Most of us got started w/ stress testing in the 90s when ou water cooling systems were either new products or user re-application of components from other sources. As a novice some Guru suggested an utility to test something and that guy knew everything ... but that was 20 years ago and things change. Today, testing with synthetics is not really an appropriate strategy. What is the point or proving that yoiur system can do something it will never, ever be asked to do again ? Why unnecessarily reduce yiur OC or decrease performance by undervolting based upon a load your system will never ever see ? "Well if it's stable in everything but P95, it's not good enough?" Why, ... if it will never, ever, ever see that type load ? If the goal to eek the best performance you can running loads your system might actually be exposed to ... or running loads it will never see ? It like picking a CPU for your next build that has "more cores" when you don't use any applications that actually benefit from more cores.
1 Would you test an new 500 pound engine hoist by lifting 2000 pounds ? It will never see that load.
2. Would you test if your SUV could handle towing your new 450 pound SkiDoo / Trailer 12 miles to the beach (@ sea level) by towing a 12,000 pound trailer up and over the Rocky Mountains ? It will never, ever see that load or that altitude ... or those slopes.
3. Living in an area that sees significant snow and sleet, would you choose all season tires based only upon testing in only one weather condition ?
4. Would you evaluate a potential baseball draft choice by only throwing him fast balls ? What if he can't hit a curve ? P95 is a 200 mph fastball.
Synthetics came into common usage when folks were testing the water cooling systems for adequacy and it became the holy grail for those aching to get their name on web site OC Leader Boards The fact remains, it's a poor tool for "Real World" CPU Stability Testing.
a) Most time an older version is run as the newer versions w/ AVX and modern instruction sets can damage the CPU
b) So the "Well it's not really stable unless it passes Prime 95" is kinda silly because if using the older version, it is not necessarily stable with modern instruction sets.
c) You can have 24 hour stable P95 OCs fail in a multitasking benchmark like RoG Real Bench in 20 minutes. Been there, done that.
d) After you do the stability test , when is the next time you have a need to run P95 ?
There's a place for synthetics. We use P95 on a new build to quickly bring CPU Temps up to levels we never will or want to see during the CPUs lifetime. P95 allows us to cycle it up to 85+ and drop it to room temp 4-5 times to cure the TIM. The laptop I'm typing from (custom built Clevo) has a slight overclock . It has survived Furmark and runs fine with no undervolting. Temps never break high 70s. It doesn't use a laptop cooler... in fact, tried on on it's predecessor and it blew out 2 of the USB ports (Note: If you use a laptop cooler, plug it in to a USB hub with its own power supply).
Application based stress tests like RoG Real Bench will allow you to obtain higher performance for the life of your system, because you are not unnecessarily gimping performance to handle load conditions that it will never be exposed to.
a) Put the CPU under a "real" application based load which is far greater than your laptop or system will ever see.
b) It's a multitasking stress test which stesses the CPU in ways P95 is incapable and much more similar to what your system might, see in its lifetime
c) It tests the CPU under different types of loads at the same time .... a far more challenging situation than 1 single load type.
In short:
1. Stylish thin and light reduces battery life and cooling performance. Pick a lappie with physical characteristics and component specs that match your needs. Custom built recommended and cheaper. These always have a key combo which lets you set fans to 100% when desired.
2 If you need a laptop cooler, use it with an external USB hub, do not connect to laptop. Putting something under the back end of ya lappie whole using works wonders. I use these cause I have a dozen of them lying around but a small strip of wood is fine.
3. Run RoG Real Bench Benchmark (8 minutes) watching temps with HWiNFO. If no issues you can perhaps loosen up your undervolting and try again. When you think you at the point you wanna be, run the RoG RB Stress test with the 50 of the - 75% amount of RAM (or 100% if ya like but 75 more realistic worse case) for 2 - 4 hours.