Chasing degrees - how to have the coldest CPU on the block

If you've purchased a high-end CPU in the last few years you will know that performance comes at a cost – heat. We recently wrote about how you can decrease the noise levels of your cooler if you are willing to compromise ever so slightly with performance.

This article is the direct opposite. We will assume that you are similarly obsessive as our in-house specialists and will do everything in your power to chase that last degree, and squeeze that last ounce of performance out of the parts you spent your hard earned cash on.

We recently got some 14900K’s in the lab and wanted to see exactly how low we could get the temperatures on these absolute number crunching monsters.

A few important notes about our testing platform in this article:

We are using our Hydro X Series of custom water cooling to keep a steady coolant temperature of 30 degrees Celsius. This is quite low, but if you are a high-end user that chases performance, chances are that you have quite the overkill setup already. We measure clock speeds and temperatures, but will not dive to much into various benchmarks and performance numbers as we just care about temperatures and thermal headroom.

For the testing itself we are running Cinebench R23 on a loop for 30 minutes, and logging the last 15 minutes. The average values over these 15 minutes are what you can see in the graphs below. The 14900K running on our MSI Z790 godlike board is with unlimited PL1 and PL2 power limits (resulting in an average power draw of the CPU of 360W).

It should go without saying, but putting your new 500$ CPU in a tiny vice and physically pulling it apart will void your warranty, so anything described in this article should be performed on your own risk. CPU's have delicate little components on them so be very careful if you decide to try this out on your own system. We take no responsibility if your CPU dies in the process, but we will feel bad for you a bit.

Now, why did we go with 30 degrees coolant temperatures in the first place? Quite simply because we saw that at this level the chip was almost able to keep full boost clocks with our XC7 Elite. So in theory, if you are comfortable with having your fans cranked up, and have a lot of radiator surface in your custom cooling system you are actually able to sustain near full all-core boost clocks on the 14900K – almost.

Probably the easiest modification you can do to increase temperatures is to mount what's called a “contact frame”. The LGA 1700 chips are a longer, more rectangular shape than previous CPU’s from intel which proved a small issue in that they would deform ever so slightly under the pressure of the standard ILM (The thing you clamp down over your CPU to hold it firmly in place in the socket). Various companies have made modified contact frames that alleviates this, and makes sure the contact between the CPU and the water block / cooler is optimized.

We tried mounting one of these frames, and saw a minor temperature drop of around 1.9 degrees celsius. When looking at the the thermal paste imprint, it's plain to see that the contact frame makes a significant difference for getting a better contact between the HIS (The actual top part of the CPU that is in contact with your cooler) and the cold plate of the cooler.

In the above two pictures you can see how the thermal paste imprint is with, and without a contact frame mounted. On the left you can see that the thermal paste on the left and right middle edge are not compressed completely. This is due to the standard ILM pressing very hard down on those exact spots on the CPU. On the right picture you can see how the contact frame provides an even pressure on the entire IHS which makes it not “bend” in the middle.

This modification is pretty straightforward, but do still require you to remove the standard ILM from the motherboard. You also have to be carefull with some contact frames how much you tighten them as too much, or too little pressure can cause issues such as missing DRAM channels etc.

Delidding is a term for when you take the “lid” off your CPU. The “lid” refers to the IHS. On modern day CPU’s these are soldered in place and can be pretty tricky to get off so you will be needing a delidding tool.

These tools comes in many shapes and forms – the ones we use above is very easy to use. You just put your CPU into the small slot, and have to tighten a screw until the IHS gets forced off the CPU. The downside of this whole operation is that both this tool, and the liquid metal thermal paste you will want to use (to get most out of your delidding process) costs a pretty penny. We spent just around 70 dollars on the delidding tool itself, so if you are only going to be delidding a single CPU, then this is a factor worth considering.

After you have removed the IHS you will need to remove the solder from the CPU die, and the IHS as well. If you have purchased liquid metal thermal paste, you can put a drop of this on as it will dissolve the solder – just rub it in well and let sit for 5 minutes and wipe it off. You might need to repeat this process a few times. Be careful not to hit the small SMD’s that are on the actual CPU PCB while you wipe the solder off. This would irreversibly murder the CPU.

You will also want to remove the black rubber like substance on the actual chip. Do not use any metal objects for this, but use plastic scrapers or cards. Again, be careful not to damage the small components on the PCB.

Depending on your particular CPU sample, you might also need to flatten out the IHS itself. Our samples had quite some high-spots in the middle which can also be seen on the above picture where the nickel plating has been sanded off, but only in the center.

You can somewhat see if you need to do this by placing the IHS (after you have removed all the black rubber and the solder) on the PCB again. If the IHS only makes contact with the PCB, then you need to either remove more rubber, or sand it down slightly. You should be able to "pivot" the IHS around the center as it will be just in contact with the CPU die.

Comparing all three scenarios (stock, contact frame & contact frame + delidding with liquid metal) we see that we got a 9.3 degree Celsius decrease in temperatures on the package from our stock temperatures to our best case scenario. On the performance cores we likewise saw around an 8.5 degree drop in temperatures.

Now, while an almost 10 degrees decrease in temperatures sounds pretty great, it did actually not really do anything for us in this scenario. Remember how we wrote that we almost hit full boost speeds with the stock mounting?

On the graph above you can see that our average speed on the performance cores from our stock bracket, and our rather expensive experiment is almost the same. In fact, beyond the contact frame there is no longer any performance gains to be found since the CPU runs as fast as it can (without any manual overclocks).

So is it all just a waste of time then? Did we spend 130$ on contact frames, delid tools and liquid metal thermal paste for no reason? Not really.

Because while we ran these tests at 30 degrees coolant temperatures, it is unlikely that your system will be able to cool a loop with a 360W CPU down to 30 degrees with a comfortable noise level on the fans. And this is where the gains are – thermal headroom.

Think of it this way – with our stock bracket we had to use 30 degree cold water in our loop to make sure that our CPU almost could hit full boost levels. For the sake of easier explanation, let’s just say that the CPU is able to JUST hit full boost levels at 30 degrees Celsius. We can’t run our fans any lower than, for example, 1600 RPM because that is what's needed to keep the coolant at this temperature.

Now, if we had 10 degrees of thermal headroom, we would not have to cool our water to 30 degrees, but only to 40 degrees. This in turn would mean that we can turn down the speed of our fans as we are not trying to cool the water down to 10 degrees above ambient temperatures, but now 20 degrees.

Alternatively, we could use the extra thermal headroom to further overclock our CPU and get some more performance out of it (with the cost of added power draw of course).

If all this is worth 130$ to you is not something we can answer – for us it's fun to play around with hardware and see how far we can push it, but in the real world the benefits are very small – especially if you mainly use your PC for gaming.

But then again, many things does not exist because they make sense, but because some members of the PC building community likes to continue to push boundaries. After all, how boring would it be if we all had the same beige box as everyone else?