CPU waterblock idea... / Computer Science / Forums

Forums

4hv.org :: Forums :: Computer Science

« Previous topic | Next topic »

CPU waterblock idea...

1 2 next

Move Thread

LAN_403

Fraggle

Sat Apr 20 2013, 11:43AM

Registered Member #1526 Joined: Mon Jun 09 2008, 12:56AM
Location: UK
Posts: 216

Hi guys, I started this thread on another forum I use and I was hoping for more input from you guys too:

Any more thoughts?

Pinky's Brain

Sat Apr 20 2013, 03:57PM

Registered Member #2901 Joined: Thu Jun 03 2010, 01:25PM
Location:
Posts: 837

Convection cooling with water is generally so efficient that anything further removed from the cooled surface than a few mm of metal is useless. So passing the wall of a heatpipe twice, going through the fins and then going into the water is just going to add a ton of thermal resistance.

You can do all the cooling near the processor, there is no need for more intermediate transfer of the heat than the standard water block provides.

Fraggle

Sat Apr 20 2013, 03:58PM

Registered Member #1526 Joined: Mon Jun 09 2008, 12:56AM
Location: UK
Posts: 216

Thanks for the input mate...

edit: Although, like I said in that thread - direct die doesn`t work so...at least a heatspreader DOES help...so, what`s going on here?

(I am a physicist so don`t worry about jargon)

Pinky's Brain

Sat Apr 20 2013, 04:18PM

Registered Member #2901 Joined: Thu Jun 03 2010, 01:25PM
Location:
Posts: 837

I think the problem with direct die is the flatness of the die, it's very hard to get a high enough flow near the die, it becomes insulated by laminar flow ... I'm sure with enough flow and turbulence direct die can work, although at some point the die won't be able to take the forces any more :)

I don't think heatspreader is exactly a good term for the base plate, it's there to conduct heat to the protrusions mostly (with most good high flow waterblocks, not true for stuff like serpentine channel blocks but those suck any way). Lateral spreading of heat will only occur over a couple of mm, water cooling of the protrusion provides most of the heat equalization over the surface of the waterblock. The parts of the waterblock more than a few mm removed from the actual die are entirely irrelevant to it's cooling ability.

Dr. Slack

Mon Apr 22 2013, 09:19PM

Registered Member #72 Joined: Thu Feb 09 2006, 08:29AM
Location: UK St. Albans
Posts: 1659

As you're a physicist ...

think 60*ln(a/b) type formula - you'll recognise this as the impedance of a coaxial line.

stay with me

Let's try to get the heat away from the hot spot on the CPU with copper.
Imagine a hemisphere of copper.
Say the CPU hotspot is 10mm radius, so 20mm across.
Let the radius of the copper hemisphere be x.
This object is spreading the heat from the 20mm area to the much larger curved surface of the hemisphere.

Now for simplification, imagine the CPU hotspot is not a plane, but a hemisphere of 10mm radius
(I know, we're straying into spherical cows and massless strings, but you're a self-confessed physicst)
The thermal resistance from the radius 10mm to the radius x will be of the form
resistance = k.ln(x/10)
As the CPU contact is planar, not spherical, there will be correction terms, especially for x approaching 10mm, but this is the assymptotic form.

In other words, the thermal resistance of the heat spreader increases monotonically with its outer radius.

This means that x should be minimised, subject to other constraints.

The other constraint is getting heat away from its outer surface. For heat transfer into a flowing liquid, the boundary layer resistance depends on the fluid thermal conductivity and thermal capacity, and volume of fluid in contact with the surface (yes volume, assuming the fins traverse the volume) and the number of fins. Obviously the number of fins can be made to increase without limit, which also increased the requirement for the fluid pump power to increase similarly. Limits to pump power and mechnical limits to fin spacing ultimately limits the thermal resistance. Microchannel plates use etching to get very tight fin spacing.

Anyhows, choose a fluid/surface geometry, and pump power, and determine the thermal resistance as a function of area, which will be 2pi.x^2. Then differentiate the total thermal resistance dx, and find the thickness of heat spreader that minimises the total.

Sounds easy. If only the choose geometry stuff didn't precede it

Fraggle

Tue Apr 23 2013, 10:24AM

Registered Member #1526 Joined: Mon Jun 09 2008, 12:56AM
Location: UK
Posts: 216

Good answers.

So, the problem really does boil down to the fact that a heatpipe is not homogeneous and can`t just be made any shape (ie, tapering to a huge area of fins) without an addition thermal interface TO those fins, whereas copper can...

Pinky's Brain

Tue Apr 23 2013, 05:48PM

Registered Member #2901 Joined: Thu Jun 03 2010, 01:25PM
Location:
Posts: 837

It's impossible to tell without lots of modeling ... but I think with ye average watercooling setup there is a good chance that simply having to pass the heatpipe walls twice is already enough to make a heat-pipe approach uncompetitive even if you assume the the pipe can be arbitrarily shaped and the thermal resistance other than the walls is 0.

I think the "problem" is that the power density is "too low" and the flow/pressure "too high" to bother trying to conduct the heat away more than a few mm before transferring it to the water ... and you don't want heatpipes for such short distances.

Steve Conner

Mon Apr 29 2013, 01:23PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Think about it this way: A heatpipe already contains one metal-to-liquid interface, as they're filled with alcohol or similar stuff. You're proposing to add a second metal-to-liquid interface in series with it. The performance will only be as good as the interface with the least surface area, and in your proposed system that is the one inside the heatpipes.

It follows from pure logic, no math required, that it would work better if you sawed off the fins and fed the cooling water straight through the heatpipes where the alcohol used to be, essentially converting the heat pipe cooler to a waterblock.

Put another way: A heatpipe might have better thermal conductivity than a solid copper rod of the same dimensions, but a pipe full of flowing water conducts heat better still.

Fraggle

Tue Apr 30 2013, 06:49PM

Registered Member #1526 Joined: Mon Jun 09 2008, 12:56AM
Location: UK
Posts: 216

but what I`m thinking is that it`s the interface to the water that`s the problem, that`s why the direct to die doesn`t work because below a certain limit of area, the water just won`t take up the heat fast enough. So then you increase the area as discussed until you reach the sweet spot where the water can pick up the energy fast enough, in return for moving the water further from the die. Once the energy is in the water it`s problem solved.

Steve Conner

Thu May 02 2013, 09:57AM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

No, I think the metal-to-alcohol interface inside the heat pipe will behave similarly to the metal-to-water interface, so making the metal-to-water interface bigger will just make the metal-to-alcohol one be the bottleneck.

1 2 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob