling pack leaves enough space for a full-size C14 connector (including EMI filter ) to go into the rear of the case - preferable over the C6 I have used in my last build, as when connected to a 110v AC circuit, a C6 would exceed its rating at the 600w the PSU can potentially draw.


The Cryorig XT140 completes the package...


...for a total 43mm build up before modding the fan - I figure I can save another mm by machining pockets into the fan to accommodate for the folding edge profile of the radiator.


That all makes for a pretty much air-tight fit in the case - practically not requiring any fasteners to hold it in place!



to verify the build's key dimensions, I insert the (fried) PSU...


...and the board into the case - seems to pan out!


Next is adding a EK-FC Rotary terminal to the GPU cooler. This will be a temporary solution - it allows for me to close the cooling loop as planned using fittings exiting the card parallel to the PCB plane rather than perpendicular to it. End game will be to use a custom machined part with G1/8 fittings to make space for full height RAM, as currently there appears to be no low-profile RAM available (let me know should you find some... >=3200 Mhz, <= CL14...) allowing to leverage the full potential of the Ryzen 3xxx platform.


Inside the box are the terminal (heavy PTFE coated forged brass - unlike the default terminal, made of acetal), a set of mounting screws, two o-rings and a matching Allen key. Some silicone oil would have been a nice add-on - it's what you'd like to put on the o-rings before mounting them.



It all fits rather nicely and makes the GPU ready for connection to the Festo loop using G1/4 push in rotary terminals...


...that should allow the tubing to be pointed in the right direction.


Time to check how it stacks up - quite literally:


...and as predicted in CAD, the inner radiator port clashes with the GPU cooler. Despite the shorter non-rotary elbow fitting working out, I'm still about 8mm too tall!


In plan view it's all more or less checking out with the GPUs rotary fittings staying well-clear of the low profile ram (you can imagine how a custom port module with smaller G1/8 fittings will allow to use full height memory modules...):


Key challenge, however, is that unfortunately milling into the perspex GPU block cover to provide a pocket for the radiator outlet is not an option. As you might be able to see below, the clash with the elbow fitting is extending beyond the o-ring line of the water block - that's a non-starter.


Conclusion: Custom inlet to the radiator...
 
Last edited:

Curiosity

Too busy figuring out if I can to think if I shoul
Platinum Supporter
Bronze Supporter
M...M...M...M...Multi-Tier...Subscriber...
Apr 30, 2016
708
827
Fantastic progress, what a wonderfully dense rad unit! For the custom rad inlet are you planning to drill a hole and thread your own g1/4 (g1/8?) Threads?

Is that risstation smaller than the acrylic top they offer? You've made me wonder if I made a mistake getting the acrylictop with my dclt haha.
 
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
Fantastic progress, what a wonderfully dense rad unit! For the custom rad inlet are you planning to drill a hole and thread your own g1/4 (g1/8?) Threads?

Is that risstation smaller than the acrylic top they offer? You've made me wonder if I made a mistake getting the acrylictop with my dclt haha.
Yeah will have to drill a hole and solder-in a G1/8 fitting... discussed that with a friend recently and it turns out that the has the mother of all metal workshops in his basement. He's currently experimenting with a turning lathe to see how small a custom fitting he can machine from a brass rod...
his kit will also come in handy when machining the acrylic GPU waterblock top and some case parts as they will have to lose quite some material!

re Eisstation: It's a little bit larger than the acrylic top but makes good for it by having an insane number of connection options - somehow fits my build better. But most importantly, it comes with a built-in reservoir and bubble filter promising to address the air bleeding challenge that my build would otherwise have faced. All still to be proven though...
 

Curiosity

Too busy figuring out if I can to think if I shoul
Platinum Supporter
Bronze Supporter
M...M...M...M...Multi-Tier...Subscriber...
Apr 30, 2016
708
827
Yeah will have to drill a hole and solder-in a G1/8 fitting... discussed that with a friend recently and it turns out that the has the mother of all metal workshops in his basement. He's currently experimenting with a turning lathe to see how small a custom fitting he can machine from a brass rod...
his kit will also come in handy when machining the acrylic GPU waterblock top and some case parts as they will have to lose quite some material!

re Eisstation: It's a little bit larger than the acrylic top but makes good for it by having an insane number of connection options - somehow fits my build better. But most importantly, it comes with a built-in reservoir and bubble filter promising to address the air bleeding challenge that my build would otherwise have faced. All still to be proven though...
I'm jealous, I've got access toa makerspace but that's the limit of my tool access unless I want to buy things. Very lucky to have that friend! Haha.

I see. I'm going to be using a single 149 just like you are, hopefully my build won't be too much of a pain to bleed lol
 
  • Like
Reactions: petricor

paulesko

Master of Cramming
Jul 31, 2019
415
322
One thing. If you happen to try the 3600 rpm version of the pump take in mind that I have found it waaaaay noisier than the 2600 rpm version. Even with matching rpm I don´t know why it makes a hell of a lot of noise. I´ve tried it in a eisbaer so it may be different with the eisstation though. If noise is a thing for you take that in mind.

Great build, I´m enjoying reading it, really. keep going!
 

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
336
1,820
One thing. If you happen to try the 3600 rpm version of the pump take in mind that I have found it waaaaay noisier than the 2600 rpm version. Even with matching rpm I don´t know why it makes a hell of a lot of noise. I´ve tried it in a eisbaer so it may be different with the eisstation though. If noise is a thing for you take that in mind.

Great build, I´m enjoying reading it, really. keep going!
yeah that's what I have heard - was going to go for the 3600 initially but thought I'd give the 2600 a shot and see whether it copes as I like it silent...
 
Last edited: