Silent power, a designed from scratch computer cooled with cooper foam

[wraper]

Talking about that particular computer implementation, not heatsink. Just noticed total bullshit. Do they intend to mount CPU and GPU on riser board, or some other mystical way above the main board where the RAM is located according to this picture.

[TMM]

A completely fanless desktop PC can be done.

Zalman have produced a few fanless cases in the past, e.g. the Zalman TNN-500. They use heatpipes to transfer the heat from the main components to the case which is basically a giant aluminium heatsink.

I doubt however that Zalmans cases from the past would be enough to cool a modern high end graphics card (e.g. GTX780). CPU's haven't really increased in thermal output so you could still run a top of the line CPU.

If you took the copper in a wire and fluffed it up so you had the SAME AMOUNT OF COPPER spread over a great cross section, then the conductivity would only be moderately worse. But that would obviously be an expensive and pointless operation since solid wire is rather easy to make and, say, twice as conductive as the foam. If the air around all those wires is actually needed for heat dissipation though, then it might just be a useful idea after all.

It's actually a terrible idea because with all that surface area the 'wire' (foam) would corrode in no time flat when placed in a moist environment.

[rob77]

Talking about that particular computer implementation, not heatsink. Just noticed total bullshit. Do they intend to mount CPU and GPU on riser board, or some other mystical way above the main board where the RAM is located according to this picture.

of course it's bullshit    they can't route DDR3 signals easily between boards (probably not even possible). they would have to place the RAM chips to the same board as the CPU or GPU.

i think it's just another SOAP like project the guys behind this "project" are too optimistic and they're suffering from lack of knowledge

[rs20]

But again heat exchanger != regular heatsink. There are very high pressure inside which pushes medium through it. They make it up to size: Duocel® copper foam is limited to 1" x 6" x 18" . As heat will come from both sides of the foam in heat exchangers, longest distance heat will need to travel through the foam is 1.25cm. So suppose in the heat exchangers surface area is more important than thermal conductivity. And this foam is much more dense than you would be able to use with simple fan and especially free air flow. If you think that such thermal conductivity is enough, it's not so simple. Even in regular heatsinks with plates which have better thermal conducticity than that foam, heatpipes are used (talking about PC heatsinks where you need to dissipate a lot of heat in restricted volume, not thick ones used in amplifiers, etc.). Just because if heat need to travel through the plate more than few centimeters, it becomes a problem as plate is hot in one end and cool in another. Yet even heatsink with such plate density as PC ones (= high surface area) are not any good for free convection.

Wow, look at all those numbers backing up your statements. I'm sure the heatsink would work better if it had iced water pumping through it, but that's not what we're doing here, so why are you mentioning it? I don't care if magical diamonds exist that defy the laws of thermodynamics, we don't have that here, so it's not relevant. If you have numbers to back up your claim that the air would have trouble escaping the foam (and I do suspect you might be right there, although it looks like quite a open, coarse foam) then I'd be interested to see that.  In any case, you were talking about the conductivity of the foam for ages, and now you've suddenly switched gear to the convection of the air. I've made you backpedal on the point that the conductivity of the foam itself is pretty close to good enough for the job, which is all I was ever claiming -- since everyone before in this thread just immediately derided any foam as "an excellent insulator". 

BTW, stock Intel heatsinks have no heatpipes inside them, and they work well enough (yes, with a fan), so heatpipes aren't relevant either.

Like I've said at the start, this is not an optimal design. But just because the design is bad doesn't make thinking it's bad for incorrect reasons (e.g. "copper foam is an excellent insulator") a good thing, or a thing not worth calling out.

[wraper]

wow, look at all those numbers backing up your statements. I'm sure the heatsink would work better if it had iced water pumping through it, but that's not what we're doing here, so why are you mentioning it? I don't care if magical diamonds exist that defy the laws of thermodynamics, we don't have that here, so it's not relevant. If you have numbers to back up your claim that the air would have trouble escaping the foam (and I do suspect you might be right there, although it looks like quite a open, coarse foam) then I'd be interested to see that.

Because if you want to directly compare regular heatsink with that foam while using regular fan, you would need to make that foam much thinner so the fan can do it's job. Reason for air resistance is because there is no direct path through it, like in the simple heatsink. http://en.wikipedia.org/wiki/Drag_(physics)
I never wanted to insult you BTW.

[wraper]

http://www.silentpcreview.com/intel-i7-hsf Intel do heat pipes actually. Their stock coolers for high end CPU's suck, but for lower end are usually pretty adequate last years. Thou intel do the trick with massive copper core in the middle, so they are not completely simple heatsinks.

[edavid]

they can't route DDR3 signals easily between boards (probably not even possible). they would have to place the RAM chips to the same board as the CPU or GPU.

I don't know about easily, but it's obvious that you can route DDR3 signals from a motherboard onto a DIMM.  CPU board -> riser -> RAM board should work too if the traces aren't too long.

[zapta]

http://www.silentpcreview.com/intel-i7-hsf Intel do heat pipes actually. Their stock coolers for high end CPU's suck, but for lower end are usually pretty adequate last years. Thou intel do the trick with massive copper core in the middle, so they are not completely simple heatsinks.

Great idea. It can function also as a solder tip cleaner.

[rob77]

they can't route DDR3 signals easily between boards (probably not even possible). they would have to place the RAM chips to the same board as the CPU or GPU.

I don't know about easily, but it's obvious that you can route DDR3 signals from a motherboard onto a DIMM.  CPU board -> riser -> RAM board should work too if the traces aren't too long.

actually i haven't seen DDR3 modules on a riser board it was common back in the times of slower RAMs, but haven't seen anything like that nowadays... especially when the RAM controllers are integrated in the CPU.
and not talking about the GPU's RAM - that's usually even faster (and even more "touchy" in terms of signal integrity).


EDIT: of course i mean DDR3 DIMM modules on a riser board.

[AndyC_772]

It might be just about OK if only one bank of DDR3 is to be supported. Against all odds, DIMMs work OK even with multiple DIMMs connected to each line, so the extra trace length and impedance discontinuity might be acceptable in a simpler system. The only way to be sure is to spend some time with a proper SI simulation package and an engineer who knows how to use an IBIS model properly.

[mzzj]

One thing that's not clear to me -- radiation and conduction to air/convection are both relevant to heatsinks, I wonder what the breakdown is for the typical heatsink? Radiation must be some part of the story if manufacturers bother to anodize the heatsinks black, but if it was all about radiation, why does the fan need to move the air around? Is the heat radiated at wavelengths easily absorbed by air? I think thinking this through would be relevant to this discussion as well.

Radiation effects are something like 5 to 40% of total heatflow depending on your case.

For example 50 mm piece of Fischer SK174 heatsink @ 300K ambient temperature(air and radiation temperature)  with 60K temperature rise:
specified thermal resistance ~6K/w
with a 60K temperature rise that works out to 12 watts or so.

Stefan-Bolzmann law gives us ~4,5W of radiation cooling assuming perfect 1.0 emissivity and  94cm2 outer surface area for the heat sink.
Radiation cooling has lesser effect on finned structures and with forced air cooling it has generally minimal net effect.

Under cloudless night sky the radiation cooling effect can account 100% of the cooling when heatsink temperature drops below ambient air temperature!

[rs20]

One thing that's not clear to me -- radiation and conduction to air/convection are both relevant to heatsinks, I wonder what the breakdown is for the typical heatsink? Radiation must be some part of the story if manufacturers bother to anodize the heatsinks black, but if it was all about radiation, why does the fan need to move the air around? Is the heat radiated at wavelengths easily absorbed by air? I think thinking this through would be relevant to this discussion as well.

Radiation effects are something like 5 to 40% of total heatflow depending on your case.

For example 50 mm piece of Fischer SK174 heatsink @ 300K ambient temperature(air and radiation temperature)  with 60K temperature rise:
specified thermal resistance ~6K/w
with a 60K temperature rise that works out to 12 watts or so.

Stefan-Bolzmann law gives us ~4,5W of radiation cooling assuming perfect 1.0 emissivity and  94cm2 outer surface area for the heat sink.
Radiation cooling has lesser effect on finned structures and with forced air cooling it has generally minimal net effect.

Under cloudless night sky the radiation cooling effect can account 100% of the cooling when heatsink temperature drops below ambient air temperature!

Interesting, thanks. Shows how insignificant emission is in many cases. It's worth noting that a finned heatsink may have a large surface area, but most of the emitted radiation will be re-absorbed by another fin of the heatsink! Perhaps you took this into account with your "outer surface area" (not sure if you were looking at the surfaces that are only facing free space). The other thing is that any surface is always absorbing radiant heat as well as emitting it, the Stefan-Bolzmann law only gives you one half of that equilibrium. For example, applying the Stefan-Bolzmann law to the human body gives an emission rate of 550W -- which would be rapidly lethal rate of temperature loss, except everything around is is at at least 273 kelvin as well so we're bathed in almost 550W of incoming radiant heat as well. So the net emission of our heatsink is ~4.5W minus incoming radiation. Again, I'm just clarifying, you did touch on this with your cloudless night sky comment.

[mzzj]

One thing that's not clear to me -- radiation and conduction to air/convection are both relevant to heatsinks, I wonder what the breakdown is for the typical heatsink? Radiation must be some part of the story if manufacturers bother to anodize the heatsinks black, but if it was all about radiation, why does the fan need to move the air around? Is the heat radiated at wavelengths easily absorbed by air? I think thinking this through would be relevant to this discussion as well.

Radiation effects are something like 5 to 40% of total heatflow depending on your case.

For example 50 mm piece of Fischer SK174 heatsink @ 300K ambient temperature(air and radiation temperature)  with 60K temperature rise:
specified thermal resistance ~6K/w
with a 60K temperature rise that works out to 12 watts or so.

Stefan-Bolzmann law gives us ~4,5W of radiation cooling assuming perfect 1.0 emissivity and  94cm2 outer surface area for the heat sink.
Radiation cooling has lesser effect on finned structures and with forced air cooling it has generally minimal net effect.

Under cloudless night sky the radiation cooling effect can account 100% of the cooling when heatsink temperature drops below ambient air temperature!

Interesting, thanks. Shows how insignificant emission is in many cases. It's worth noting that a finned heatsink may have a large surface area, but most of the emitted radiation will be re-absorbed by another fin of the heatsink! Perhaps you took this into account with your "outer surface area" (not sure if you were looking at the surfaces that are only facing free space). The other thing is that any surface is always absorbing radiant heat as well as emitting it, the Stefan-Bolzmann law only gives you one half of that equilibrium. For example, applying the Stefan-Bolzmann law to the human body gives an emission rate of 550W -- which would be rapidly lethal rate of temperature loss, except everything around is is at at least 273 kelvin as well so we're bathed in almost 550W of incoming radiant heat as well. So the net emission of our heatsink is ~4.5W minus incoming radiation. Again, I'm just clarifying, you did touch on this with your cloudless night sky comment.

yeah, calculated the surface area based on outer dimensions exactly for the reasons you said.
And yes, i did take incoming radiation also in consideration. (Thats why i mention 300K ambient radiation temperature.)

[T3sl4co1l]

Also, since radiation has a Tabs^4 dependence, you get vastly more thermal conductivity at elevated temperatures.  That is, if you want a lower thermal resistance due to radiation, you MUST run it HOT!

Vacuum is a rather awful conductor, which gives some clues about the design of vacuum tubes (back in the day).  The cathode in the center is hot, closely positioned near the grid, which must remain cool (otherwise it goes leaky).  The grid is usually molybdenum wire wound on copper supports, with radiator flags tacked on the top.  So the grid can stay somewhat near outer glass temperatures, being heatsunk by those flags.  The anode surrounds everything, and gets hot (it's where all the electrical power is going!), sometimes awfully hot (transmitter tubes made with graphite or tantalum anodes are intended to run red-hot!), hotter than the grid should be.  And around everything, there's a glass envelope, which typically runs over 200C at the hottest point -- so anything inside the vacuum is running nearly double room temperature (absolute scale), which helps greatly with radiation -- as long as the baseline temperatures are tolerable!

In vacuum lab equipment, you have to use non-outgassing PCBs (RF teflon stuff, Rogers whatever -- FR4 is no good), which conduct more poorly to begin with.  Thin copper traces don't account for much, so you can very easily have a jellybean op-amp dissipating not 100mW, yet burning up because there's no short path out!  Often, a huge copper braid is used, not so much for grounding as heatsinking: over such length, even though it's copper, the thermal resistance sucks, but vacuum sucks more (ha..), so it's a net win.

Lessons not all that applicable to consumery stuff (< 150C silicon, enough watts that you need a fan), but good to keep in mind all the same.

Tim

[tszaboo]

You actually have problem with the copper foam?
I7: 300 USD
GTX760: 250 USD
500GB SSD: 300 USD
8GB memory: 100 USD
Put it together, and it becomes 930 USD, what they are asking for. Without power supply, motherboard copper foil, case, and most important, the design.

[macgyver0815]

You actually have problem with the copper foam?
I7: 300 USD
GTX760: 250 USD
500GB SSD: 300 USD
8GB memory: 100 USD

Well these are consumer prices. First, if you buy these as a reseller you get volume discounts (the reseller has to make money from selling you stuff, often the resellers earn way more than the manufacturers).
Second, if you buy the components in >100pcs quantities and manufacture the cards yourself you can get it even cheaper (the manufacturer of the normal retail cards also have to make money from something).
Then these guys sell it directly, without resellers in between.

But I still expect very low margin as their production volume is not *that* high.

Oh and it is of course possible to put the DDR3 memory on a second PCB through a high speed connector. Not the best idea IMHO, but it can work (I have done DDR3 PCB layouts, so I know a bit about this stuff).
Also creating such a complex board with a small team is not impossible (several small industrial embedded PC vendors proof that). Very hard work, very hard to get right the first time, but very possible if you spent enough time.
The cost that the large manufacturers have here is paying several good engineers (say 100k$/a each) over several month fulltime work. This cost is 0 if you work for free.

That copper "foam" is actually a quite interesting material. But I also do not think that it will work @>100W without good airflow. I see the thermal design as the major technical issue of this project - the rest is not unrealistic (though I don't know if they actually have enough experience to do it properly...).

[Fsck]

you can find ddr3 on risers in high density multisocket servers, it's just annoying to do. also as macgyver0815 pointed out, not the easiest thing in the world. also if the nvm and ram are on the same board, you're probably going to be in a world of hurt when it comes to QC...

[rob77]

you can find ddr3 on risers in high density multisocket servers, it's just annoying to do. also as macgyver0815 pointed out, not the easiest thing in the world. also if the nvm and ram are on the same board, you're probably going to be in a world of hurt when it comes to QC...

actually i work with high density servers (HP blade servers) and all the DIMM sockets are packed as close to the CPU sockets as possible on the main board (and we have machines with half a terabyte of ram).

[corrado33]

Don't quite know if this was mentioned, but a heatsink needs proper contact with the hot device in order to drain the heat away.

I doubt this foam is making proper contact.

I mean hell, I've had computers overheat before because the heatsink wasn't connecting on a freaking corner, with 3/4 or it connecting just fine...

[T3sl4co1l]

Presumably there's a base plate it's soldered or brazed to...

With forced air cooling, and some solid fins to bring heat up into the center of the foam, it wouldn't be too bad.  But you're probably better off with the regular array of fins construction at that point, so who cares?

Tim

[Fsck]

you can find ddr3 on risers in high density multisocket servers, it's just annoying to do. also as macgyver0815 pointed out, not the easiest thing in the world. also if the nvm and ram are on the same board, you're probably going to be in a world of hurt when it comes to QC...

actually i work with high density servers (HP blade servers) and all the DIMM sockets are packed as close to the CPU sockets as possible on the main board (and we have machines with half a terabyte of ram).

sorry, I type before I think too often. forgot that blades even exist as they're usually low memory, but are definitely the highest density.
was thinking high end multisocket systems. the linked board will take 6TB of memory on risers over 4 sockets. the 2010-era nehalem-ex 8-sockets would take 1TB.

http://www.supermicro.com/products/motherboard/Xeon/C600/X10QBI.cfm

[rob77]

the linked board will take 6TB of memory on risers over 4 sockets. the 2010-era nehalem-ex 8-sockets would take 1TB.

http://www.supermicro.com/products/motherboard/Xeon/C600/X10QBI.cfm

but it looks like there is some kind of buffer/controller on those memory boards (the 2 heatsinks between the connector and the actual DIMM slots).. so it's not just routing the DDR3 signals to a different board.

[T3sl4co1l]

Hmm, hard to say.  They'll definitely want VREF drivers on the daughterboards, but they may well have more chips than that alone.  Wish the pictures had zoom.

I don't know what the fanout of SSTL_15 looks like, can't imagine it's too pretty after a hundred though.  Drivers per card array seem more likely than direct connecting everything.

Tim

[Fsck]

the linked board will take 6TB of memory on risers over 4 sockets. the 2010-era nehalem-ex 8-sockets would take 1TB.

http://www.supermicro.com/products/motherboard/Xeon/C600/X10QBI.cfm

but it looks like there is some kind of buffer/controller on those memory boards (the 2 heatsinks between the connector and the actual DIMM slots).. so it's not just routing the DDR3 signals to a different board.

correct. aka: annoying to do.