Return to Level1Techs.com

Prime95 and Ryzen 3rd gen overheating issues

System:
Ryzen3600, 16gb lpx 3200mhz ram, MSI Tomahawk Max, Scythe Ninja 5 with 2xArctic P120 fans

OC to 4.45hz, 1.375v stable enough for:

  • Intel Burn Test- Temp 83C - CPU Power 95W
  • Sandra bench 235gops, 177gflops,
  • CinebenchR15/20 - 79c/82w and 80C/86w respectively

… but crashes instantly in Prime95 Small FFT

On lower OC, Prime95 Small FFT:

  • 4.4Ghz, 1.375v - runs but goes instantly to 92C and then I have stop it afraid it’s gonna fry.

  • 4.35Ghz, 1.325V - runs at 87C/97W - impossible temp to test for stability for longer periods.

4.3 and 4.2 Ghz with lower voltages 1.28/1.275 still produce unfeasible high temps (83-84C) for few days stability testing.

These super high temps and wattages only happen in Prime95 Small FFT. I stress tested in Aida64, Blender bench, Sandra, CPUz… etc.

My questions are:

  • why is Prime95 considered the definitive stability test, many recommend it as the “proper” way to test OC stability, when it induces totally unrealistic levels of load… and why is the feared Intel Burn test much milder in terms of power draw/temps?

  • Is there a KNOWN issue Prime95 and 7nm Ryzens not getting on well? I heard something mentioned that Prime95 is being worked on to address this “issue”…

  • Does successful completion of multiple passes on the “High” setting in Intel Burn test validate stability of my CPU or do I need to put it through 3 days of Prime95 Small FFTs to be sure?

It isn’t unrealistic.

As an example, an upcoming game called Dual Universe uses AVX256 to do the procedural generation of its world. It is heavily multithreaded. It will crash any system that won’t pass Prime95 AVX.

Other games in the past have tried to release code using AVX but had to give up because of the loud entitled complaints of a tiny percentage of their users.

I believe professional tools which do heavy physics simulation like heat transfer, virtual wind tunnels and fluid flow also make heavy use of advanced vector instructions.

As for Ryzen, it sort of sounds to me as if you are using a fixed clock OC. It would be much more flexible if you went back to using PBO, then it could adapt to stay under the heat limits.

Thanks for your answer!

I found this explanation here (although this refers to Intel CPUs):

I wonder if the procedural world generation based on AVX in the named upcoming game would load the CPU in similar fashion as Prime95 (AVX instructions on top of the 100% load).

I did all my THERMAL testing with Prime95 AVX disabled for now. With AVX on and all core overclock, I reckon no cooler would keep up. It’s not just about wattage (goes up to a mere 160W) but the very small die size - a lot of heat concentrated in a tiny space and it’s also awkwardly positioned towards the edge of the heatspreader… silly stuff!

When running Prime95 with AVX ON at STOCK - it’s interesting how Ryzen Master TDP reading is showing well below the chip’s factory rated 65watt! I guess that’s a form of throttling.