What happened when I reseated the Antec Kuhler 620's water block?

In the previous post I mentioned that there was a problem with cooling of my new CPU. The Antec Kuhler 620 CPU that I had been using to cool it was not up to the task. The CPU was touching TJ-max kind of core temperature when stress tested using OCCT 4.4.

There was one thing for me to do: reseating of the water block.

I had some free time in the last weekend as the baby went into his regular evening nap, so I went down to business.

Almost all the guides out there recommend that you use a tiny amount of thermal paste in the middle of the heat spreader and let the pressure from of the heatsink mounting mechanism spread and thin the paste. I followed the same guide. But when this method is employed, the spread become a circular area and the edges don’t get proper contact with the heatsink. I thought the thermal issues that I had were caused by this contact issue. I cannot remember if I mentioned this before, but the pressure from the mounting mechanism seemed insufficient. The spread could be better with other heatsinks with better mounting pressure.

My Devil's Canyon chip runs really hot!

In the previous post I mentioned how I overclocked my new Devil's Canyon chip to 4.6GHz. Going from 4.5GHz to 4.6GHz was difficult and the settings are not even finalized yet. It requires further tuning.

Finding the settings that would make the CPU stable was one thing. Keeping the CPU cool while stress testing was another. My Antec Kuhler 620 was terrible at it. Even at the conservative settings that I used for 4.5GHz overclock, the core temps hit 90C once in a while. Devil's Canyon chips are supposed to run about 8C cooler than the original Haswell. I was stupid not to test the temps with my old CPU before the switch.

Plus, these temps were not observed while running something crazy like Prime 95 v28.5 or Linpack. The stress test that I used was OCCT 4.4. It is not a complete slouch at stressing CPUs as it, I believe, does heat up the CPU more than AIDA64 - the one originally recommended for Haswell. Still, this shouldn’t have happened.

There is no need to emphasise that there is a problem.

Assuming it was inadequate contact between the CPU and the water block, the water block was reseated once, which unfortunately made no difference. However, I'm not so confident in my application of the thermal paste. I felt there was too much paste the last time. This time it could be too little. There is no way to know without examining the foot print of the spread, but that means taking if off again. Time is something I don't have these days.

Before thinking about getting a better cooler, I think I should try reseating it one last time. Like I mentioned earlier, it could be simply my incapability to properly apply thermal paste. If the cooler has gone bad because it wasn't been used for 2 years, then there is no choice but to replace it. It wasn't a great cooler to begin with.

Overclocking my Devil's Canyon Core i7 4790K

It's Haswell overclocking all over again. In the previous post I mentioned that I was able to boot into Windows at 4.6GHz with just 1.25V for Vcore with my new Devil’s Canyon chip. I ran OCCT for just a minute and felt great about it as it is did not crash. But I indeed did not run long enough. I wanted to see if it can do better.

Next I tried 4.8GHz at 1.25V. Alas, it gave a BSOD while booting Windows up. Sigh! Then I increased the Vcore up to 1.30V and tried. This time it booted into Windows without any drama. But OCCT crashed instantly. Again, sigh! So it definitely is not a super chip.

Then it was time to start overclocking properly. The stress test of my choice was OCCT 4.4. While it is not as hardcore as Prime 95, which at version 28.5 seems to be an impossible stress test to pass without delidding - even with the "Next Generation Polymer" based thermal paste that Intel is claiming which performs much better than the crap that they use with non-Devil's Canyon parts. Since it seems that I would be stuck with this CPU probably until I leave Japan for good, I didn't want to see it degrade. Hence, this time around, I am very conservative with the length of stress tests and how much heat I put it through.

While I know that the CPU must be well capable of at least 4.6GHz, I wanted to start low. I dropped the multiplier to 44x, Vcore to 1.2V, set memory speed to XMP, set Ring voltage to 1.1V and let everything be automatically determined by the motherboard. This set the Uncore multiplier to 40x and Input Voltage to 1.75V. And this deemed stable in OCCT. Good! Now, we have a starting point. It is possible that the CPU didn't need 1.2V for Vcore, but I didn't want to check the absolute best. I only ran OCCT for like an hour.

4.4GHz settings

• Vcore: 1.20V
• Ring Voltage: 1.10V
• Uncore multiplier: 40x (AUTO)
• Input Voltage: 1.75V (AUTO)
• Load-Line Calibration: Level 8 = MAX (AUTO)
• Memory: XMP (2400MHz/11-13-13-35-2T/1.65V)