Associate
- Joined
- 25 Sep 2018
- Posts
- 8
I am heavily overclocking an Intel i9-7980XE CPU and decided to use liquid metal. I delidded the CPU and am using Der8auer's Direct Die Frame allowing me to put the cold plate of my custom water cooling loop directly on the chip silicon.
The problem is that the silicon die itself is slightly convex at its top. This apparently is a common issue for Skylake-X CPUs (especially the many core ones, as pointed out by no other than Der8auer himself in a video he made on the issue) - the CPU chip itself not being perfectly flat. It is less of an issue with conventional thermal paste, which is much thicker than liquid metal and can be applied as a thick coat to make up for any die/heat-sink/cold-plate warping or imperfections.
With liquid metal though, and I am using Thermal Grizzly Conductanaut, the general recommendation is to apply a thin coat to the silicon and a thin coat to the heat-sink. This accepted approach does not work for me. Only an oval portion of the chip, representing approx. the central 33% of the total area of the top of the die is coming into contact with the perfectly flat top-notch mirror finished cold plate I am using, an Aqua Computer Cuplex Kryos Next Vario with Vision. The periphery of the silicon around this oval center, due to the convexity of the chip (i.e., the ever so slight central bulge), is not coming into contact with the cold plate under tension (regardless of how much tension I place, within reason). I figured this out by leveling the cold plate on the CPU with LM applied, putting a lot of downward force on it, and seeing that only the central 33% of the die had touched the cold plate. This was easily noticeable due to the disturbance of the LM on both the die and the cold plate, forming an oval like fingerprint at the area of actual contact.
The result of applying two thin layers of LM to the CPU die and cold plate, respectively, is that some cores, probably the ones along the periphery of the die, are either not making contact or are making poor contact with the cold plate and jump to the Tj MAX temperature (or close to it) at load. Other cores near by are also extremely hot. There is an almost 60 degree C disparity between properly cooled cores (near the center of the die) and the overheating ones at the farthest point in the periphery.
I have been able to (temporarily?) resolve this issue on my motherboard which is sitting horizontally on a test-bed by applying a much thicker Conductanaut coat to both the die and cold plate. At this point, I see quite a bit of Conductanaut LM pooling on both surfaces, which goes completely against all recommendations for how to apply LM, but the temps are uniform across cores and very very good when mounted. So good in fact, that I am now motherboard VRM temperature limited in my ability to overclock (MB: ASRock Fatal1ty X299 Professional Gaming i9 XE) as I hit over 800 W at the wall, while the CPU cores remain at a temperature of about 65-70 C.
My concern is what will happen when I actually mount the motherboard vertically in a case, after I am done testing, and that the liquid metal will pool over time to the bottom facing edge of the CPU die due to the effects of gravity. This is despite the great surface tension and capillary action between the cold plate and the die keeping the Conductanaut mostly in place. I have protected the CPU PCB with many layers of nail polish so that any spilled over LM will not short out the tiny resistors and caps on the CPU PCB, but as soon as the LM pools and shifts off of the top facing part of the CPU die when the motherboard is mounted vertically, the cores situated at the top of the die will again lose contact with the cold plate and/or make poor contact, and begin to overheat at load. This effect may be very gradual, assuming it happens at all.
How can LM pooling with a thick coat be prevented and/or avoided, short of sanding down the silicon to make it perfectly flat and risking damage to a $2000 CPU (especially over time, as the protective layer at the top of the silicon will be sanded off)? Milling the cold plate (or laser etching it) to make it concave to match the CPU die is pretty much out of the question since this will be a very complicated expensive process and way beyond what I am willing to do to go forward.
I think for as long as I keep the motherboard horizontal, none of this will be an issue, but as soon as I mount it vertically into a case which is the goal, then the problems may start (after some unknown period of time, depending on how slowly the LM flows/transitions to the bottom).
I am left with just an LM solution of some sort, I think. Will the pooling of Conductanaut be an actual issue or will capillary action and the surface tension of the LM be enough to prevent pooling due to gravity over time despite the two thick coats of it applied to the chip die and cold plate, respectively.
Switching to a non-liquid-metal TM results in much worse temperatures - worse on the order of around 20-30 degrees C under load.
The problem is that the silicon die itself is slightly convex at its top. This apparently is a common issue for Skylake-X CPUs (especially the many core ones, as pointed out by no other than Der8auer himself in a video he made on the issue) - the CPU chip itself not being perfectly flat. It is less of an issue with conventional thermal paste, which is much thicker than liquid metal and can be applied as a thick coat to make up for any die/heat-sink/cold-plate warping or imperfections.
With liquid metal though, and I am using Thermal Grizzly Conductanaut, the general recommendation is to apply a thin coat to the silicon and a thin coat to the heat-sink. This accepted approach does not work for me. Only an oval portion of the chip, representing approx. the central 33% of the total area of the top of the die is coming into contact with the perfectly flat top-notch mirror finished cold plate I am using, an Aqua Computer Cuplex Kryos Next Vario with Vision. The periphery of the silicon around this oval center, due to the convexity of the chip (i.e., the ever so slight central bulge), is not coming into contact with the cold plate under tension (regardless of how much tension I place, within reason). I figured this out by leveling the cold plate on the CPU with LM applied, putting a lot of downward force on it, and seeing that only the central 33% of the die had touched the cold plate. This was easily noticeable due to the disturbance of the LM on both the die and the cold plate, forming an oval like fingerprint at the area of actual contact.
The result of applying two thin layers of LM to the CPU die and cold plate, respectively, is that some cores, probably the ones along the periphery of the die, are either not making contact or are making poor contact with the cold plate and jump to the Tj MAX temperature (or close to it) at load. Other cores near by are also extremely hot. There is an almost 60 degree C disparity between properly cooled cores (near the center of the die) and the overheating ones at the farthest point in the periphery.
I have been able to (temporarily?) resolve this issue on my motherboard which is sitting horizontally on a test-bed by applying a much thicker Conductanaut coat to both the die and cold plate. At this point, I see quite a bit of Conductanaut LM pooling on both surfaces, which goes completely against all recommendations for how to apply LM, but the temps are uniform across cores and very very good when mounted. So good in fact, that I am now motherboard VRM temperature limited in my ability to overclock (MB: ASRock Fatal1ty X299 Professional Gaming i9 XE) as I hit over 800 W at the wall, while the CPU cores remain at a temperature of about 65-70 C.
My concern is what will happen when I actually mount the motherboard vertically in a case, after I am done testing, and that the liquid metal will pool over time to the bottom facing edge of the CPU die due to the effects of gravity. This is despite the great surface tension and capillary action between the cold plate and the die keeping the Conductanaut mostly in place. I have protected the CPU PCB with many layers of nail polish so that any spilled over LM will not short out the tiny resistors and caps on the CPU PCB, but as soon as the LM pools and shifts off of the top facing part of the CPU die when the motherboard is mounted vertically, the cores situated at the top of the die will again lose contact with the cold plate and/or make poor contact, and begin to overheat at load. This effect may be very gradual, assuming it happens at all.
How can LM pooling with a thick coat be prevented and/or avoided, short of sanding down the silicon to make it perfectly flat and risking damage to a $2000 CPU (especially over time, as the protective layer at the top of the silicon will be sanded off)? Milling the cold plate (or laser etching it) to make it concave to match the CPU die is pretty much out of the question since this will be a very complicated expensive process and way beyond what I am willing to do to go forward.
I think for as long as I keep the motherboard horizontal, none of this will be an issue, but as soon as I mount it vertically into a case which is the goal, then the problems may start (after some unknown period of time, depending on how slowly the LM flows/transitions to the bottom).
I am left with just an LM solution of some sort, I think. Will the pooling of Conductanaut be an actual issue or will capillary action and the surface tension of the LM be enough to prevent pooling due to gravity over time despite the two thick coats of it applied to the chip die and cold plate, respectively.
Switching to a non-liquid-metal TM results in much worse temperatures - worse on the order of around 20-30 degrees C under load.
