Processors

[alpere]

Hi, In prime95 small fft testing, when i select 24 thread workload temperatures are just fine not passing 68C with PBO auto.
In Cinebench R20b, it stays below 65C.

However, when i select 4 thread workload, small fft in prime95, temps hit the 92C wall, under 3 seconds and stay there. (most likely thermal throttle there).

I tried it with PBO off/on/auto/advanced with negative offsets, It always hit 92C in first 2 seconds. The only way to keep temp down in this workload is disabling CPB which in return 55C in same test.

My cooler is arctic freezer 34 duo, and case is aerocool airhawk duo. I have never encounter any thermal issues in real life applications, however having the possibility of some random workload could potentially damage the cpu is bugging me.

Is this normal behavior or some thing is broken in my system?

[trek]

This is correct function of processor, you can read about it in Anandtech review, page 8 https://www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-dive-review-5950x-5900x-5800x-and-5700x-te...

You can read it if you want and we could discuss this topic further, maybe some interesting contributions from others will follow.

[alpere]

so idea is, even total power draw drops vs full core load for 4 core load, individual core draw much more  versus full core load.

however, what we i see here is that temperature of cpu scales weirdly that, we can manage single core load and 6 or above core load yet not able to with 2-4 core load. 

My cooler is not aio liquid cooler still it is relatively good cooler and the airflow of case superb yet, cpu could hit maximum temp in matter of seconds in certain workloads which consist fever cores.

So my question ultimately evolves to this: how can we limit per core voltage. I can  sacrifice some performance yet not want to disabling CBO. Can we achieve sweet spot which my cooler able to manage temperature at fewer core load but not with entirely disable cpu boosting.

[trek]

Temperature does not  scale weirdly in your case, everything works correctly. What you write is a misconception that causes the same countless threads and questions.

Processor works similarly as GPU boost nowadays. The processor clocks itself to achieve maximum performance under given circumstances, one of these circumstances is temperature called in BIOS as Platform thermal throttle limit, which is set to 90C, which means that as soon as temperature reaches this limit processor built in overclocking will not proceed further.

This value was set by AMD as 90C is a perfectly safe working value, there is an additional milestone and this is 95C at which the processor starts real throttling.

Your processor cannot damage itself around 90C, it wants to run as fast as possible within its operating limits. If you like cooler temperatures, you can add cooling or change Platform thermal throttle limit in BIOS to e.g. 80C, so CPU will stop increasing its performance around this temp.

There is nothing to fix or repair, because the processor is working correctly.