-
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
- Thread starter petricor
- Start date
Update #2: The Summer '22 edition of my ventilation layout!
As many here note, cooling the Supermicro PSU is a challenge in it's own right - and who would have ever thought that cooling PSUs would become an ongoing issue?
So, after having moved from a super-silent tiny Noctua 40mm whisper fan (worked with a 2080TI - but not with a 3090 FE) to bypass-powered bleed air from my Noctua Industrial 140mm monster, I am now at a point where I conclude that I need to look at the PSU independently from my main fan for the water loop, and with some more beef to it.
Key reason is that I have realised that I need to keep the main fan cooling the radiator at full blast as soon as the system starts to pull moderate power for the PSU not to run into a thermal shutdown - theoretically a workable option, but leading to intolerable noise levels at almost all time. And, on the other hand, I noted that from about 1,600 RPM upwards, the radiator's performance doesn't significantly increase anymore: Running the build at full tilt with synthetic loads on CPU and GPU the system sees temps levelling out at about 60% fan speed with water in the low 50's (C), and CPU/ GPU in the mid-80s and below throttling.
That means: The PSU now drives the noise level from the main radiator fan - and that it could run at just 60% of that if I were to cool the PSU separately. And unless I crank it to 11 I look at thermal PSU shutdowns during intense tasks (well, gaming, really!).
Another thing that isn't quite ideal is that the Aorus B550 board I have chosen for it's compactness and avoidance of the x570 cooling challenge does not come with external temperature sensor headers - meaning that I have to drive the radiator fan from either CPU or GPU temperature or an arithmetic mix of both, but not from water temperature - that means I am foregoing the potential to keep noise down by using capacity of the coolant to buffer temperature/ performance spikes.
Long story short: I need independent fan controls, temperature sensors telling me what's going on in PSU and water loop, and a way to cram this into my case as my mobo wouldn't want to talk to external probes.
Translated into parts it looks like this:
From left to right:
The fundamental challenge coming with this is that, unfortunately...
...the only place where the Aquacomputer would fit has only half of the height that the bare part takes...
...and that doesn't even account for all the connectors that would eventually stick out of it to connect fans, sensors, power (the ATA power plug is particularly nasty) - so this will need some work.
What you see x-ed out above is a power connector I thought to use to power it directly from the PSU, tapping into the 5 and 12V lines of it...
...I quickly ditched that idea after remembering that I still have a spare SATA power header on my Pico PSU that just seems to have been waiting for this.
And regarding the size of the Quadro: Lucky enough, all I'm really after is in a rather neat and flat PCB package...
...so some unsoldering and cutting later...
...I look at a low profile version of my Aquacomputer that should fit the bill - assuming I just solder anything I need directly to the PCB.
A little later, I am looking at this:
The cable on the upper side of the photo plugs into my Pico PSU, the two large connectors to the fans I want to drive, and the two two-pin interfaces are for temperature probes I want to hook up to the module. The USB cable connector coming with the Aquacomputer fortunately is small enough to not get in the way and can be used as intended.
A test fit confirms:
Should to the trick, and a few minutes later I take things for a test drive before replacing the PSU fan:
It's alive!
So, next is the PSU:
Here you already see the Noctua A4x40 replaced by an Evercool EC4015SH12BP. The specs and use case could not be more different from the Noctua: It's essentially be biggest punch I could find combining 15mm depth (that's exactly how much space I have left for a fan) with PWM speed control - idea being: Turn it down if you don't need it, crank it up if you do - and it's still likely to be significantly less noisy than a Noctua A14 Industrial running at full tilt. For comparison, the Noctua A4x10 revs up to 5,000 rpm and pushes 8.9 m3 airier hour - the Evercool does up to 10,000 rpm and 25 m3 of air per hour... slightly different proposition!
It's worth noting that I still need to keep my signal generator mod:
Despite not controlling the fan from within the PSU anymore, I still need to simulate the feedback from the original monster fan that comes with the Supermicro PSU for it to stay happy and not refuse operations - so I disconnect if from the fan but still keep it wired to the PSU's controller.
And here is where I got a bit lazy taking photos:
Next step is placing a film temperature sensor on thePSU's air heat sink - about where the blue mark is below...
...and a sticky silicone pad filling the gap between case and sensor on top of it to keep it attached and pressed against the heat sink at all times - see the sketch above.
The finished result looks like this...
... with the PWM fan control (the coloured cable) and the temperature sensor cable (black) being routed to the outside through one of the fan's mounting holes. The blue box in the top right corner above the power connector is the signal generator used to spoof the fan logic.
Last thing missing is attaching a temperature probe to the water return of the loop: Very similar to how I have attached the probe to the PSU's heat sink...
...I use a strip of thick adhesive silicone tape to stick the sensor directly onto the steel fitting, and secure everything with a cable tie to ensure there is always pressure on the sensor, keeping it firmly in place. NB that these sensors are fragile - the (transparent) silicone tape I use for padding is 4mm thick, reasonably soft, and distributes the pressure from the cable tie avoiding breaking it.
A bit of tidying up - and that's it!
Independently operated fans...
...with a software interface that allows me to employ different profiles for different tasks, and targeting heat sources directly.
Makes for pretty silent operations at light loads - and has fans kicking in properly when hitting it hard.
And, most importantly: No more thermal shut downs! And for those struggling with Supermicro temps: My thermal sensor data tells me that it trips at about 72 C measured at the top of the internal heat sink - may be a helpful data point.
As many here note, cooling the Supermicro PSU is a challenge in it's own right - and who would have ever thought that cooling PSUs would become an ongoing issue?
So, after having moved from a super-silent tiny Noctua 40mm whisper fan (worked with a 2080TI - but not with a 3090 FE) to bypass-powered bleed air from my Noctua Industrial 140mm monster, I am now at a point where I conclude that I need to look at the PSU independently from my main fan for the water loop, and with some more beef to it.
Key reason is that I have realised that I need to keep the main fan cooling the radiator at full blast as soon as the system starts to pull moderate power for the PSU not to run into a thermal shutdown - theoretically a workable option, but leading to intolerable noise levels at almost all time. And, on the other hand, I noted that from about 1,600 RPM upwards, the radiator's performance doesn't significantly increase anymore: Running the build at full tilt with synthetic loads on CPU and GPU the system sees temps levelling out at about 60% fan speed with water in the low 50's (C), and CPU/ GPU in the mid-80s and below throttling.
That means: The PSU now drives the noise level from the main radiator fan - and that it could run at just 60% of that if I were to cool the PSU separately. And unless I crank it to 11 I look at thermal PSU shutdowns during intense tasks (well, gaming, really!).
Another thing that isn't quite ideal is that the Aorus B550 board I have chosen for it's compactness and avoidance of the x570 cooling challenge does not come with external temperature sensor headers - meaning that I have to drive the radiator fan from either CPU or GPU temperature or an arithmetic mix of both, but not from water temperature - that means I am foregoing the potential to keep noise down by using capacity of the coolant to buffer temperature/ performance spikes.
Long story short: I need independent fan controls, temperature sensors telling me what's going on in PSU and water loop, and a way to cram this into my case as my mobo wouldn't want to talk to external probes.
Translated into parts it looks like this:
From left to right:
- An Aquacomputer Quadro capable of driving four fans and four temperature probes
- An Evercool EC4015SH12BP 40x15 mm 12V PWM Fan
- the USB cable connecting the Aquacomputer Quadro to a main board header
- and two film temperature probes for PSU and Water Loop
The fundamental challenge coming with this is that, unfortunately...
...the only place where the Aquacomputer would fit has only half of the height that the bare part takes...
...and that doesn't even account for all the connectors that would eventually stick out of it to connect fans, sensors, power (the ATA power plug is particularly nasty) - so this will need some work.
What you see x-ed out above is a power connector I thought to use to power it directly from the PSU, tapping into the 5 and 12V lines of it...
...I quickly ditched that idea after remembering that I still have a spare SATA power header on my Pico PSU that just seems to have been waiting for this.
And regarding the size of the Quadro: Lucky enough, all I'm really after is in a rather neat and flat PCB package...
...so some unsoldering and cutting later...
...I look at a low profile version of my Aquacomputer that should fit the bill - assuming I just solder anything I need directly to the PCB.
A little later, I am looking at this:
The cable on the upper side of the photo plugs into my Pico PSU, the two large connectors to the fans I want to drive, and the two two-pin interfaces are for temperature probes I want to hook up to the module. The USB cable connector coming with the Aquacomputer fortunately is small enough to not get in the way and can be used as intended.
A test fit confirms:
Should to the trick, and a few minutes later I take things for a test drive before replacing the PSU fan:
It's alive!
So, next is the PSU:
Here you already see the Noctua A4x40 replaced by an Evercool EC4015SH12BP. The specs and use case could not be more different from the Noctua: It's essentially be biggest punch I could find combining 15mm depth (that's exactly how much space I have left for a fan) with PWM speed control - idea being: Turn it down if you don't need it, crank it up if you do - and it's still likely to be significantly less noisy than a Noctua A14 Industrial running at full tilt. For comparison, the Noctua A4x10 revs up to 5,000 rpm and pushes 8.9 m3 airier hour - the Evercool does up to 10,000 rpm and 25 m3 of air per hour... slightly different proposition!
It's worth noting that I still need to keep my signal generator mod:
Despite not controlling the fan from within the PSU anymore, I still need to simulate the feedback from the original monster fan that comes with the Supermicro PSU for it to stay happy and not refuse operations - so I disconnect if from the fan but still keep it wired to the PSU's controller.
And here is where I got a bit lazy taking photos:
Next step is placing a film temperature sensor on thePSU's air heat sink - about where the blue mark is below...
...and a sticky silicone pad filling the gap between case and sensor on top of it to keep it attached and pressed against the heat sink at all times - see the sketch above.
The finished result looks like this...
... with the PWM fan control (the coloured cable) and the temperature sensor cable (black) being routed to the outside through one of the fan's mounting holes. The blue box in the top right corner above the power connector is the signal generator used to spoof the fan logic.
Last thing missing is attaching a temperature probe to the water return of the loop: Very similar to how I have attached the probe to the PSU's heat sink...
...I use a strip of thick adhesive silicone tape to stick the sensor directly onto the steel fitting, and secure everything with a cable tie to ensure there is always pressure on the sensor, keeping it firmly in place. NB that these sensors are fragile - the (transparent) silicone tape I use for padding is 4mm thick, reasonably soft, and distributes the pressure from the cable tie avoiding breaking it.
A bit of tidying up - and that's it!
Independently operated fans...
...with a software interface that allows me to employ different profiles for different tasks, and targeting heat sources directly.
Makes for pretty silent operations at light loads - and has fans kicking in properly when hitting it hard.
And, most importantly: No more thermal shut downs! And for those struggling with Supermicro temps: My thermal sensor data tells me that it trips at about 72 C measured at the top of the internal heat sink - may be a helpful data point.
While the work on the hardware is impressive (as always), I equally admire the temperature troubleshooting steps explained on these paragraphs.
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
- 724
- 838
The most insane feature is that you can control RGB based on temp sensors. In my watercooled rig I'm using it to make the FE waterblock terminals glow between green and yellow (up to red) depending on the water temperature.
Just came across your build - nice one! Did you get any thermal shutdowns with the A4x20?
Nope, I only got issues when I tried to pull way too much power on the CPU+GPU (around 600W with overclocking and stress-testing PSU+GPU), but I always thought the motherboard itself was the cause, the back of it was getting really hot. In gaming (with full GPU load and normal CPU load) I never had any sudden shutdowns.
So do we know what is generating the most heat on the PSU?
Once my new revision comes I'm going to open up the side of the PSU and have a 40*20mm fan to suck out more air.
Also I'm putting a graphite pyrolytic graphite sheet between the PSU and front panel and all metal to meal contact points to turn the PC case into a hot boi
Once my new revision comes I'm going to open up the side of the PSU and have a 40*20mm fan to suck out more air.
Also I'm putting a graphite pyrolytic graphite sheet between the PSU and front panel and all metal to meal contact points to turn the PC case into a hot boi
Every time I come back and look at this thread my mind is blown.
You could make a post tomorrow describing a plan to install dual GPUs and I wouldn't be shocked.
As for the PSU, maybe there is a more compact GaN unit that could be modified.
Or maybe you've already ruled that out and I didn't backread enough.
You could make a post tomorrow describing a plan to install dual GPUs and I wouldn't be shocked.
As for the PSU, maybe there is a more compact GaN unit that could be modified.
Or maybe you've already ruled that out and I didn't backread enough.
So i have learned how to design custom PCBs, the hard part is finding the correct components. I use JLCPCB for the manufacturing and EasyEDA for the designing, let me know if i can help.
Edit: also with the PCH fan and the A4x10 fan i haven't had a thermal shutdown yet. Only if there was a way to receive the temps that the PSU sees then i can run it through a RPI Pico and data log it. I wonder what else we can use an RPI Pico for
Edit 2: i totally forgot i had a second 800w PSU, let me know if you are interested in finding the PSU thermistor.
Edit 3: i think i found it, its green circle thing surrounded by some heat shrink next to the bridge rectifier heatsink and some inductors or whatever those coiils are called. i tested it and the hotter it is the lower the resistance ( 21 ohms ambient vs closer to 1 ohm with a mini torch. after cool down it is back to 21 ohms.) so its possible my heatsink mod helped bring
down the PSU temp/ reading.
here it is peaking out a little.
Edit: also with the PCH fan and the A4x10 fan i haven't had a thermal shutdown yet. Only if there was a way to receive the temps that the PSU sees then i can run it through a RPI Pico and data log it. I wonder what else we can use an RPI Pico for
Edit 2: i totally forgot i had a second 800w PSU, let me know if you are interested in finding the PSU thermistor.
Edit 3: i think i found it, its green circle thing surrounded by some heat shrink next to the bridge rectifier heatsink and some inductors or whatever those coiils are called. i tested it and the hotter it is the lower the resistance ( 21 ohms ambient vs closer to 1 ohm with a mini torch. after cool down it is back to 21 ohms.) so its possible my heatsink mod helped bring
down the PSU temp/ reading.
here it is peaking out a little.
Last edited:
For the 800w platinum,I added a 40x20 fan here ontop of the PCH fan near the connectors (i desoldered some Mobo connectors). I was going to do a dual 40mm slim fan on it but i messed up on my secondary PSU cover. i need another cover to attempt it again. I can probably fit another 40mm fan.
I have 35mm of clearance from the side of the 40mm fan to the PCH fan.
Yeah i don't even know if this is necessary for my setup since i never had a thermal shutdown before and after modifying the rectifier heatsink to a thicker copper one with only the PCH fan (from a gigabyte x570 Mobo)
Edit: yeah i have gone too far with this mod. I need to find a top cover for this PSU. Or let it free. ( Or create my own cover.)
Last edited:
Petricor, your stuff is just fantastic to read and I really appreciate you taking the time to document it so clearly for us all.
I am very interested in the idea of high efficiency builds at the moment, and have one on the go (A2000 into a S4T), and another planned. However, you have got me thinking about a sub 250w maximum power (for that wattage constraint) build in my S4T.
Here is what I am thinking: A4000 in a waterblock (140w max), 65w CPU with a Eisbaer pump block, LPX or VLP ram, the new 20mm (ST20) 140mm radiator from Alphacool, 25mm Noctua, 250 watt HDPlex GaN. I am not sure if it will fit, but it would generate less heat and may thus be able to run quieter. I will plan it out in Fusion and see what it looks like before I spend any money, as my wife may kill me otherwise.
I am very interested in the idea of high efficiency builds at the moment, and have one on the go (A2000 into a S4T), and another planned. However, you have got me thinking about a sub 250w maximum power (for that wattage constraint) build in my S4T.
Here is what I am thinking: A4000 in a waterblock (140w max), 65w CPU with a Eisbaer pump block, LPX or VLP ram, the new 20mm (ST20) 140mm radiator from Alphacool, 25mm Noctua, 250 watt HDPlex GaN. I am not sure if it will fit, but it would generate less heat and may thus be able to run quieter. I will plan it out in Fusion and see what it looks like before I spend any money, as my wife may kill me otherwise.
I mean S4M for this build. Obviously wouldn’t fit in the S4T
So rtx 4090 and ryzen 7000 is out, you probably know what Im thinking
I'm relieved to hear that Petricor and their 140mm radiator will heat all of Europe for the duration of the energy crisis.
Would you perhaps be able to share who you used to get these made. I downloaded the files and might try to get one made albeit with a couple modifications.
Similar threads
