this is the situation: i'm a very happy owner of a threadripper CPU - 1950X, cooled by a Noctua NH U14S, Cpu is at default with a quad channel of ddr4 2133 - Asus Prime X399A with latest bios. Rig is very rock solid. Temps are very good (Tdie reading not Tctl): for example after 1 hour of kill_ryzen.sh (script to verify ryzen's segfault,which i conferm not existent in threadripper CPUs) max temp is 55°When i run cinebench or other benchmark temp is 50 or 51° max. Cpu cooler is installed properly and the thermal paste is ok.
So everything seems fine and is fine except for this: i don't understand why when i run some stress test, for example y-cruncher, during the execution of the single stress test temp is max 50°or 60° (depending of the stress test), but when a stress test finish and start another, during this interval of time i see spikes of 75° with fans suddenly ramping up speed, then suddenly temp goes down.. these spikes happens for 3 seconds...
Another example: yesterday i've tried aida64 cpu stress test: during the first 5 minutes i see max temps of 55°- 60° (with xfr working correctly 3.5 - 3.6 ghz)...then suddenly i hear fan ramping up and i immediately stopped the test: hwinfo was showing max cpu temp 70° and fans were at 100%. I restarted the same test and this didn't happen again, no spikes!....only max temp of 60°....Why? I have to say that this spike temps are confirmed in both dies: Tdie readings within the two dies are very similar, near identical...and identical when cpu is in full load (hwinfo shows two Cpu Nodes).
It's a problem related to hwinfo? Continuos "pinging" sensors can create errors?
I have to say that regarding to Ycruncher, this spike temps was happening with my previous 1700X cooled by Noctua Nh-d14 too. So this spikes are not related to a physical problems (bad cpu cooling), and are not related to a bad cpu or bad tdie sensors....three different ryzen dies showing same behaviours...under certain conditions.
Can someone explain me why this happens?
I suspect that tdie is not a simply value indicating temp, but it's a number calculated with other variables like voltage, type of instructions, load of the core, etc etc