Is version 26.6 Prime95 more realistic?
Is version 26.6 Prime95 more realistic?
Hi guys, I've been pushing my i5 6600K to its limits quite a bit recently. I built the system last week and started testing with Prime 95's latest release. During one test, the CPU hit 82°C! Others have mentioned that version 26.6 of Prime95 gives more accurate results, so I'm using that now. After further overclocking, my max temperature came down to 66°C. There seems to be a difference between the two versions, but is 26.6 really more realistic or are people exaggerating?
For a long time, 26.6 was seen as the final version that doesn’t cause local overheating and stops modern Intel processors! This might be due to it being the last one without AVX extensions for Sandy Bridge. The first stable release with AVX support was 27.7, and 28.5 was the first to include AVX2 and FMA3 for Haswell processors.
Stability checks and thermal evaluations aren't always the same. If your workload doesn't involve AVX code, testing for it might be unnecessary because a stable system could handle higher overclocks without affecting stability at normal temperatures. Tools like Prime95 v28.9, AIDA64 FPU test, Intel Burn Test (not an Intel tool), and LinPack can increase TDP beyond 120% when using Auto or Adaptive Vcore settings, leading to excessively high temperatures. Intel benchmarks their chips at full Thermal Design Power (TDP), aiming to match real-world conditions closely. For stability assessments, Asus RealBench is the best option, offering a realistic AVX workload with Core temperatures comparable to P95 v26.6 Small FFT results. You can find more details here: http://rog.asus.com/rog-pro/realbench-v2-leaderboard/. Also, check out the Intel Temperature Guide for further insights.