passive cooling [Finished] The number cruncher: Triple Xeon passive mineral oil cooling [Update 12: Final pictures and summary]

[nmil]

1 hour ago, thekeemo said:

That should make it about equal and with superior heatpipes it should make it even better (to the point of filling it with mineral oil?)

I don't think heat pipes alone can beat out a full tank of oil. the pipes only contact the main heat generating areas of the components, but the oil will contact everything. Plus the vast amount of oil also gives you a tonne of thermal capacity, to get the same in a heat pipe you need a significant amount of speed which I don't think is possible via convection alone.

[Stefan1024]

The project is still alive and some finished touches are needed.
 However, my timeto work at the PC is basicly non existent at the moment so you need to wait a little longer, sorry.

[Stefan1024]

I finished the build today!

 

I have no time for a proper blog entry by now but here you have a sneak peak:

 

[pagani123]

awesome!!!

[Trey222]

Ooohh that looks so nice. 

[Stefan1024]

Update 12: Finally finished!!!

 

So this week I got the glas cover and finished my build over halfe a year after I started building it. And I have to say I really like the outcome

Have a look by yourself:

 

 

 

 

 

 

 

 

 

As the build is running oil cooled for almost two month now I can already give you a summary:

 

The good....

In the beginning I was not sure if I'm able to master this risky project. Despite a long planing phase, there was a high probabillity ending up with a design flaw and it wouldn't work at all. But the cooling works well, on idle the CPUs are at about 42 C and the GPU at 33C. Now I'm folding at full power on CPU and GPU, both overclocked (CPU 3,1 GHz all cores, GPU 1.277 GHz) and the CPUs are at 64C and the GPU at 73C. I'm only folding since 45 minutes so the temperature will rise a bit more, but that's fine, considering I have the max. possible load and burning over 600 watts.... Also, passive cooled and overclocked Titan X (Maxwell)!

Now I have a quite powerful PC that I don't need to upgrade for several years. And by the way, when I will upgrade I have to dissassembe the entire PC because I can't reach several cables. But I will worry about that in 4 years or so

 

... the bad....

There are a few things that didn't turn out like they where intendet. From an aestetic point of view I should have switch the position of the Xeon Phi and the GPU so the DP cables would be on the side and not in front. I assumed there must be a 90 degree DP cable somewhere, but didn't finde any, even in the Shenzhen electronic market, and they have about 3 floors with only display cables. But the PCIe slot allocation is better this way, that's why I choosed this placement in the first way.

Speaking of the Xeon Phi, I'm able to use the auto-offload feature and speed up calculations when using MATLAB, but only when the jobs are big enougth. As I usually have not very high maxtrix / array sizes and low arithmetic desity, hardly any calculation gets send to the Phi. On the other hand the two E5-2680 are already crushing my old machine (i7-4790). Everytime I boot the PC, the Phi will boot as well and stay in the P0 power mode (about 200 watts!) and I have to manually connect to it so it will go dwon to P3. But even then it still burns about 70 watts while waiting for work. This is a lot of energy and I decided to unplug the PCIe cable to turn the card completly off.

When I have a big project in the future I will reconnect it (takes be about 5 minutes + 5 minutes to clean my hands).

Also the adhesive of the cable ties got disolved in the oil and they get lose. The once with screwes are much better, but I liked to drill as less holes througth the heat sinks as possible. By the way, the rubber dampenings on the noctua fans are completly fine.

 

... and the ugly.

I used to much thermal paste. Usually this is not a problem as I use nonconductive one. But not the oil dissolved it and it felt doen onto the RAM slots.

 

I tried to clean it, but it just desolved into the oil even more and make it clowdy. I hope it will fall out over time and accumulate on the bottom where it is invisible.

But my worst mistake was to use silicon sealing. As Christoph told me, it is hydrophobical but lipophilic. So it will seal good for water but bad for oil. The oil can go througth even the most narrow gap and even diffuses through solid silicone. As a result my build is leaking a little bit. It is not much at all and all gets captured by the catch basin, but still. After one month the oil level in the basin was about 1 mm. I can handle it as I have enougth oil, so I will not disasseble everything to fix it. If you work on a similar project keep this detail in mind.

 

But in the end I'm really pleased with the outcome.

[Stefan1024]

On 2016-09-22 at 0:44 AM, Maybach123 said:

awesome!!!

 

On 2016-09-22 at 9:01 PM, Trey222 said:

Ooohh that looks so nice. 

Thank you very much

[Heesleemer]

So basically we won't see another passive pc from you for another 4 years when you need upgrading =P

[MoonlightSylv]

6 hours ago, Stefan1024 said:

-snip-

That has to be one of the coolest PCs I have ever seen, nice work!

[Space Reptile]

On 2.2.2016 at 8:58 PM, Stefan1024 said:

Xeon Phi 31S1P

 57 cores?  

 

[Energycore]

4 minutes ago, Space Reptile said:

 57 cores?  

 

They're Atom cores, not particularly great

[Space Reptile]

39 minutes ago, Energycore said:

They're Atom cores, not particularly great

im running an FX , im used to it  

[Energycore]

2 minutes ago, Space Reptile said:

im running an FX , im used to it  

You have no idea how bad Atom Cores are. Not even an FX compares.

[Space Reptile]

Just now, Energycore said:

You have no idea how bad Atom Cores are. Not even an FX compares.

oh i do , i ran an netbook that had a single core atom on it , running XP , im just having a laugh , since Phi´s are co-cpus . wich are more like network resource than anything 

[Energycore]

1 minute ago, Space Reptile said:

oh i do , i ran an netbook that had a single core atom on it , running XP , im just having a laugh , since Phi´s are co-cpus . wich are more like network resource than anything 

My mother has an Atom powered laptop. I keep telling her we can upgrade for cheap, but she says it's "fine".

 

In my experience working with that computer, performance is not "fine".

[Space Reptile]

2 minutes ago, Energycore said:

My mother has an Atom powered laptop. I keep telling her we can upgrade for cheap, but she says it's "fine".

 

In my experience working with that computer, performance is not "fine".

it was enough for 480p youtube

[Energycore]

Just now, Space Reptile said:

it was enough for 480p youtube

I bet it was

 

But jesus, it was wholly incapable of running office and chrome at the same time.

[comicsansms]

30 minutes ago, Energycore said:

I bet it was

 

But jesus, it was wholly incapable of running office and chrome at the same time.

My $1,000 netbook is going to become my remote desktop computer so I can browse reddit at school. It is very useless, it compares to a P4 uselessness.

[Mr.Meerkat]

7 hours ago, Energycore said:

My mother has an Atom powered laptop. I keep telling her we can upgrade for cheap, but she says it's "fine".

 

In my experience working with that computer, performance is not "fine".

Single core Atoms were complete and utter shit but the dual core atoms back then were usable for word and web browsing...

 

On the other hand baytrail CPUs are called Atoms and come on, they are fine for even 720/1080p streaming  

[Stefan1024]

14 hours ago, Heesleemer said:

So basically we won't see another passive pc from you for another 4 years when you need upgrading =P

Unless I sell this one or finde somebody who pays me to build one, no I won't build an other one. No money for constantly building a new PC

12 hours ago, ShadowTechXTS said:

That has to be one of the coolest PCs I have ever seen, nice work!

Thank you!

12 hours ago, Energycore said:

They're Atom cores, not particularly great

For many core computing they are fine as you only look at performance / watt. Single core performance is not important.

[Rodniz]

I love it. I wish i had the skills, or time, or money to build something like this

 

[feha]

Although I doubt I have the time and energy to actually make passive coolers myself, I still got a tad curious about some things when reading about your builds.

 

How do you know what size of the heat-sinks are sufficient? For example your Silent Box had this paragraph

Quote

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 about 80°C with a max. room temperature of 30°C. As a result 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.

This makes me sort of confused as I am pretty sure the TDP of the components you cool should matter (otherwise those heat sinks would suffice no matter what you want to cool, as long as the delta-temperature is ~50 c). Either way, I am not quite managing to find any calculations for this and it seems to me as if some heat sinks with same size differ in tdp they are supposed to be able to handle?

 

How much margin should one have when selecting heatsinks? For example, consider the i7 6700k. Tdp of 91w, but as it is unlocked one can expect some overclocking and an article I read suggest 110w as around the highest tdp one can expect. What size would a heatsink cooling that have to be (assume same 50 c temp-diff)?

 

And a question about heatsinks in general, is there any indication of how fast they spread the heat along themselves? As in, say one just pastes a cpu or gpu onto a regular heatsink, at around what tdp would the warm thing exceed, say, 100 c simply because heat sink fails to spread the heat through itself (I dont know what materials are actually used, but for example copper should have a limit to how fast it spreads the heat no matter the size of heatsink if the object generates enough heat. Related to this, what materials are commonly used?)?

If one reaches such a generation of heat, to what extent would it be possible to improve things by using heat-tubes connected to the cpu and spread over the entire surface of the heatsink?

[Stefan1024]

On 2016-09-26 at 1:44 AM, feha said:

- snip -

Sorry for late answer, I was busy.

 

The most important thing is that a heatsink can't dissipiate x watt, but a thermal resistance, for example 0.1 K/W. This means when it has to dissipiate 200 watts, it needs to heat um 20 degrees C to do so. For a GPU you overall thermal resistance needs to be below 50C (or K, it's the same delta) / 250 watt = 0.2 K/W.

But keep in minde this is the overall resistance including the coppling from the chip to the heatsink. In my systems this is about 0.1 K/W with direct coppling and even a bit more then I use heat pipes.

So I need a heat sink of about 0.1 K/W that is an impressive performance for a passive heatsink. And then I looked for suppliers of heat sinks and they measured this parameter.

 

Add a buffer as big as you can afford,  > 30% at least. And with OC you can double or even triple the heat output of a CPU. But passive cooling is not design for high end OC.

 

The heatsinks I buy also have a thick baseplate (for the silent cube 15 mm!) of pure aluminium to spead the heat. And remember ther thermal resistance from the aluminium to the air is much bigger than the one of the aluminium itself. So the heat distribution within the heat sink is fine. Still I used 5mm of copper for spreading the heat from the chip. But I also need the material to bridge the gap because of the big components on the PCB.

The limitting factor is the transfer of the heat from the heat sink to the air. The heat distribution is quite equal once the thermal capacity is used up.

 

[Stefan1024]

On request from @Bananasplit_00 some more information about the GPU location:

 

It is attached to the front heat sink and on the opposite side of the MoBo. The two big cabled in the front are the DP cables.

 

 

 

As it is sandwiched between two heat sinks one can't see it annymore.

 

 

By the way, the system is still up an running. The oil leakage is not increasing and stays on a very low level.

[Bananasplit_00]

13 minutes ago, Stefan1024 said:

-snip-

ok, thanks! copleatly clears that up