Disapating 750W of MOSFET heat for under $100

[thm_w]

The Rth of a common CPU cooler, costing about $20, is about 0.5°C/W. Let's say you want a max delta T = 50K (Between junction and air), this leads to 100W dissipated for $20, 750W for $150. That doesn't take into account all the custom hardware to mount your FETs, the assembly work, etc.

For this range of dissipated power, there's no any heatsink cheaper per watt dissipated than CPU coolers, they're manufactured by millions. Any solution with industrial parts will cost more than that.

Evo212 slightly better at 0.17°C/W for $30: https://www.anandtech.com/show/10500/stock-cooler-roundup-intel-amd-vs-evo-212/6
Wonder how a corsair H60 or similar compares, probably better with a high power fan, but I can't seem to find any specs.

[Mr.B]

The Corsair H60 is what I used in my design - see previous post.
I could not find any specs on it, so contacted Corsair directly.
Their response was that it was confidential information.

What I interpreted was "We have no idea"...

[Mechatrommer]

for $40-60 you can already get water based cooling, much powerful heat transfer and movable heat dissipation element. for this kind of power and if compactness/configurable is preferred, thats a wiser solution imho, ymmv.

The Corsair H60 is what I used in my design - see previous post.
I could not find any specs on it, so contacted Corsair directly.
Their response was that it was confidential information.
What I interpreted was "We have no idea"...

i have $40+ segotep halo blue china brand (cant find in ebay anymore) with 12x12cm coil/fan put on 95W cpu, no need big fan on cooling coil, everything runs cool to the touch with only slight air flow. i believe it can take 200-400W easily with more powerful fan. because it so affordable, i bought a spare. for something like this project. but for configuration in OP i think he will need more than one either way.

[Nerull]

A Corsair H115 can barely keep up with a 210W 9900k.

[37electrons]

Paralleling MOSFETs in linear mode is not a good idea. You'd need a separate gate control signal for each of them if you want true & equal load sharing.

[Mr.B]

A Corsair H115 can barely keep up with a 210W 9900k.

Because Corsair would not, or could not, supply me with any specs I had to make an educated guess at what the H60 was capable of.
Assuming it can be fitted to a reasonably grunty CPU, I figured it should be good for 200W.
I am using Noctua PPC NF-F12 PPC-3000 fans.
Up to 3000rpm - but bloody noisy at full speed.

[digsys]

Having made and sold many high power linear power supplies, using FETs in linear mode, I can add the following -
There is a LOT of good cautionary advice here ! Adding to them -
1/ DO NOT believe any of the datasheets. I had to destroy literally many dozens of FETs to make my own SOA curves. As posters note, even the linear specified
FETs are estimates / or plain bogus.
2/ De-rating is VERY important, both Temp and Voltage (current)
3/ Use a independent sense / share drive for each FET. There are ICs that do just that. Hot-spotting is bad enough, and you don't want one FET taking off.
4/ Select a proper thermal paste ! not just the crappy standard one people use. Arctic silver is a good start. If you have to isolate each FET, there are MUCH
better types available than the standard Silpad. This is very important.

There is at least one thread here going over all of this stuff. There's more to add, but I have to run :-)  .. bbl

[T3sl4co1l]

The better solution is to use resistors, which are smaller and cheaper than heatsinks for the same wattage.

You'd think so, but the exact opposite is true. MOSFETs rated for 200W can be found for about $1, but chassis mount resistors rated to 80W cost around $5.

Fortunately I wasn't talking about transistors.

How much will the heatsink cost?  $20 per 100W?  How much time will it take to cut to length (if applicable), drill and tap holes, grease the parts and bolt them in?

If you aren't having a Chinese worker do it en masse, the answer may be surprising.

Now, the resistors need to be mounted too, but that can be with much simpler means, like clips and self-tapping screws, or entirely with CNC'd custom sheetmetal.

If you're starting from an off-the-shelf box and cutting and drilling that, your labor costs will about wash out, and then other considerations may apply, like physical size.  (Resistors run hotter, so they can dissipate more heat for the same volume of air moved.  The air moves faster when it's hotter, too.)

Don't forget the knock-on costs, either.  Resistors probably need wiring.  You may be able to arrange a PCB to do it -- but mind issues of temperature, thermal expansion and other mechanical tolerances, and the cost tradeoff for a relatively large PCB, or a multi-board design.  Transistors need support components (at least a beefy source resistor; preferably an op-amp and support passives), and protection.

Tim

[nAyPDJ]

Evo212 slightly better at 0.17°C/W for $30: https://www.anandtech.com/show/10500/stock-cooler-roundup-intel-amd-vs-evo-212/6
Wonder how a corsair H60 or similar compares, probably better with a high power fan, but I can't seem to find any specs.

That's a list of stock coolers, which have a reputation for being crap.

AnandTech also reviewed some aftermarket air coolers, with 0.18°C/W being fairly typical

And they also reviewed some liquid coolers, and while the Corsair H60 isn't in the list, I'd expect it to be 0.08°C/W or so

"Disapating 750W of MOSFET heat for under $100"

The Rth of a common CPU cooler, costing about $20, is about 0.5°C/W. Let's say you want a max delta T = 50K (Between junction and air), this leads to 100W dissipated for $20, 750W for $150. That doesn't take into account all the custom hardware to mount your FETs, the assembly work, etc.

For this range of dissipated power, there's no any heatsink cheaper per watt dissipated than CPU coolers, they're manufactured by millions. Any solution with industrial parts will cost more than that.

Your numbers are a little off here. The cooler I've ordered (and not yet received) advertises 0.18°C/W & cost me $16 each ("bulk" discount--I bought two). You can't super trust the ratings, but it's about as hefty as my Wraith cooler in my current computer, and that's 0.18°C/W, so it seems about right. This comes out to 0.01125°C/W$, which is more than half of 0.025°C/W$

[magic]

The Corsair H60 is what I used in my design - see previous post.
I could not find any specs on it, so contacted Corsair directly.
Their response was that it was confidential information.

What I interpreted was "We have no idea"...

And I interpret "it's confidential". These aren't audiophools, they must know how their design performs

I suppose it's a matter of
- it depends on operating conditions and they don't want to publish a single number or a set of plots that Johny Overclocker wouldn't understand anyway
- they don't want to be beaten by other vendors using crooked methodologies when the game gets reduced to Johny Overclocker comparing one number printed on the box (see also: PMPO)

As is customary with computer hardware, you look for performance data in 3dr party reviews, which specify testing conditions and methodology. If your application is different than the reviewer's, you have to do the math yourself.

[Mechatrommer]

it should be around 120W ballpark official figure they knew that already. but with slight mod and working at higher than rated temp, i guess it can be made 200 - 400W, at limited time interval if not continuous, water is a great specific energy container. https://www.ebay.com/itm/ID-COOLING-ICEKIMO-7V-120W-Pure-White-AIO-Water-Cooler-CPU-Radiator-Cooling-Fan/113695211003?hash=item1a78c33dfb:g:QvUAAOSwLvxbq0mc

[nAyPDJ]

How much will the heatsink cost?  $20 per 100W?  How much time will it take to cut to length (if applicable), drill and tap holes, grease the parts and bolt them in?

If you aren't having a Chinese worker do it en masse, the answer may be surprising.

Plan is to use commercially available CPU cooling gear for the majority, which can be obtained at about 0.01125°C/W$ from china.

If this goes anywhere (big "if" here!), I'd expect this to go a little lower@100s. At that point, I'd expect it to few a couple extra dollars ea. to have a childadult worker fabricate the mounting brackets (Mr.B's clever design means that the heatsinking can be symmetrical & the clamps can be just tapped metal plates).

However, your point is taken, thank you.

Having made and sold many high power linear power supplies, using FETs in linear mode, I can add the following -
There is a LOT of good cautionary advice here ! Adding to them -

Thanks for the advice! It's great hearing from someone who has done this before.

There is at least one thread here going over all of this stuff.

I think this is it, I'll just link it here for anyone following along: Programmable Electronic Load, 0-5A.

The discussion is mostly centered upon making it high precision, but OP is also working on (and having difficulty) paralleling the mosfets. I might steal some parts of his control design so that I can focus on the heat dissipation for now

The advantage that dropping power across a dumb resistor is of course the fact that the resistor, being a dumb passive element, can run at very elevated temperatures without failure. 

For example, here is a 50 watt "resistor" that operates at something like 2,500 degrees centigrade

I really like the idea of dissipating my heat using exposed 500°C wires, but while the electrical design is more straightforward, the mechanical design is quite tricky. Roughly 300cm of 22AWG nichrome wire is needed to get 10ohm.

- the wire needs to be mounted on something non-conductive & conveniently shaped. An 80mm ceramic square would be perfect, but it needs to be fabricated & I can't even find prices for ceramic 3-d printing.
- the copper wires need to be bonded to the nichrome in a way that doesn't fail at high temperature. A bolt & nut would work, but then where would the bolt and nut be mounted?

Light bulbs wouldn't work since their resistance goes from ~16ohms cold, to hundreds of ohms hot. I haven't done the math exactly, but I figure I'd need hundreds of bulbs as an order-of-magnitude approximation.

[digsys]

This post shows a method I use to have several FETs track almost perfectly. 2x is shown, but easy to add more.
https://www.eevblog.com/forum/beginners/linear-psu-with-modern-components-will-it-matter/msg2041252/#msg2041252

Added: Also discussion here -
https://www.eevblog.com/forum/beginners/linear-mode-operation-and-mosfet-fbsoa-question/msg1235696/#msg1235696
Maybe not specific to your question, but helpful info

Added: more discussion of issues
https://www.eevblog.com/forum/beginners/question-on-mosfet-in-linear-mode-does-_hot-spotting_-a-serious-problem/msg99612/#msg99612