Which version of Prime95 for Kaby Lake?

[1405]

Is the latest version of Prime95 safe to use with Kaby Lake processors, or should we still be using version 26.6? Or was that problem just a Haswell thing?

 

[CompuTronix]

1405,

The issue with Prime95 versions later than 26.6 is AVX code which adversely affects all processors that have AVX instruction sets. This includes 2nd through 7th Generation i7's, i5's and i3's. Pentiums and Celerons are not affected, as they do not have AVX instruction sets.

For thermal testing, please continue to use only Prime95 version 26.6 Small FFT's.
For stability testing, I recommend Asus RealBench - https://rog.asus.com/rog-pro/realbench-v2-leaderboard/

Also, you might want to read this Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT

 

[Ecky]

Any are "safe" to use, but anything later than 26.6 (as CompuTronix says) put unrealistic loads on your CPU, and aren't a good stress test because it's not anything you'd typically ever run.

 

[1405]

Thanks. Great answer. But that raises more questions for dumb ol' me. (1) We should avoid any P95 ver. past 26.6 for Sandy Bridge forward? (2) I just read where AMD Bulldozer forward also has AVX instructions. So same with them?

 

[Ecky]

I only have a vague understanding of Bulldozer's AVX pipeline, but I believe that, though these instructions are technically compatible, they run at half speed so they don't cause the same heat and power increase as on Intel chips - it was more of a checkbox "yes our CPUs can run this code too". It certainly doesn't hurt to use 26.6 on them too though.

 

[CompuTronix]

(1) "This includes 2nd through 7th Generation i7's, i5's and i3's." Sandy Bridge is 2nd Generation, so yes, from Sandy Bridge forward.

(2) 7th Generation has an AVX offset adjustment in BIOS, which is designed to reduce clock speed when AVX code is encountered, so that Vcore and temperatures don't spike, and stability is maintained so as to prevent blue screen crashes. AVX offset can be "Auto" or manually set to reduce the multiplier by increments of 100 MHz.

 

[1405]

Thanks guys. I think it is best if I just stay with 26.6 for every processor I test from now on. Although technically, the latest version should be able to work fine on 7th gen Intels and AMDs, right?

 

[Ecky]

It will run fine, but it will also use instructions that are not generally used in normal desktop loads, and thus makes a poor stability tester.

 

[CompuTronix]

Thanks guys. I think it is best if I just stay with 26.6 for every processor I test from now on. Although technically, the latest version should be able to work fine on 7th gen Intels and AMDs, right?

Correct. Intel tests their processors under carefully controlled conditions at 100% TDP. Prime95 Version 26.6 Small FFT's is the standard for CPU thermal testing, because it's a steady-state 100% workload which provides steady-state Core temperatures, and runs Core 2 processors and Core i variants with Hyperthreading typically within +/- a few % of TDP at stock settings. No other utility so closely replicates Intel's test conditions. This is also the utility that Real Temp uses to test Core temperature sensors.

CT

 

[1405]

...Intel tests their processors under carefully controlled conditions at 100% TDP. Prime95 Version 26.6 Small FFT's is the standard for CPU thermal testing, because it's a steady-state 100% workload which provides steady-state Core temperatures, and runs Core 2 processors and Core i variants with Hyperthreading typically within +/- a few % of TDP at stock settings. No other utility so closely replicates Intel's test conditions...

CT

Not even Intel Burn Test?

 

[CompuTronix]

Intel Burn Test was not written by Intel; it was written by someone who arrogantly calls himself "AgentGOD". Further, IBT is not a steady-state workload, nor is it a 100% workload; it's a repackaged version of Linpack, which forces your processor to intermittently cycle through 120% workloads, which is beyond Prime95 version 26.6 Small FFT's. While IBT is OK for stability testing, it is not accurate, realistic or suitable for thermal testing.

Before you ask yet another question, in my first post I gave you a link to the Intel Temperature Guide. The question you just asked is explained in detail in Section 13 - Thermal Testing at 100% workload - complete with images so you can visualize and better comprehend the explanations given. I ask that you please read it.

Thank you,

CT

 

[mrobscura]

i havent ran the newest prime on my 7700k but i have ran occt and while realbench and xtu are stable for hours at 4.8ghz/1.28v and a max temp of 84 or even 4.9ghz/1.29v and a max of 88, occt will not run stable at anything past 4.7ghz/1.28v and temps approach 90c. 4.8ghz fails at even 1.33v and temps get crazy by then... with a h115i.

compu, can you take a look at my thread ("7700k bad chip?") and weigh in on my issue? id appreciate it.