Temperatures: Stock Vs. Delidded Vs. Direct Die Cooling
With our platform built and the components prepared, it was time to quantify the thermal differences between delidding a CPU and replacing its thermal paste, and cooling the die directly. In each of the following charts, Without Modification means that no changes were made to the CPU; it's a stock sample. Delidded indicates that the processor's integrated heat spreader was removed and then replaced after using higher-quality thermal paste. Delidded and Cleaned means that we also took the time to remove Intel's adhesive residue, in addition to replacing its thermal compound with better stuff from our lab. Of course, the Direct Die entry corresponds to removing the IHS and mounting our thermal solution on the die itself.
Establishing A Baseline
For our first pass, we stuck with default settings and tested the processor using three workloads. Cinebench R15 is one of the most common benchmarks for quantifying the effect of an overclock. It can utilize one core or all of them, and it's incredibly easy to run. We specified a static clock rate of 4000 MHz and dialed in a 1.06V supply voltage. Those parameters should be easily attainable by anyone.
As expected, the stock CPU with no modifications posted the highest temperatures. Delidding helped shave off 5 to 6°C on average, and cooling the die directly helps even more.
Prime95 v.26.6 does not utilize AVX instructions. But it's still one of the most taxing workloads you can hit a CPU with.
Our Core i9-7900X continued operating at 4000 MHz at 1.06V. Not surprisingly, the finishing order remains the same, though the impact of more effective cooling is amplified.
Switching over to Prime95 v.29.1 adds AVX instructions to the mix. The brutality of this workload automatically forced our CPU down to 3600 MHz at 1.0V (due to our CPU ratio offset of -4). But despite the more taxing benchmark, less aggressive clock rates and voltages lead to slightly lower temperatures than the preceding run without AVX support.
If we stopped benchmarking right there and had to draw a conclusion based on these results, we'd recommend leaving your Core i9 alone. The risk of damaging a $1000 CPU, voiding its warranty, and sinking money into extras like the delidding tool just aren't worth a slightly lower operating temperature. However, the gains become more significant once you start overclocking to higher frequencies.
The Impact of Overclocking
Next, we ran the same three workloads using higher clock rates and voltages. In our pursuit, we sought settings that put our chip's temperature as close to 100°C as possible.
In Cinebench R15, that meant a frequency of 4400 MHz at 1.25V. The overclock imposed significantly warmer temperatures, which increased from 46 to 75°C out of the box. And the delta grew as we improved the configuration's ability to dissipate heat quickly.
With or without adhesive, delidding facilitated gains of 11 to 12°C on average, but also reduced the hottest core's temperature by 15°C. Direct-die cooling helped even more, yielding an 18°C drop compared to the manufacturer’s stock heat sink and fan.
We used the same settings in Prime95 v.26.6 (without AVX), and were treated to significantly warmer temperatures. One core even got as hot as 96°C (never mind our high-performance water cooler and fairly modest 1.25V voltage setting). This is where we realized the benefit of delidding: gains of 15°C after replacing Intel's stock thermal paste and 22°C thanks to the Direct Die Frame are huge.
We had to lower the clock rate and voltage significantly under Prime95 v.29.1. At 4100 MHz and 1.15V, it was already getting worryingly close to our 100°C target. While executing AVX instructions is known to impose warmer temperatures, we've never seen anything like this. And the Core i9-7900X only has 10 cores. Just imagine the 18-core Core i9-7980XE!
MORE: Best CPUs
MORE: Intel & AMD Processor Hierarchy
MORE: All CPUs Content