With power density being what it is, and the IHS it’s self being insufficient to really take advantage of modern CPU designs, how long until we see CPU IHS with integrated vapour chamber?
My understanding is that this would be a substantial improvement in the CPU-IHS heat transfer bottleneck, going off what we see in laptop designs today.
Edit: adding to the OP to get a more interesting discussion going.
I actually watched that the other day, and it is pretty interesting stuff.
But, there’s also an LTT video of using an industrial chiller to overclock a 13900k, which still managed to hit TJmax with just a bit of extra voltage/speed, at around 370w or so I think? Suggesting that the actual IHS/TIM might be more of a bottleneck than it seems, even for modern tower coolers.
There’s also a video on solid state cooling where it was mentioned that vapour chambers/heatpipes will remain in use because of their ability to spread heat horizontally far more effectively than any solid material, due to the evaporation/condensation current flow pushing heat into the corners of the chamber to condense/wick back to the center, essentially making them some of the most thermally conductive systems physically possible. Take A Lab Tour Of This Solid-State Cooling Tech - YouTube
Of course, the heat still travels through copper, but the idea is that it can travel through a very thin single piece of copper to a vapour chamber directly, underneath the CPU cooler. If the IHS contains a direct-die vapour chamber, it could help to address the power density problem, which comes from the contact surface that is much smaller than the IHS it’s self.
My understanding is this is already the norm in any high-TDP microdevice, such as thin and light laptops, high power laptops, and even high-end phones and tablets.
The goal isn’t so much to transfer heat vertically from the surface of the CPU to the surface of the cooler more efficiently, but to transfer heat horizontally across the surface of the IHS, to then have a larger surface area to transfer heat to the tower cooler/heatpipes more efficiently. I think it probably wouldn’t have mattered much for desktop CPU designs from even 3~5 years ago, but only just the most recent desktop CPU designs, which have such a high power density that the copper heatspreader it’s self is becoming a bottleneck.
Likely never, it’s been tried by OEMs in the past and did not improve performance. The strength of vapor chambers are that they can spread heat over certain temperature ranges over a good distance, they aren’t good for transferring heat short distances.
Current IHS-es (“integrated heat spreaders”) are made out of treated copper, and they provide a bit of mechanical robustness, and you have heat pipes in the cooler/fan typically.
I suspect laptops will get it before desktops, despite being lower power, because of the overall price/performance considerations and the fact you’re making money on an integrated and more expensive product.
Would it cost more than $10/CPU to get rid of the desktop IHS?
I actually watched that the other day, and it is pretty interesting stuff.
But, there’s also an LTT video of using an industrial chiller to overclock a 13900k, which still managed to hit TJmax with just a bit of extra voltage/speed, at around 370w or so I think? Suggesting that the actual IHS/TIM might be more of a bottleneck than it seems, even for modern tower coolers.
There’s also a video on solid state cooling where it was mentioned that vapour chambers/heatpipes will remain in use because of their ability to spread heat horizontally far more effectively than any solid material, due to the evaporation/condensation current flow pushing heat into the corners of the chamber to condense/wick back to the center, essentially making them some of the most thermally conductive systems physically possible.
Of course, the heat still travels through copper, but the idea is that it can travel through a very thin single piece of copper to a vapour chamber directly, underneath the CPU cooler. If the IHS contains a direct-die vapour chamber, it could help to address the power density problem, which comes from the contact surface that is much smaller than the IHS it’s self.
My understanding is this is already the norm in any high-TDP microdevice, such as thin and light laptops, high power laptops, and even high-end phones and tablets.
The goal isn’t so much to transfer heat vertically from the surface of the CPU to the surface of the cooler more efficiently, but to transfer heat horizontally across the surface of the IHS, to then have a larger surface area to transfer heat to the tower cooler/heatpipes more efficiently. I think it probably wouldn’t have mattered much for desktop CPU designs from even 3~5 years ago, but only just the most recent desktop CPU designs, which have such a high power density that the copper heatspreader it’s self is becoming a bottleneck.
I 100% agree the ihs is a bottleneck in current intel systems because of their insane power density. The solution I think is just to delid them though, anything you add between the die and what is actually cooling the cpu is bad.
Liquid cooling is significantly more thermally conductive than vapor chambers or heat pipes at even modest flowrates. Also vapor chambers (and heat pipes) have a thermal flux limit where they just stop effectively transferring heat after a certain point which has already been reached for practical designs in server chips or pretty much anyone overclocking.
I think as power densities continue to climb that direct die liquid cooling will have to take place, but maybe that technology will never reach us consumers.
I agree with @twin_savage, don’t think a vapour chamber in place of an IHS is the solution.
In laptops and GPUs, you’re starting to see applications of phase change material in place of traditional single phase thermal compount between the “bare die” and heat sink/heat pipes.
I experienced this with a 10900K and NH-D15. Even with 3000RPM 140mm iPPCs, you could tell that it struggled to deal with power draw north of 280W effectively. Otherwise it could handle 260W all-day at a +50C delta with minimal heat soak inside the wind tunnel of a P500A loaded with iPPCs.
@twin_savage nailed it, the vapour chambers, like the heatpipes, have a maximum wattage they can handle, this is called in the industry the “dry-out wattage” referring to the point where you have no more liquid in your heatpipe/vapour chamber.
One big issue of IHS is that you add thermal interfaces, each coming with its own thermal resistance.
On our coolers, we have thermal resistance IHS to air that go down to 0.05 °C/W, even lower, in other words, at 300W, the IHS temperature is just 15°C above ambient, but still, the CPU hits 90°C, simply because of the thermal resistance Die to IHS, the thermal density is here a serious problem, but I am not sure the issue is with the IHS so much as it is with the interface to IHS.
I believe that the multi-die solution has much more future, at least for the heat management part, a Freezer 4U kept a Xeon 270W TDP at 54°C core temperature in our testing…
because that would be a disaster in power consumption… Peltier takes about 1W of power for 1W of heat removed…
I think it’s probably even worse efficiency than that
You also don’t get as dense of cooling you’d need, higher power ones are bigger than ihs, not to mention the heat is concentrated on one area of ihs
You won’t see a vapor chamber on your CPU any time soon. However you may see hose fittings at some point, or a ring-gasket that installs around the edge of an IHS specifically slotted to flow liquid. The future of cooling is likely going to be microfluidic cooling or micro-channel capillary cooling; higher pressure water flow to push coolant through the rear silicon layer.
The traditional IHS will probably not go away because of the mechanical purpose it serves.
Gamers Nexus just did an AMD lab tour and showed off a prototype IHS vapour chamber for a 7900x, there was a claim of a 3C improvement over solid copper IHS… although I’d imagine this benefit would only apply to specific lower wattage scenarios and turn into a negative as the heat is ramped up.
The same video has AMD clamming a 60c improvement by removing the IHS and cooling the dies directly.
Coming back around to this topic, I wonder if a sealed IHS could use water or thermally conductive fluid to the same effect through similar mechanisms.
The only practical solution is properly optimizing IHS thickness, IHS contact surface flatness, gap between IHS and bare die, and TIM material choice (soldered will always have a benefit over paste, I’ll never forget the Intel 4770k era).
I do not see micro-fluidics to be viable in mass produced hardware in my lifetime.
With that said if AMD continues to use their current IHS profile with Zen5, I’ll be skipping the next generation as well. So much benefit left on the table to maintain cooler mounting back-compat.
I never thought I’d see a truly necessary return to direct-die mounting for what used to be understood as acceptable at-load, long term temperatures.
Interesting that the new Intel i9-14900KS can be run delidded with warranty for select PC builders
I’d happily buy a delidded CPU and take care to install the appropriate cooler, but the idea of delidding myself seems a bit intimidating with such a high value component, even if der8auer latest video makes it look easy, and the gains worthwhile.
It was really interesting that der8auer brought up that letting the CPU block sit flush with the PCB degraded memory performance because it was getting too physically close to the memory traces.
Makes me wonder how long it’ll be before we get low-k spacers for direct die blocks.
