CPU Cooler Testing Methodology – A Long Road
Forward Onto A New Methodology
The intent of our CPU cooler testing methodology is to recreate real world yet repeatable testing conditions in order to comparatively benchmark competing heatsinks, AIOs and other cooling devices. We have decided to move well beyond the “slap a heatsink on a test bench and log results”.
Whereas we previously tested on an open, horizontally oriented test bench which relied on ambient air temperatures to create a stable testing environment, we have decided to move towards a different approach. After consulting with a mechanical engineer* who specializes in heat transfer as well as several OEM heatsink manufacturers we have validated the open system with the motherboard in a horizontal position can create artificially inflated results in certain situations.
This is due to the following factors:
- The cooler placed vertically can allow it to wick away heat from the processor in a more efficient manner versus it being placed in a more natural horizontal position like you would find in a typical ATX case. It can also obfuscate the negative heat transfer effect brought upon by poor heatsink base finishing or an improperly designed mounting system. This can improve the results of products that would normally underperform.
- Certain heatsink designs create localized heat zones around key motherboard components or run contrary to a enclosure’s typical airflow patterns. An open test system would not allow us to calculate the negative effect this has on key items such as inner case & VRM temperatures.
- AIO’s can aid in case airflow in many situations (though not in a configuration where its pulling air into a case) by providing additional outwards-bound air movement and by exhausting much of the CPU’s heat immediately outside the case. This is one of their main benefits since it can drastically lower case temperatures but installing it in an open system will completely and unrealistically negate this benefit.
While testing on an open bench will still yield apples to apples results, they are not the ones we are looking for here. Rather, we wanted ours to represent more of a real world scenario, though under very tightly controlled testing conditions to remove sample variance.
With this in mind, we have decided to use a completely closed test system in the following format:
– Fractal Design Meshify S (tempered glass side window)
– 2x Intake 140mm fans operating at a constant 600RPM
– 1x Exhaust 140mm fan operating at a constant 600RPM
– If an AIO is installed, it is placed on the top radiator bracket, exhausting air outside the case
Other Test System Components
CPU: Intel Core i9-10980XE
– Stock Speed = ASUS Optimized Defaults (~150W – 165W)
– Overclock = 4.5GHz all core rate, constant @ 1.25V (~210W – 225W)
– Threshold for thermal throttling modified to 110°C
This may not be a common CPU but it allows us to test a wide range of thermal output conditions from 150W to approximately 225W (consumed power, not CPU thermals). If a cooler passes these tests, you can safely assume it will be more than adequate for more efficient processors from AMD and Intel.
Thermal Compound: Arctic MX-4
Motherboard: ASUS X299 Edition 30
GPU: ASUS RTX 2080 Ti STRIX OC
Memory: G.Skill Trident Z @ 3200MHz
Storage: Samsung 860 EVO 1TB NVMe SSD
PSU: be quiet! Dark Power Pro 11 750W
Interior & Exterior Environmental Conditions
In order to maintain consistent temperature conditions, all testing is done within a room equipped with a dedicated HVAC unit and the capability to maintain ambient temperatures with a variance +/- 0.5°C.
With that said, ambient temperature is maintained at 22°C throughout testing and if it dips below or goes above that point by more than 1°C, testing is halted until things stabilize. This is monitored by a calibrated Extech Thermometer equipped with dual Type K thermocouples and an alarm set to notify if temperatures move outside our deltas.
Before testing begins, the system is left at idle until interior case temperatures reach 25°C as monitored by a Type-K thermocouple placed ~1.5” away from the middle of the case’s upper radiator bracket.
Unlike exterior temperatures, interior temperatures are allowed to rise throughout the test. However, after each test is complete and before beginning the next one, the system is left in an idle state to naturally come back to 25°C.
A Note About Mounting
Due to a number of factors, results can be highly variable from one cooler mount to the next. This could be due to a poor mounting method, insufficient / excess thermal compound, etc. In order to partially remove this variance, the results you see are the best achieved between two separate mounts of each cooler.
In addition, in order to normalize testing Arctic MX-4 thermal compound is used all comparative testing. However, all coolers are initially tested with their packaged TIM. If there is a large difference in results, we will report the variance within the review.
Load Conditions & Logging
The load placed upon the i9-10980XE is through AIDA64 Engineer Edition System Stability Test (CPU test only). While logging is done through HWInfo64.
This is done for a constant 30 minutes are logs are pulled from HWInfo after this time has expired. The 30 minute load has been chosen since within this timeframe most coolers / AIOs reach their peak thermal saturation point or liquid temperature.
The following conditions are logged (only some reported):
– Max CPU package temperature
– Stabilized CPU package temperature after 30 minutes (logged & reported)
– Max individual CPU core temperature
– Max VRM temperature
– Stabilized VRM temperature after 30 minutes
– Interior case temperature after 30 minutes (logged in Extech thermometer via Type K thermocouple)
Note: VRM & inner case temperatures are always logged and if there are concerning results, they will be reported within the review. Otherwise, for simplicity’s sake results will be strictly kept to CPU-centric temperatures.
CPU Cooler Fan Speeds
ASUS Default Fan Profile:
The default fan speed profile on the X299 Edition 30 dynamically modifies the fan speed (a percentage of the fan’s maximum RPM level) based on CPU package temperature. The levels are:
<40°C = 20% Fan Speed
>40°C / < 60°C = 20% to 55% Fan Speed
<60°C / < 80°C = 55% to 100% Fan Speed
> 80°C = 100% Fan Speed
Fixed RPM Modes:
In an effort to normalize results while testing various noise levels & cooling potential the default fan profile is overridden with two separate fixed fan speeds. They are:
1000 RPMs – All cooler fans set to a constant 1000 RPMs regardless of their maximum RPM levels or CPU temperature
Full Speed Fans – All cooler fans are set to maximum level
While AIO coolers sometimes have built-in defaults with their shipping software, all default tests are conducted with their fans at the speeds indicated above. Meanwhile the pump speed is left at its stock level.
If the AIO does have specific preset performance profiles, they will be tested separately in its individual review.
Reported Results & Acoustics
All results in our charts are the stabilized CPU package temperature (as reported by HWInfo64) after 30 minutes of load. In addition, special care is given to monitor for any processor frequency throttling within that time span.
Optimally, our CPU would operate at maximum boost frequencies at stock and our predetermined fixed speed during the overclocking tests. Therefore, a cooler will fail a test (and we will indicate so within the charts) should speeds be lower than expected.
The thermal throttling conditions are:
Stock CPU: < 3.8GHz on any core
Overclocked CPU: < 4.5GHz on any core
All acoustic results are done with a calibrated decibel meter placed at roughly 30” away from the closed test system and roughly at the same level an individual would be sitting when using the PC.
Like the temperature results all Acoustic results are logged once the 30 minute timer has expired.
As a general guideline, these are the approximate result outcomes:
< 35dB – Whisper quiet
35 – 40dB – Barely noticeable noise
40dB – 45dB – Noticeable noise
> 45db – Loud
Note: For reference the test system with no CPU fans running and the 3x system fans at their constant 600RPMs measures a constant 32.8dB.
* Source: Matthew Brassard (M.Eng) Thermal Analyst; MDA Systems