News Core i9 14900KS heatspreader transformed into CPU water block - clever machining yields a functioning water block

[Admin]

A Chinese modder has machined their CPU's IHS (Integrated Heat Spreader) into a functional water-block for liquid cooling.

Core i9 14900KS heatspreader transformed into CPU water block - clever machining yields a functioning water block : Read more

 

[A Stoner]

You would want the water to enter at the center of the die and work its way outward where it then discharges.

 

[bit_user]

An obvious problem I can see with the concept is that removing material from the IHS impairs its ability to distribute the heat. I'd expect an additive approach to work better, where channels are built up atop the IHS. Or, at the very least, don't remove (much) material right above the die, itself.

Also, I don't know much about water block design, but the path cut for those channels seems non-optimal.

 

[bit_user]

You would want the water to enter at the center of the die and work its way outward where it then discharges.

I think that's not what waterblocks typically do and I think I understand why. If you recall Newton's Law of Cooling, the rate of heat transfer is proportional to the temperature difference. If the center of the IHS is at 70 degrees and the edges are at 55 degrees, then having water hit the center first will quickly warm up the water (let's say to 50 degrees), which is almost the same as the edges. At this point, contact with the edges of the IHS won't extract much additional heat from it.

On the other hand, if you pump in the water at the edge and let it work its way up to the center, then it's extracting a more constant amount of heat, the whole way. So, maybe it gets to 40 degrees at the edge and reaches 45 degrees by the center. At this point, the delta is still enough for good heat extraction, but it's already taken heat from the edge.

Which scheme is better probably depends on a lot of specifics and would require proper modeling to know for sure.

 

[A Stoner]

I think that's not what waterblocks typically do and I think I understand why. If you recall Newton's Law of Cooling, the rate of heat transfer is proportional to the temperature difference. If the center of the IHS is at 70 degrees and the edges are at 55 degrees, then having water hit the center first will quickly warm up the water (let's say to 50 degrees), which is almost the same as the edges. At this point, contact with the edges of the IHS won't extract much additional heat from it.

On the other hand, if you pump in the water at the edge and let it work its way up to the center, then it's extracting a more constant amount of heat, the whole way. So, maybe it gets to 40 degrees at the edge and reaches 45 degrees by the center. At this point, the delta is still enough for good heat extraction, but it's already taken heat from the edge.

Which scheme is better probably depends on a lot of specifics and would require proper modeling to know for sure.

I do not think the water warms up anywhere near as fast as you think it does. And, once the system has been running for a while, the water temperature should settle at a near specific delta over room temperature. Typically about 8 to 10 degrees depending on volume of air circulation and whether or not the whole gets any fresh air. Basically, if your case is not expelling the heat out, the ambient temperature inside the case will be higher and will increase whole system temperature accordingly.

If your room temperature is typical 22C/72F then you can expect the water temperature in a well sized cooling loop to be about 32C and even under load that should remain constant. A typical computer system will pump 0.04 liters of water per second. A CPU putting out an exceptional 320 watts would warm that water by 2C as it passes over. Not really a huge delta from internal to external.

The reason to have the water coming in at the center is not to use the delta, but the force of the flow coming in will reduce the amount of 'SKIN' that is present at the hot spots. As liquid flows through channels, the liquid in the center of the channel flows the fastest and that at the boundaries of the channel flow the slowest. Slower moving fluid is less capable of absorbing thermal energy and moving it away from the hot spots. So, you want the most turbulent flow in the heat spreader to be happening at the hottest spots.

 

[ElMoIsEviL]

An obvious problem I can see with the concept is that removing material from the IHS impairs its ability to distribute the heat. I'd expect an additive approach to work better, where channels are built up atop the IHS. Or, at the very least, don't remove (much) material right above the die, itself.

Also, I don't know much about water block design, but the path cut for those channels seems non-optimal.

The water replaces the copper material that is removed. The surface area thus remains the same.

I think that's not what waterblocks typically do and I think I understand why. If you recall Newton's Law of Cooling, the rate of heat transfer is proportional to the temperature difference. If the center of the IHS is at 70 degrees and the edges are at 55 degrees, then having water hit the center first will quickly warm up the water (let's say to 50 degrees), which is almost the same as the edges. At this point, contact with the edges of the IHS won't extract much additional heat from it.

On the other hand, if you pump in the water at the edge and let it work its way up to the center, then it's extracting a more constant amount of heat, the whole way. So, maybe it gets to 40 degrees at the edge and reaches 45 degrees by the center. At this point, the delta is still enough for good heat extraction, but it's already taken heat from the edge.

Which scheme is better probably depends on a lot of specifics and would require proper modeling to know for sure.

High performance waterblocks do target the hot spot area as water atoms colliding with the copper will absorb more heat due to the vortex created. This is a lesson waterblock engineers learned a long time ago. DangerDen started with their Maze series, which is what this waterblock here does, and where beaten soudly by the Storm series of blocks which injected the water right at the center of the block with a restriction plate causing the water to have the same effect as placing your thumb at the end of a water hose to cause restriction and a strong stream of water. The temperature difference of the water is not relevant because it's a closed loop system. The entire water in the loop reaches an equilibrium.

 

[Pierce2623]

This is a really cool idea but I think he should’ve left more material and gone from the center out just based on what I know of chip cooling best principles. That being said, as far as this is from my true “wheelhouse”, even though I have an engineering degree, I’m probably wrong lol.

 

[bit_user]

The water replaces the copper material that is removed. The surface area thus remains the same.

That's not how it works. Water isn't nearly as heat-conductive as copper. I'm reading it's like 0.6 W/mK, whereas copper is about 400 W/mK. So, when you remove copper from the IHS, you're impeding its ability to spread out that heat.

The idea of a heat spreader or a water block is to spread the heat over the greatest surface area, so that heat can transfer more efficiently into the water.

 

[YSCCC]

This looks like a fun project. But since it tis warranty void I woud thinkg it's better just to delid and get what better optimized jet channels a direct die water block can do. at leaset you don't try drill channels into the IHS which can accidentally cut into the die anyway

 

[ezst036]

I like this idea.

 

[ElMoIsEviL]

That's not how it works. Water isn't nearly as heat-conductive as copper. I'm reading it's like 0.6 W/mK, whereas copper is about 400 W/mK. So, when you remove copper from the IHS, you're impeding its ability to spread out that heat.

The idea of a heat spreader or a water block is to spread the heat over the greatest surface area, so that heat can transfer more efficiently into the water.

No you're not. You are removing the heat with the water. Having a thick IHS doesn't help if you have no mechanism to remove the heat. It may take longer for a thicker piece of metal to become saturated with heat but it will nonetheless become saturated with heat. The water removes the heat. Your calculation doesn't take into consideration the constant flow of water. You're also comparing Copper to the Water when you should be comparing Water to Air.

If what you say is true, then "delidding" (removing the IHS) wouldn't lead to better cooling performance. In fact direct water contact, if such a thing was possible, with the CPU die would cool better than having a copper medium in between. In the waterblock, you generally want a "heatsink" like set of fins over the area sitting atop the CPU die. This improves the surface area of the metal making contact with the water. Then you want a restrictive mechanism (Jet impingement) to constrict the flow of water over those fins to spread the water over the fin area in a more even fashion. Water takes the heat away from the CPU die and dumps it into the radiator. The constant flow of water is key.

This is how modern waterblocks are designed. The "Maze" design, by this user, is ancient and from the early days of watercooling.

 

[bit_user]

No you're not. You are removing the heat with the water. Having a thick IHS doesn't help if you have no mechanism to remove the heat.

So, let me ask you this: why aren't waterblocks smaller? They're as big as they are, and have fancy channels, because more surface area is needed to effectively transfer heat into the water.

The water removes the heat.

By coming into contact with the heated surface.

Your calculation doesn't take into consideration the constant flow of water.

Sure it does.

You're also comparing Copper to the Water when you should be comparing Water to Air.

That makes zero sense. The IHS is not designed for heat transfer directly to air.

If what you say is true, then "delidding" (removing the IHS) wouldn't lead to better cooling performance.

The point of mounting a water block directly on a bare die is that waterblocks are already copper, so having a copper IHS and then a copper waterblock is fairly redundant. The point of this exercise is to use the IHS as the waterblock, but there's still a waterblock.

In fact direct water contact, if such a thing was possible, with the CPU die would cool better than having a copper medium in between.

If that were true, then waterblocks would be smaller and have a thinner base.

In the waterblock, you generally want a "heatsink" like set of fins over the area sitting atop the CPU die.

The reason for that is because those fins have higher heat conductivity than water. If not, they'd be pointless.

 

[Notton]

That looks like a fun project.
It seems like the only advantage is the low profile. The channels aren't designed for turbulent flow which increases heat transfer rate.
https://news.mit.edu/2020/how-fluids-heat-cool-surfaces-0428

 

[PeterStark]

I think that's not what waterblocks typically do and I think I understand why. If you recall Newton's Law of Cooling, the rate of heat transfer is proportional to the temperature difference. If the center of the IHS is at 70 degrees and the edges are at 55 degrees, then having water hit the center first will quickly warm up the water (let's say to 50 degrees), which is almost the same as the edges. At this point, contact with the edges of the IHS won't extract much additional heat from it.

On the other hand, if you pump in the water at the edge and let it work its way up to the center, then it's extracting a more constant amount of heat, the whole way. So, maybe it gets to 40 degrees at the edge and reaches 45 degrees by the center. At this point, the delta is still enough for good heat extraction, but it's already taken heat from the edge.

Which scheme is better probably depends on a lot of specifics and would require proper modeling to know for sure.

It generally is configured like that.

Inlet of coolest to core locations, hotspots, then it circles around the edge and out.

Point is that manufacturer needs to perform thermal modelling and figure out the best flow for specific chip. Did you know that on GPUs we mostly direct-die cool?

Assumption is that if water enters at say 35°C (system alresdy under load), we want to remove most difficult heat first, to bring down the average temp, to spread the heat out a bit and quickly extract to rad.

Water cooling 4090 or 5090 over say 3-4hrs and you notice that memory controller needs cooling but nowhere near as much as chip itself (difference between 10s of W to 100s of W/hr). E.g. water at 50C still cools memory controller enough.

Issue with design above is matter of flowrate and dissipation. It helps to add chunky bit of say copper to provide for more of thermal buffer to slow down peaks.

Intriguing experiment none theless, we would not have known if it is good or not without trying. Thanks to brave machinist.

 

[Krieger-San]

An obvious problem I can see with the concept is that removing material from the IHS impairs its ability to distribute the heat. I'd expect an additive approach to work better, where channels are built up atop the IHS. Or, at the very least, don't remove (much) material right above the die, itself.

You are correct - I have experienced this phenomenon with thermal dynamics before working with different thickness heat spreaders integrated into water blocks. The thinner the material of the heat-spreader in the water block, the higher the likelihood of thermal runaway. Also, block material is CRITICAL. See comment below as to why:

That's not how it works. Water isn't nearly as heat-conductive as copper. I'm reading it's like 0.6 W/mK, whereas copper is about 400 W/mK. So, when you remove copper from the IHS, you're impeding its ability to spread out that heat.

This right here.
In my experience tinkering with an ancient GTX 680 with an old CPU waterblock, I needed to shim the block so that it would clear some of the board components.
I tried using aluminum (1/4" plate, laser cut to fit), but this caused some thermal spiking under load.
When switching to copper, the spikes were completely eliminated. Bigger copper plate helped smooth temps a little more with running extreme tests like Furmark.

Also, I tried creating a thermo-electric cooler to cool water (water block to water block, two water loops, TEC between the blocks) doesn't work unless you have a decent absorption contact for the heat sources (ex.: a copper shim that's at least a few millimetres thick). If the block doesn't have the thermal capacity (ex.: low W/mk), it simply cannot extract the excess heat from the source. It actually work better to cool the liquid of one loop directly using a heatsink or radiator than another loop that has a radiator, etc.

 

[bit_user]

Thanks for the info!

Also, I tried creating a thermo-electric cooler to cool water (water block to water block, two water loops, TEC between the blocks) doesn't work unless you have a decent absorption contact for the heat sources (ex.: a copper shim that's at least a few millimetres thick).

I had an idea that maybe TEC could be integrated into the secondary half of a radiator, in order to chill the water temperature to near or below ambient. I guess, if you went below ambient, you'd run a risk of condensation, so it should be controlled by a thermostat or both a thermostat and hygrometer.

 

[georgebaker437]

Some people ask why, others ask why the hell! There is a huge difference between "can" and "should".

 

[jonaswox]

Thanks for the info!



I had an idea that maybe TEC could be integrated into the secondary half of a radiator, in order to chill the water temperature to near or below ambient. I guess, if you went below ambient, you'd run a risk of condensation, so it should be controlled by a thermostat or both a thermostat and hygrometer.

Water evaporation is the answer my friend. Nobody wants a cooler that uses even more power.
Have some evaporation going on however you want, and cool your water pipes with that energy sink. Its basically free and cool addon is that you can legitemaly say that your CPU is Einstein cooled 😉(this is basically the principle of the first conceived refrigerator,thought up by none less than Einstein himself.)

 

[bit_user]

Water evaporation is the answer my friend. Nobody wants a cooler that uses even more power.

Yeah, but I also don't want my room being incredibly humid. Plus, it seems like you'd need to use distilled water, if you didn't want a lot of buildup to accumulate on the evaporator.

As for the energy thing, yes. That's a fair point and it's actually why I'd use TEC on the second half of the radiator. When the water first hits the radiator, that's when it's hottest and easiest to cool. In the latter part is where the TEC could add a bit more temperature delta to help draw even more heat out of the water. You probably wouldn't need much.

 

[jonaswox]

Yeah, but I also don't want my room being incredibly humid. Plus, it seems like you'd need to use distilled water, if you didn't want a lot of buildup to accumulate on the evaporator.

As for the energy thing, yes. That's a fair point and it's actually why I'd use TEC on the second half of the radiator. When the water first hits the radiator, that's when it's hottest and easiest to cool. In the latter part is where the TEC could add a bit more temperature delta to help draw even more heat out of the water. You probably wouldn't need much.

I did assume that you would either capture the vapor and revaporise it , or simply doing the evaporation on the roof for example. I would not personally mix the vaporation system liquid with the pc cooling liquid. Doing indoor evaporation into the air is impractical unless you live in amazingly dry conditions in which this solution would perform splendid. I personally would run a pipe to a structure similar to a roof solar cell, where I would evaporate water with the sun + breezing air - and capture the cold with my pc water. If you use copper pipes in the evaporator for both flow lines, and use a good thermal interface in between you will have amazing conductance of heat - and the supplied heat will even help the evaporator evaporate even more. So like a heatpipe we are actively using the heat to improve our cooling system and actively remove heat from the system with the energy itself. A heatpipe goes insane because it can run completely on its own - but all we need to expand is a simple water pump akin to the ones used in floor heating. As long as you can keep your evaporator moizt , you will have strong performance. And who does not love any kind of solution that performs best when moist? 😉

The evaporator itself can just be a black sponge exposed to sun and fresh air. If you have any sort of supply of dry air , this is like golden jetfuel for this system - you can accelerate osmosis massively. I would love to personally test accelerating air tunnels in such an evaporator system since have the chill factor of moving air. So it would be interesting to see how much you can improve the performance from a strong ventilation system capable of pushing the air through tight nozzles that forces the air to speed up. Maybe it would just be wasteful because now your ventilation has to work against a big resistance. But it is the same principle that keeps planes in the air and formula 1 cars on track , so it does have some special powers sometimes.

Also sorry, this is getting long. I forgot to mention in the earlier discussion that remember that both heat conductance and capacity is relevant and they are not the same. Water for example has better heat capacity than copper. And when it is even moving water .... So im not adding anything meaningful to the discussion besides saying that heat capacity alone can not account for the spikey temperature , as water has higher heat capacity than copper.

 

[Krieger-San]

Thanks for the info!



I had an idea that maybe TEC could be integrated into the secondary half of a radiator, in order to chill the water temperature to near or below ambient. I guess, if you went below ambient, you'd run a risk of condensation, so it should be controlled by a thermostat or both a thermostat and hygrometer.

I think this has maybe already been done. I recall these systems from back in my systems builder days, the CoolIT Systems Freezone

 

[jonaswox]

I think this has maybe already been done. I recall these systems from back in my systems builder days, the CoolIT Systems Freezone

It has both proven to be , and is blatantly a bad idea. If you want to go the way of an actually active cooler , why not just use a compressor system on top of a water system? Remember that the ridiculous efficiency of heatpipes comes from the fact we are using evaporation and condensation -> the exact principles of einsteins fridge, and every fridge on the planet today. In modern earthheat systems -> they go even a step further with supercritical water, that goes something like 100x even more insane at moving and dumping heat. The whole thing about temperature standing still on phase change is the key to this power. The phase change itself contains energy -> besides the temperature.

The problem is that these kind of coolers are not able to withstand high loads, and even if you manage to do it, it would use ridiculous amounts of power - compared to a tried an tested method of cooling. And the design has this inherent flaw that if it is overloaded it turns almost useless.

Besides that , the genius solutions does , like heatpipes either provide the energy themselves (it runs off of the energy its suppose to dissipate, its beyond genius) , or uses some super cheap endothermic reaction to extract the heat, like evaporation.

If you want just high dissipation ability, external radiators, copper piping and a good amount of water does an amazing job, and you can deliver the waste heat wherever you want in the house. Maybe derp in Florida but massive in norway. You can cool the pipes by digging them into the earth , you can evaporate stuff in proximity to the pipes, only your imagination and needs sets the limits really.

As there is a big discussion in this thread on the subject of block vs water. Im will throw in my take, because I feel like its simple and actually helpful in this situation of creating an overview of the situation.

The ability to shortterm cool the chip is more or less deductible from the temperature of the block at the intersection surface. So the temperature difference decides what the Watt of transfer per second will be -> away from the chip.

Then you have the block heat capacity (like battery capacity) - and the block also has a number for how many watts to rise temp by 1 degree. I hope its fairly obvious that these numbers easily combine into a formula that calculates the short term capability of the cooler.

To then determine the long term capability we need to find the equilibrium with the ability to remove the heat. That is; the transfer coefficient between the block and the water -> reliant on the temperature delta again.

So while setting the whole system up in your head is fairly simple, actually doing the calculations involve system of differential equations. And is the reason why thermodynamics in general is a hard field, and something like aerodynamics is just massively expensive to do right. The simulations involve differential equations systems and is a pain to solve - so in practice one is always pushing the resolution ALL You can to even make the calculation feasible. This maybe works fine for everyday tasks with your aero software, sure, but do you actually NEED high precision? good luck 😛