500 watt full featured gaming PC - passive cooled [Update!]

[Stefan1024]

The silent cube

Pushing passive cooling to the limit


After I could gain experience with my first passive build with a TDP of ~250 watts, I'm building an even more insane system. It will be the ultimate passive build because it's the best I can achive with the current technology without going absolutely crazy. And with crazy I'm mean something like using two PSUs, more than 50 kg heat sinks or something along thos lines.

The most powerful fanless PSU aviable nowadays is the Enermax Digifanless 550 watts, with a peak power of up to 605 watts (one minute). Under this power envelope I will fitt:

- CPU: i7-4790K (90 watts)
- GPU: 2x EVGA GTX 980 SC ACX 2.0 in SLI (380 watts)
- MoBo: Asrock Z97M OC Formula (15 watts)
- RAM: 16 GB DDR3 2400 (5 watts)
- SSD 1: Intel 750 PCI-E, 400 GB (25 watts)
- SSD 2: Samsung Evo 256 GB in M.2 format (5 watts)
- HDD 1: 2 TB HDD, TBD (10 watts)

The HDD can turned on and off with a switch and will only used for back-ups. I'd love to get rid of a HDD, but it's silly to make back-ups on a SSD and it will be hardly ever turned on.

As you can see the maximal power draw is around 530 watts and very close to the limit of 550 watts. But in normal operation it's very unlikely to push all components at the same time. For peak power draw I have a buffer of 75 watts, what I can feel comfortable with.
Thankfully the PSU has a very high build quality (it it also very expensive, but never cheap out on the PSU) and I can measure the load and temperature in real time.
EDIT: The measured peak load pulled from the PSU is about 510 watts.


Now I made sure the PSU don't blew up, I have to get rid of the heat somehow. The upper temperture limit for me is 80°C with a max. room temperature of 30°C. So I have a temperature difference of 50°C to play with. So it's time for number crunching and it turned out I need three heat sinks of 400x300x84 mm (12 kg each). So I designed the system around them.

1.jpg

The system building went on...

GPU4.jpg

12_7.jpg

... and is now nearly finished.

12_6.jpg

12_8.jpg

The temperature are also fine. During full stress test the CPU is around 85°C and the GPU's around 74°C (26°C ambient). During normal usage they are significantly cooler.

If you are interested in more informations go to http://linustechtips.com/main/topic/351 ... -pictures/ to view the full build log.

UPDATE:
I was able to borrow a 5000$ thermographic camera and confirm the build.

low_1.jpg

Look at the post at the end for more pics.

[quest_for_silence]

What's the role of the FX-70?

[Stefan1024]

What's the role of the FX-70?

I underestimated the transport capacity loss when installing heatpipes upside down. The three 8mm are not enougth to transfer the 100 watts to the big heatsink, despite they are capable of >200 watts when installed with the heat source down.

The FX-70 helps me to get rid of a couple of watts so I don't overpower the heat pipes and the CPU don't overheat.

[quest_for_silence]

Thanks for sharing your (stunning) experience.

[Stefan1024]

Thanks for sharing your (stunning) experience.

You are wellcome!

The system isn't finished by now, I will install some peltier modules to recover electrical energy using the Seebeck effect.

I appologize for not redoing the whole build log in this forum, but I can only upload <256kB pictures one by one. I need to rescale every image and you can't read the information of a full screen-shot anymore.

But nevertheless I will share the stress test results with you: Furmark and Prime95 (on 4 threads only so I don't disturb furmark) for 90 minutes

t11.jpg

t12.jpg

[Vicotnik]

Nice system.

PNG is the image format of choice for screenshots like that btw. Small file size and perfect quality.

[Stefan1024]

Nice system.

PNG is the image format of choice for screenshots like that btw. Small file size and perfect quality.

Thx

Well the 2560x1600 screen shot in PNG is 4 MByte. Downscaled to <256kBytes it's only 506x286.

t20.png

Am I doing something wrong with compressign the image?

[Vicotnik]

Well, a full screenshot with background etc (Furmark image) will be best as an JPEG, but something like the HWMonitor screen, with few colors, will be best as PNG.

You could put the relevant windows side to side, take a screenshot, crop the image to remove all the background, optimize the number of colors etc (optional really) and save as PNG.

[Stefan1024]

Well, a full screenshot with background etc (Furmark image) will be best as an JPEG, but something like the HWMonitor screen, with few colors, will be best as PNG.

You could put the relevant windows side to side, take a screenshot, crop the image to remove all the background, optimize the number of colors etc (optional really) and save as PNG.

Ok, good to know. I guess I'm to young (23) that I encountered exessive problemes with >1MByte pictures

EDIT: HWmonitor only, 80kByte (low usage / browsing usage temperatures)

Screenshot 2015-09-14 23.01.45.png

[Vicotnik]

Cropped it a bit, lowered colors to 256 and ran it though optipng. Got it down to ~30kB.

Sorry for all the OT and congratulations again on your very nice system.

[Stefan1024]

Cropped it a bit, lowered colors to 256 and ran it though optipng. Got it down to ~30kB.

Sorry for all the OT and congratulations again on your very nice system.

Haha yes with manually optimizing you can do much better than me with the lazy ALT + PrintScreen.
As I'm working a lot with embedded and ultra low power systems (but hardly ever with image processing yet) it's nice to see to what you can go down to without really loosing quality. Because every bit you have to transmitt costs you battery life.
So I don't mind

[pcgeek]

Wow, that's amazing dude! Very nice system!

[Stefan1024]

Wow, that's amazing dude! Very nice system!

Thank you!

[Stefan1024]

CPU VRM:

low_1.jpg

GPU VRM:

low_3.jpg

PSU:

low_4.jpg

The software from Enermax allows to read out the temperature of the PSU. But the sensor placement is a bad joke. It was showing 60°C while the PSU was actually running at 95°C. Full 35°C difference!?!

[quest_for_silence]

The software from Enermax allows to read out the temperature of the PSU. But the sensor placement is a bad joke. It was showing 60°C while the PSU was actually running at 95°C. Full 35°C difference!?!

Which FLIR camera is, and how did you calibrate it?

[Stefan1024]

The software from Enermax allows to read out the temperature of the PSU. But the sensor placement is a bad joke. It was showing 60°C while the PSU was actually running at 95°C. Full 35°C difference!?!

Which FLIR camera is, and how did you calibrate it?

It was an Optris Pi 450.
I didn't calibrate it by myself, but I borrowed it from the lab where it get's calibrated periodically. Also the temperatures measured on other parts of the system seemed reasonable, based on software readings from the components (except the PSU) and measurements takes with a temperature probe.
And I used E = 1.00. So it is likely the parts where even a few degrees hotter.

[Stefan1024]

So I finally finished my build:

SPCR_17_10.jpg

[CA_Steve]

So, it's about 40kg? I'm guessing the corner of the floor behind the PC won't get dusted often.

Cool concept.

[quest_for_silence]

Cool concept.

I would say "hot".

[Stefan1024]

So, it's about 40kg? I'm guessing the corner of the floor behind the PC won't get dusted often.

Cool concept.

There is a big enougth gab between the PC and the wall so I can (or at least could) clean the corner.
Also the total weigth is ~50kg.

But thanks for the kudo

[Stefan1024]

So today a woman of the government was here to measure the noise of the street in from of your flat. They had to collect data for some kinde of study and/or statistics.
The nice thing is that she brought a high quallity and well callibrated sound meter with here. And she was keen enougth to hand it to me for a few minutes.

I made three measurements each with the PC off and with Unigine Heaven runnign at ~100FPS. The noise floor of the device itself is 15 dBa.
The measurement distance was 60cm. I know the standart distance to measure is 1m, but I perfered to use the actuall distance to my head.

PC off: 20.9 dBa, 19.7 dBa, 19.8 dBa (mean value 20.1 dBa)
Heaven Benchmark: 21 dBa, 20.3 dBa, 20.2 dBa (mean value 20.5 dBa)

As you can see the noise floor in my room is pretty high. A good yound chamber is <14 dBa. So I was not able to measure the effective sound level of my PC because it hardly exceeds the noise floor of my room. But that was the goal of the project to have a PC that is a quiet as the room Also it is <21 dBa for sure.
But when a car drives by the sound level is about 33 dBa (with all windows closed)

Hopefully I can move to a quieter place sometime in the future. But the flaw here is nice and cheap. Just the street is very close.

[zhilin]

Hey man, great build!
I'm working on 0db PC myself and studied thermodynamics and heat sink optimization a lot, hope to post build log or gallery in here too. I have a straight out suggestion for you: find a way and remove every single rib in your profile - that would increase efficiency of the whole thing big time. These beasts were not designed for vertical convective cooling but for horizontal (ribs up), so in fact you're blocking the flow with that dense and package so the system works just by "brute force overkill".
Cheers and keep up the great work! (Passively cooled SLI, f**k me!)

[Stefan1024]

Hey man, great build!
I'm working on 0db PC myself and studied thermodynamics and heat sink optimization a lot, hope to post build log or gallery in here too. I have a straight out suggestion for you: find a way and remove every single rib in your profile - that would increase efficiency of the whole thing big time. These beasts were not designed for vertical convective cooling but for horizontal (ribs up), so in fact you're blocking the flow with that dense and package so the system works just by "brute force overkill".
Cheers and keep up the great work! (Passively cooled SLI, f**k me!)

Hi Zhilin

Sorry for late answer.
Don't get tricked by the picures. The heat sinks are optimized for natural convection. The finns are wide apart and textured fo have even more surface area.

profil.PNG

Also as far as I know vertical finns are more effective than horizontal once due to the chimney effect.