• Save 15% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!

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

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Gulp... 87 C at 460W from the wall. I am able to undervolt to 390W and keep a decent 15224 (17k graphics score) Timespy score.
I think I am going to add a waterblock to the outside of the PSU where the internal heatsink is. I am hopeful that will keep the temps down while not impacting the rest of the loop too badly. the interesting thing about the g-stick is the fan is in the center of the PSU , splitting the PSU into two separate sections. The problem is the fan is way underpowered. I've thought of replacing it.
Otherwise I'll go the supermicro route.
Well that is perhaps a tad too warm… can you squeeze in an A4x10 on one end to move some more air through?
 
Last edited:

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Time to progress matters a little further - on the menu today:

Using this to complete my chipsed cooling solution!

First up, a longer version of my 6mm VRM pipe - the 250mm one should cut it:


Starting with the right length for the VRM block...


...it folds pretty much exactly to end within the chipset block. Practise helps I guess!


Next I need to apply some thermal pads to the VRM blocks - using 0.5mm 6W/K*m gel pads in a double layer facing the VRMs, and in a single layer to fill the gap to to the chokes.



Fits, and the 1mm pad leave the block standing above the upstands, so the screws will be able to apply some pressure onto the pad and VRMs.

Next point of order is fitting the fin stack onto the fan:


I start by cutting some slots into the fan's mounting hole channels allowing me to fix it to the case using cable ties (will become clearer later...)


Next, I mount the fin stack onto the A4x10 using some wire...

...to tie it to the fan's screw holes.



Now I can slot-in some cable ties that will make for means of attachment to the case:


Once I insert the 5x150mm heat pipe, the fin stack loses its wobbliness...


...and allows me to place it into the case for a little test-fit and confirmation of my bending hypothesis.



A few minutes later the first pipe looks about right...

...and I can tackle number two...



...to land on the chipset block. Having tested this in 3D really helped to get an idea of the radii I need to achieve, and once the spacing of the bends is set out on the pipes (that's the black marks you see), it becomes surprisingly easy.

That looks about right!
As you see I have re-installed the board into the case, as the VRM heat pipe covers one of the screws attaching the board to the sub-frame - so it has to be mounted before I can fix the pipes to the board.

With that sorted, I can start to cut some thermal pads to measure:

I use a 0.5mm 6W/m·K silicone gel sheet, with a double layer on the chipset block matching the original 1mm pad on the ASRock cooling solution.

The same double-layer approach helps me to pad up the 6mm grooves in the chipset block to properly make contact with the two 5mm pipes connecting it to the fin stack.


With the heat pipe assembly fully equipped with thermal pads...


...I can fit everything into the case and zip the fan to the case's frame (that one came out a bit blurry...)


...and do a first fix of the wiring within.


I have installed the water block from radiator to CPU to make sure I have enough space at the junction over the x16 slot- as you can see, it's getting rather busy there:



Next step is applying some pressure to the block over the central VRM bank: As the original screw holes in the PCB are covered by the heat pipe, I resort to cable ties...

...that I thread through the original screw holes...


...and tie on the underside of the PCB...

...pulling down the heat pipe and block onto the VRMs.

With the heat pipe system installed...


...the question remains: Does it work?

I could just take off the chipset cooler and see whether I see a nice imprint of the chipset in the silicon pad - but I much rather prefer to take some temperature readings. That implies that I need to switch it on - and that, in turn, means:


TIme to slap on my brand new 16 core district heating plant!

Not much of an unboxing ceremony here - the box contains 98% air, 2% chip,


...which finds its way into the socket rather quickly. I have to say that installing an intel CPU is a much more relaxing experience - the prospect of potentially breaking off a pin of a GBP1k CPU does not contribute to steady hands.

So far, so good!

To make sure things don't evaporate, the water block needs to go back on...


...together with some MX-4, applied in a highly scientific X-pattern.



That looks usable - up next: Taking my other S4M apart for a GPU transplant, and some temperature readings!
 
Last edited:

harveyy_sh

Chassis Packer
Nov 14, 2020
14
7
@petricor nice work!!!

quick tangent question. I have a server psu that gives out 12vsb. Do you have any suggestions to convert it to 5vsb? Would a step down work or would it cause problems in the future? Thanks!
 
  • Like
Reactions: petricor

Valantar

Shrink Ray Wielder
Jan 20, 2018
2,201
2,225
Now, admittedly I absolutely suck at any math whatsoever, let alone geometry, but doesn't adding two .5mm thermal pads to a 5mm diameter heatpipe add 1mm to the radius (not the diameter) and thus increase the diameter to 7mm? Wouldn't a single .5mm pad be sufficient, and likely to perform better too?

Beyond that, great work! Really looking forward to seeing this all come together.
 
  • Like
Reactions: petricor

riba2233

Shrink Ray Wielder
SFF Time
Jan 2, 2019
1,776
2,359
www.sfftime.com
Now, admittedly I absolutely suck at any math whatsoever, let alone geometry, but doesn't adding two .5mm thermal pads to a 5mm diameter heatpipe add 1mm to the radius (not the diameter) and thus increase the diameter to 7mm? Wouldn't a single .5mm pad be sufficient, and likely to perform better too?

Beyond that, great work! Really looking forward to seeing this all come together.

Yeah but they do compress somewhat, I just don't know by how much.
 
  • Like
Reactions: petricor

reallamer

Chassis Packer
Jan 17, 2019
15
11
Time to progress matters a little further - on the menu today:

Using this to complete my chipsed cooling solution!

First up, a longer version of my 6mm VRM pipe - the 250mm one should cut it:


Starting with the right length for the VRM block...


...it folds pretty much exactly to end within the chipset block. Practise helps I guess!


Next I need to apply some thermal pads to the VRM blocks - using 0.5mm 6W/K*m gel pads in a double layer facing the VRMs, and in a single layer to fill the gap to to the chokes.



Fits, and the 1mm pad leave the block standing above the upstands, so the screws will be able to apply some pressure onto the pad and VRMs.

Next point of order is fitting the fin stack onto the fan:


I start by cutting some slots into the fan's mounting hole channels allowing me to fix it to the case using cable ties (will become clearer later...)


Next, I mount the fin stack onto the A4x10 using some wire...

...to tie it to the fan's screw holes.



Now I can slot-in some cable ties that will make for means of attachment to the case:


Once I insert the 5x150mm heat pipe, the fin stack loses its wobbliness...


...and allows me to place it into the case for a little test-fit and confirmation of my bending hypothesis.



A few minutes later the first pipe looks about right...

...and I can tackle number two...



...to land on the chipset block. Having tested this in 3D really helped to get an idea of the radii I need to achieve, and once the spacing of the bends is set out on the pipes (that's the black marks you see), it becomes surprisingly easy.

That looks about right!
As you see I have re-installed the board into the case, as the VRM heat pipe covers one of the screws attaching the board to the sub-frame - so it has to be mounted before I can fix the pipes to the board.

With that sorted, I can start to cut some thermal pads to measure:

I use a 0.5mm 6W/m·K silicone gel sheet, with a double layer on the chipset block matching the original 1mm pad on the ASRock cooling solution.

The same double-layer approach helps me to pad up the 6mm grooves in the chipset block to properly make contact with the two 5mm pipes connecting it to the fin stack.


With the heat pipe assembly fully equipped with thermal pads...


...I can fit everything into the case and zip the fan to the case's frame (that one came out a bit blurry...)


...and do a first fix of the wiring within.


I have installed the water block from radiator to CPU to make sure I have enough space at the junction over the x16 slot- as you can see, it's getting rather busy there:



Next step is applying some pressure to the block over the central VRM bank: As the original screw holes in the PCB are covered by the heat pipe, I resort to cable ties...

...that I thread through the original screw holes...


...and tie on the underside of the PCB...

...pulling down the heat pipe and block onto the VRMs.

With the heat pipe system installed...


...the question remains: Does it work?

I could just take off the chipset cooler and see whether I see a nice imprint of the chipset in the silicon pad - but I much rather prefer to take some temperature readings. That implies that I need to switch it on - and that, in turn, means:


TIme to slap on my brand new 16 core district heating plant!

Not much of an unboxing ceremony here - the box contains 98% air, 2% chip,


...which finds its way into the socket rather quickly. I have to say that installing an intel CPU is a much more relaxing experience - the prospect of potentially breaking off a pin of a GBP1k CPU does not contribute to steady hands.

So far, so good!

To make sure things don't evaporate, the water block needs to go back on...


...together with some MX-4, applied in a highly scientific X-pattern.



That looks usable - up next: Taking my other S4M apart for a GPU transplant, and some temperature readings!
what unstoppable man
 
  • Like
Reactions: petricor

Questors

SFF Lingo Aficionado
Oct 28, 2018
124
82
Very nice! The level of customization and personalization is very interesting.

I would love be able to hack my cases to bits to create what I want. Well, I mean, I have years ago, but not anymore. The wife would skewer me if I blundered into ruining $225 to $450 boutique case.
 

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
@petricor nice work!!!

quick tangent question. I have a server psu that gives out 12vsb. Do you have any suggestions to convert it to 5vsb? Would a step down work or would it cause problems in the future? Thanks!
Hi there, there various options for that - easiest and cheapest is a Pico plugin PSU (just google it, there are plenty of options and wattages available), delivering +5V (amongst others) to a 24-pin ATX plug from where you could divert it. I use a modded 160W Pico for my build.
There are also fancier and more performant versions by @guryhwa or @Kmpkt (have a look at the "vendor" section). That's all assuming you want to divert some power to a 5v application - for all of it, you may want to look for a dedicated 5v PSU.
 

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Now, admittedly I absolutely suck at any math whatsoever, let alone geometry, but doesn't adding two .5mm thermal pads to a 5mm diameter heatpipe add 1mm to the radius (not the diameter) and thus increase the diameter to 7mm? Wouldn't a single .5mm pad be sufficient, and likely to perform better too?

Beyond that, great work! Really looking forward to seeing this all come together.
It would indeed, but I'm only adding them at the bottom half, so that shifts the pipe up by one mm - 6mm total vertical cross section. That, of course, leads to a radius mismatch in the bottom half - but here is where compression comes to play as @riba2233 points out... and some of that is desirable to make good contact!
 

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Little update from the workshop: Frankenstein day!


Looks like my loyal 8700k-build (left) will have to play organ donor and go to hibernation for a while to fire-up my new Ryzen :(
BTW did I mention that I am still looking for a GPU? 🥳

Aim is to transplant what you see in the upper left corner (my apparently still desirable 1080)...

...to the bottom right and to run some temperature tests on my south bridge cooler.

Nothing new here, really - but I like this photo ;)


A few screws later, I have successfully isolated my modded 1080 (that's the shroudless thing under the big Noctua in the bottom-left corner...)

Also a nice display of the progress in PSUs made over the past 3 years - that dual combo here delivers 360W, compared to the 800W of the Supermicro below, with 2/3rds of the volume...

As, predictably, the 1080 wouldn't fit, I'll have to park it next to the case. The power cable I have built for the 2080ti fits, so I should be good to fire things up...

...pending an improvised water loop being installed to keep the CPU from melting.

Fires up! (phew.. always a bit of a moment with a new CPU in a modded build)

So, long story short, the cooling solution appears to work. Running Timespy, the south bridge goes up to 60 degrees celsius max, and grilling the Ryzen with Prime95 gets me to 71 degrees tops - and note that the heat from the VRMs, which will run at full tilt during this test, blends in into this reading. That's considerably better than the 102C(!) I had before (and that's pretty certainly the point where it throttles), on the south bridge alone...
I'll still have to hit it with a few games and VR (pushing a lot of data over USB3 and thus stressing the southbridge), but I'm fairly confident that this is ticking the box.


As finishing touch I add some copper fins to the VRMs/ chokes, interestingly they have little effect on the south bridge heat readings (down from 71 to 70 in Prime95), which appears to indicate that the bottleneck of the cooling setup is the heat transfer from x570 to the cold plate.
 

harveyy_sh

Chassis Packer
Nov 14, 2020
14
7
Hi there, there various options for that - easiest and cheapest is a Pico plugin PSU (just google it, there are plenty of options and wattages available), delivering +5V (amongst others) to a 24-pin ATX plug from where you could divert it. I use a modded 160W Pico for my build.
There are also fancier and more performant versions by @guryhwa or @Kmpkt (have a look at the "vendor" section). That's all assuming you want to divert some power to a 5v application - for all of it, you may want to look for a dedicated 5v PSU.
Yes I want to ue the same pico psu that you use but the power supply I have doesnt provide 5v, so I was wondering if using a buck converter to convert the 12v standby signal to 5v signal would work well. Then of course connect it to the pico psu. Thanks!
 
Last edited:
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Ah gotcha @harveyy_sh. Yes that should work in theory... there are plenty of circuits of various capacities available OTS. I gather you'd need 2A for a compliant current. For a simpler solution, you could keep the PSU always-on by pulling PS-ON low with a resistor directly at the PSU, and then feed the Pico straight from the AC-DC. With a wifi-plug, you can then shut down everything properly when not using the machine (and set it to boot-on-power, allowing you to start everything via app).
 
Last edited:

harveyy_sh

Chassis Packer
Nov 14, 2020
14
7
Ah gotcha @harveyy_sh. Yes that should work in theory... there are plenty of circuits of various capacities available OTS. I gather you'd need 2A for a compliant current. For a simpler solution, you could keep the PSU always-on by pulling PS-ON low with a resistor directly at the PSU, and then feed the Pico straight from the AC-DC. With a wifi-plug, you can then shut down everything properly when not using the machine (and set it to boot-on-power, allowing you to start everything via app).
Nice! I’ll look into that. Thanks!
 
  • Like
Reactions: petricor

petricor

Airflow Optimizer
Original poster
Bronze Supporter
May 12, 2018
347
1,851
Did you heard about Intel ATX12VO? This thing can make server psu much more usable :) (no need for Pico PSU).
Looks promising, indeed... given how much progress has been made in CPUs/ GPUs in past decades, the component interface standards seem somewhat behind...