My i7 8700k is getting unstable when I try to speed it up, and things are going wrong lately.
My i7 8700k is getting unstable when I try to speed it up, and things are going wrong lately.
That's a good change, but 200 Watts means your 8700K is running AVX / AVX2, which Prime95 v29.8 lets you turn off. • Prime95 v29.8 Even though apps like HandBrake use AVX code, the peak jumps and power usually won't go higher than steady 100% work of Prime95 Small FFTs when all AVX tests are turned down. Click on the AVX options that aren't grayed out so all three AVX boxes are checked, as shown above, then try it again.
Here is a breakdown of the text:
A feature that has not yet been tested properly: you aren't finished with just Prime 95. It only checks if your CPU cooler stays cool under heat. There isn't proof yet that your overclocking voltage works well. Prime 95 doesn't test both things effectively because its workload is too steady for the voltage check. Instead, run Cinebench R20 in "Infinite Loop" mode. To set it up: go to File > Preferences and find Minimum Test Duration. Enter a number you can't possibly reach. Run the test for an hour - the longer runs are better though.
And there it is. Good job! AVX works much better when used for transcoding than when running in P95 mode, because P95 uses a lot of work. So if you turn off AVX in P95, your power use gets close to what it would be just doing transcoding. That makes sense and is exactly what we expect. To keep P95 stable with AVX on, you'd have to raise the voltage (Vcore) or slow down the clock speed by 200 to 300 MHz. Both of those things use more power and make heat worse. You could change your BIOS settings to lower the AVX option just so it works in P95, but since your current Vcore setting seems fine while doing transcoding, why would you slow down just for that? That's one reason we don't do it. However, as Phaaze88 mentioned earlier, we need to check if your voltage and power supply are stable. Download and run the HWiNFO tool with P95 Small FFTs but without AVX. Then look at the minimum and maximum Vcore numbers while it's working steadily. This will tell us how much the voltage changes during normal use. Letting the CPU, cooler, socket, motherboard, and VRMs sit idle for 10 to 15 minutes is more than enough time for everything to reach a safe operating temperature. It makes no sense to let them heat up longer than that. Run HWiNFO with just "Sensors" selected. Once we know how the voltage behaves in P95, then take another transcoding session using HandBrake, which you used when you first found the problem. Look at the minimum and maximum Vcore numbers from that run too, and compare them to see if the fluctuations match. A difference of 16 to 24 millivolts (or about 0.016 to 0.024 volts) in P95 suggests good voltage control, but values higher than 48 millivolts point to bad regulation and a likely unstable motherboard VRM.
I ran some longer tests on prime95 and got unstable (BSOD). So I changed the offsets again and kept going until I found a safe spot at an offset of +10 with no BSOD. Here are my results: Prime95 uses about 1.2 watts. Handbrake also uses power but shows different numbers because it's cooler. The average voltages for Prime95 are around 1.19 volts, while Handbrake averages higher at about 1.24 volts when the temperature is lower. I think my voltage regulation isn't great right now. Are there any other steps I can take to make this better?
Power usage and Core temperatures look okay for now, but your CPU is showing a lot of wobble: 64 millivolts in the Vcore during steady workloads like Prime95. That means your voltage regulator isn't sticking straight. Just to be sure, please run another round of Prime95 Small FFTs without AVX. Before you start recording with HWiNFO, make sure P95 is already running and wait until it stops. Then, record the lowest and highest Vcore values while that test is going on so you only see what happens during the Prime95 run. The average Vcore doesn't count here, so don't write that down either. Transcoding makes workloads swing around a lot more than P95 does, so if your CPU was working well before, it might show much bigger swings (around 144 millivolts) during transcoding compared to just 64 millivolts in Prime95. We don't need to repeat HandBrake for you right now, so please just run the P95 test one more time if that helps.