For our CPU cooling tests, we use the same hardware, overclock and configuration for every test, to minimize environment variables in testing. This allows for all results across all coolers tested on the platform to be viable as side-by-side examination for direct compare/contrast.
CPU | AMD Threadripper 2990WX all 32 cores @ 3.40Ghz @ 1.20v |
Row 1 - Cell 0 | (3.00Ghz stock speed, single core boost @ 4.2Ghz) |
Motherboard | Gigabyte X399 Aorus Xtreme (bios vers. 8A07BG03) |
Memory | Corsair Vengeance LPX, 32GB (4x8GB) DDR4-3000 |
Storage | Corsair MP600 m.2 2280 NVMe, 500GB |
Graphics | Gigabyte GTX 1050 Ti |
Power Supply | be quiet! Dark Power Pro11 1200w |
Chassis | Corsair Graphite 760T |
Monitoring | CrystalFontz CFA-633-TMI-KU, 4x Dallas One Wire WR-DOW-Y17 sensors |
Fan Control | Corsair Commander Pro, 100%/50% PWM Speed profiles (liquid cooling pump always @100%, if applicable) |
OS | Windows 10 Pro 64bit |
Networking | Disconnected, not used |
Thermal Compound | Arctic MX-4 |
Data comparisons are based on data collected from testing performed on our new AMD Threadripper 2990WX system, including re-visiting many previously covered products which were originally covered on the prior testing platform which pivoted around a Threadripper 1900X (4.10ghz @1.40v).
All data reported for this article has been collected on the current AMD 2990WX TR4 platform and will be maintained as like-for-like evaluation of ongoing cooling coverage. We’ve recently swapped the taller, Corsair Dominator RGB DIMMs with Corsair Vengeance LPX for lower-profile memory, allowing for higher cooler compatibility for testing.
Prime95 v29.4b8 (no AVX) is used for two-hour intervals, one managing fans at 50% PWM and the other at 100% PWM with RPM measurements being taken every 3 seconds and averaged across the duration of each 2-hour capture. Omitting AVX instruction sets allows for accurate, 100% loads at chosen clock speeds, while allowing AVX instructions would provide higher, albeit, unrealistic synthetic CPU loads and excessive heat production, less indicative of real-world use.
This also allows for a greater range CPU coolers to be tested and compared without the need to configure the system differently for smaller coolers which may not handle the excessive thermal loads being generated during testing, while larger coolers might be better equipped to manage heat output produced by the 32 beastly cores from the 2990WX.
HWInfo64 is used for real-time core temperature readout, thermal throttling alerts, motherboard power consumption, CPU speed and logging of data, while a CrystalFontz CFA-633-TMI-KU is used to monitor and later average both ambient room (2 probes) and motherboard voltage regulator heatsink (2 probes).
With these temperature readings collected, CPU Core is defined as an average of all CPU core temperatures reported once per second for the entirety of the testing run from HWInfo64 data. This value is then taken as an offset difference from the reported ambient room temperature collected from the CFA-633-TMI-KU (also once per second and averaged). CPU PWM is defined by the voltage regulator heatsink direct temperature probe as an offset different from the same reported ambient room temperature reading for the same CPU Core test (100%, 50%). This helps define a working model of how well the evaluated cooler performs as a process of also cooling nearby hardware under loads, like our overclocked motherboard voltage regulator heatsink.