Trying to boost performance causes overheating problems with an older 6700K processor.
Trying to boost performance causes overheating problems with an older 6700K processor.
Hello everyone, sorry for the beginner vibe of this message. I’ve been running a 3570k since 2012, connected to an MSI Z77 M-Power board and cooled by a Noctua NH-D14. It’s been working smoothly for over seven years at 4.6 GHz with speedstep turned on and a variable v-core—no major problems so far. I’ve always preferred gaming, so upgrading wasn’t urgent.
In the meantime, I’ve started buying iMacs from Facebook Market occasionally. Sometimes you find some really cheap models and sell them for big gains. Recently, I got a 5k iMac with an i5 6500, and switching to an i7 6700k added a lot of value. I found a 6700k for £120 on the site and bought it. The seller had a £30 Asus Z170-A board, so I decided to go ahead and replace the Mac with my gaming PC. After assembling everything, I set the XMP profile to 3200MHz for my RAM. It booted up right away, but I noticed the CPU temps spiked to 90–100°C regularly—often jumping above 90°C in just one second, then dropping to around 40.
Using CPU-Z, I saw the v-core reading 1.392v under load, which explained the high temperatures. After some tweaks in the BIOS and trying ASUS Easy Tuning, the issue persisted. I checked the BIOS for a setting about air vs water cooling, but it was set to water cooling—even though the auto-tune thought otherwise.
After more adjustments, I finally got stable temps around 4.5GHz with a 1.296vcore. Still, they’re in the high 90s, which is pretty high. I’m wondering if this CPU needs a full rebuild or maybe a liquid metal upgrade. But honestly, my 6700k has been reliable enough for gaming, and I haven’t had any real issues.
Looking at the photos I’ve uploaded, do you think there’s something obvious missing? Or is this just a tough CPU that could use a serious overhaul? I’m open to your thoughts!
Thanks for the update. I quickly removed that option once I noticed the cooling concerns I mentioned earlier. I discovered my Z77 MSI board had a decent OC Genie—it didn’t boost CPU performance drastically, unlike the Asus model that just cranked everything up. It’s possible the CPU was faulty; the person who bought it used an AIO, but I’m sure most AIOs aren’t much better than a well-built NH-D14 with two fans. One thing I wanted to note is that while the heatpipes felt cold during testing, they should have been warmer. It seems we now know the right steps are to desoldering the board.
It's just Dumb95 if you keep running SmallFFT over long stretches (this is mainly for extreme temperature scenarios, real-world conditions rarely reach that intensity). Blend is better for stability checks—it handles smallFFT for a shorter duration and covers many other tests too. It can also stress the RAM. If P95 isn't your preference, ASUS Realbench's stress test works well for stability assessments. AIDA64 with FPU disabled is another solid option. With it on, performance resembles P95 SmallFFT. Games are really bad for stability testing unless you're playing them intensely for hours while covering all possible scenarios solo and multiplayer. If instability occurs and settings change, you'll need to repeat the process. Now that it's stable in one game, it's time to try another that handles CPU differently. P95 and Realbench provide consistent loads on CPUs, making them ideal for stability testing. You can schedule an 8-hour run and then assess performance later. Even after passing, unexpected random events might cause crashes. Or you might only complete a 4-hour test, but if it fails during a weekend session, running Realbench for 6-8 hours can reveal instability at the 6.3-hour point. In short, to achieve 100% stability, either run a single game for extended periods or use a dedicated stability suite while working elsewhere. That's why this software is designed for reliable use.
Alright, here’s a revised take on that:
So we finally had the courage to check our CPU. Everything worked smoothly—impressive how the IHS adds weight but the bare chip feels lighter. After swapping in fresh MX4 paste and adjusting the layout, temperatures dropped noticeably. The Core temps fell from 88 to 84, and so did the others, which is a nice improvement. Still not quite cool, but it’s a solid step up. The build feels a bit warm, though, probably because of the heatsink design. There’s still room for a more robust upgrade, maybe with copper, but that would require some extra work to fit the socket properly. Let me know your thoughts!
They provide copper IHS replacements for LGA1151 CPUs. However, I wouldn't recommend them in general. I'd use liquid metal and meticulously prepare the surrounding area around the die, then apply it precisely to the die and IHS—since it's conductive you must handle it carefully. After that, secure the IHS with superglue so it stays fixed. This should improve heat dissipation, and there are plenty of videos and guides explaining the process. LMs tend to clash with copper (it absorbs into it), so I wouldn't suggest a copper IHS. A regular paste plus a copper IHS would actually lead to worse temperatures compared to using the original chip with LM.
Thanks for the response. I've thought about using liquid metal, though I'm not sure it's the best way to handle the temperatures it generates. I'll monitor how the MX4 performs over the coming weeks and might rethink it. The advantage of a good paste is avoiding the need to re-glue the IHS and eliminating the risks associated with liquid metal in the CPU area. Honestly, there don't seem to be any components under the IHS that are a concern. Cheers!
I often rely on Realbench too, but it keeps crashing after about two hours of use. I'm gradually increasing the voltage until it becomes stable for a while. It really bothers me when my PC freezes during games, especially with the RVII adding so much random chaos. I don't want the CPU handling that kind of instability. If you're okay adjusting settings again in case it randomly crashes, then multiple long runs aren't necessary. Otherwise, if you're just looking for quick benchmarks like CB20 scores, you only need a stable setup—something simple enough to check with CPU-Z.