The tests were done on 2 CPUs across 2 different Intel Haswell platforms:
1) LGA 1150: Intel i7 4770k
This 4 core, 8 thread unlocked CPU is still one of the most popular mainstream CPUs from Intel. The newer i7 4790k is based off the same platform and performs the same clock to clock, while perhaps running a bit cooler. If anything, the older 4770k would benefit more from a custom loop. Sorry, no Skylake hardware here!
Motherboard: Asus ROG Maximus VI Formula
RAM: Corsair Dominator Platinum DDR3 1866 MHz (2x8gb)
CPU frequency: 4.7 GHz at 1.4 Vcore
2) LGA 2011-3: Intel i7 5960x
The behemoth 8 core, 16 thread unlocked CPU has a very high heat load due to 8 unlocked cores. Running at ~$999, it is one that benefits from a custom loop for sure.
Motherboard: Asus ROG Rampage V Extreme
RAM: Corsair Dominator Platinum DDR4 2666 MHz (4x4gb)
CPU frequency: 4.4 GHz at 1.3 Vcore
Pump: Swiftech mcp35x2 set to 1.2 GPM
Controller: Aquacomputer Aquaero 6 XT
Radiator: HardwareLabs Black Ice Nemesis 480GTX with Noiseblocker NB-eLoop B12-3 fans at full speed
TIM: Gelid GC-Extreme
Everything required was placed inside the hotbox and the ambient temperature set to 25 ºC. TIM cure time was taken into consideration and 5 separate mounts/runs were done. For each run, a 90 minute Intel XTU stability test was performed. XTU is a stability test from HWBot that uses a custom preset of Prime 95 to ensure the load is uniform on each run. CPU core temperatures were measured using Aida64 and average core temperature was recorded at the end of each run. Loop temperatures were recorded using 2 inline and 1 stop plug type temperature sensor connected to the AQ6 and the average loop temperature was recorded at the end of each run. A delta T of CPU core and loop temperature was thus calculated for each run with an average delta T then obtained across all 5 runs. This way the cooling solution is taken out of the picture. The measurement cycle was done for both blocks in both orientations and the average is reported below with standard deviation accounted for.
As a reminder, this is the regular orientation:
and this is the goofy orientation:
Now on to the results:
Note that these results are with the corrected installation without the locking nuts on LGA 2011(-3) which did help get a better TIM spread aka better contact with the CPU IHS. The Eisblock XPX now performed better, and on par with the leading flagships of 2016 which is good news. As always, note that these results are specific to my testing setup with my CPUs and your own findings may vary so refer to as many reviews as possible to get a general idea on where these components stack up across a larger sample size of blocks and CPUs alike.