Scythe Fuma CPU Cooler Review: A Little Off the Top

Test System and Benchmark Results

Test System:

Test was conducted on a DimasTech MiniV1 open-air test-bench for parity performance between tower-style coolers and C-type downward or upward blowing coolers. Ambient temperature is measured at the fan intake and kept as constant as possible (room controlled temperature). All case and heat sink fans were benchmarked with fan control settings disabled and running at 100% unless otherwise specified.

To get closer to real world performance, a retail Intel i7-4770K CPU was used (non-delidded). Stock settings use a constant 1.1Vcore voltage while overclock settings use a constant 1.2Vcore voltage with EIST disabled (1.28Vcore OC test added for high-end air coolers). The Intel core i7-4770K integrated graphics was used for display output and no discrete graphics card was installed to prevent additional heat sources from contributing inside and outside the case. Multi-core enhancement is enabled in the BIOS by default and left on (ASUS Turbo Core policy used instead of Intel standard). Note that performance results on this setup are not directly comparable with any benchmarks from some of the previous reviews as the UEFI BIOS on the motherboard has been updated and now has a much tighter control on the Vcore voltage.

Mushkin Stealth DDR3 modules ensure full compatibility with any CPU cooler tested as its heatspreaders only add ~1.5mm to the standard JEDEC module height of 30mm compared to other modules which limit cooler compatibility.

Temperatures are averaged (last minute) from individual core temperature results monitored by AIDA64 after 15 minutes using the default CPU, FPU, Memory and Cache simultaneous load. AIDA64 is able to use the latest instructions including AVX and AVX2, etc unlike other older CPU load tests so it is also a lot more “future proof” as more software start to utilize it. FPU-only load average is used to simulate worst case scenario load levels similar to Intel Burn Test or OCCT. Please keep in mind that this test is brutal and not even close to real-world load (especially not that constant for that amount of time), so not many CPU coolers are expected to pass this test but the ones that do are exceptional. Results marked “100″ and in red means thermal limit was reached and the CPU was throttled, even for just but a second. This includes results where even just the first core reached the limit and even if it briefly happened. It is marked as 100 in red in the review if it happens three times. Three runs are conducted per cooler and a fourth run is done after a remounting to verify. Last minute average is taken instead of peak because it represents the averaged behaviour of the thermal performance instead of worst-case scenario or a snapshot. Temperature delta results are used to account for variance since not all heatsinks can be conducted on the same day.

The Corsair HX850W power supply’s fans only ramp up when system load is past 20% making it an excellent power supply to use for when testing the CPU cooler’s noise levels. Any load under that and the fan does not spin at all, effectively acting in passive mode.

The American Recording Technologies SPL-8810 meter is placed 20-inches/50cm from the source fan to measure sound level. The entire test unit is moved to an acoustically treated room to get the ambient noise as low as controllably possible for real-world results. All other fans are disabled to eliminate sound sources that are not from the cooling unit itself that is being tested. This includes all case fans and all other component fan is shut off. The boiler during winter or the air conditioner during summer is also shut off to eliminate ambient noise further and sound testing is conducted from 2AM to 5AM so sound traffic from the outside is minimized as much as possible. Fan levels are controlled manually via software and at a separate time from the temperature testing.

Temperature Benchmarks:

The default Scythe Fuma setup with two fans actually makes the cut-off, surviving the FPU-only load at 4.2GHz:

The single fan (middle) set up of the Fuma fought valiantly to keep from throttling but eventually buckled at 7 minutes which is still fairly impressive:

Next level is to push the Vcore voltage to 1.28V and the CPU to 4.4GHz which only the Scythe Ninja 4 and Noctua NH-D15/D15S have survived 15-minute load so far on my heatsink collection. The Scythe Fuma (two fans) throttles immediately unfortunately:

Here is the default load (NOT FPU-only) at 4.4GHz with 1.28Vcore for those interested in seeing it in a more practical sense. I left this running for 18 minutes, 3 minutes longer than the standard 15 I use for tests. Peak temperature is 89C on the first core but peak average is 79.8C which is still fairly good:

Just in case you skipped reading my methodology above, the FPU-only load is only meant to simulate a worst-case scenario that is impractical in any sense because that kind of protracted FPU-only load is non-existent on any real-world programs.

Thermal behaviour (white with red bias are the hot parts, black with blue bias are the colder parts):

Sound Profile (Fan RPM vs Noise Level) – Room ambient noise is 34.8dBA: