[SOLVED] CPU Power Issue - Spiking During Stress Testing

Jul 28, 2019
4
0
10
0
I'm currently having some strange issues with my machine (which has been stable for months).

When I start a stress test, everything works perfectly for 4-5 minutes. CPU (9900k) temp is stable around 50C @ 4.8GHz (custom loop cooling), CPU power is stable at like 110W, voltage is locked at 1.23V, and everything is working great.

After a few minutes, the CPU Power draw starts jumping (nearly 200W), the temperature spikes up to 80C, and I kill the test. Meanwhile... the actual core voltages haven't moved... which means the CPU must start drawing more current for some unexplained reason? Never run into something like this one before... all seems beautifully stable, until the power draw jumps up?

Anyone have any ideas what may be causing this?
 
Jul 29, 2019
2
0
10
1
I think this is due to the test (I'm assuming you used Prime95 blend). Change from blend to small or large fft and see if it 'starts' at that 200w spike.
 
Jul 28, 2019
4
0
10
0
So that's the confusing part... Nothing today.

I DID have my AIO pump die a few weeks ago, which I replaced with a custom loop. That did result in the CPU hitting thermal limits and actually shutting the machine down, as it happened when I wasn't around.

Weirdly, I saw no issues resulting from that until now... it's been running fine since that happened, until today.
 
Jul 29, 2019
2
0
10
1
I think this is due to the test (I'm assuming you used Prime95 blend). Change from blend to small or large fft and see if it 'starts' at that 200w spike.
 
Jul 28, 2019
4
0
10
0
Interesting thought. I'll test that tonight when I get back home.

You're guess is correct in that I'm using P95 blend. Your theory is it's switching test modes after 5 minutes? Definitely not something I ever would have thought of, but definitely worth trying.
 
Jul 29, 2019
2
0
10
1
Right, I'm thinking a software change at the 5-6 minute mark. At that point, more wattage, and probably a temperature spike to go with it.
 
Jul 28, 2019
4
0
10
0
I can confirm you are entirely correct... the P95 blend test clearly starts with large FFTs (memory intensive), and then after a few minutes switches to either small or smallest FFTs, which is much more CPU intensive... so while the CPU is pinned at 100% usage the entire time, the workload is VERY different, as is the wattage the CPU draws.

Definitely not something I ever would have thought of, thanks for the help!
 

ASK THE COMMUNITY

TRENDING THREADS