How do you disable AVX in Prime95
 

For stress testing I would like to disable AVX.

Read undoc.txt : it contains the necessary parameters. One thing that must be added is that you have to stop prime95 before modifying the configuration files. You do not have to exit.

Jacob

CpuSupportsAVX=0 or 1, so 0 is off and 1 is on thanks?

I also suppose also that if CpuSupportsXXX is 0 the support for XXX is off and if it is 1it is on.

So to spell it out you must add a line with CpuSupportsAVX=0 in the file local.txt.

If you look at the data provided by going to the Options / CPU menu you will see what features of your CPU are used.

If you are stress-testing your computer you should leave AVX on: otherwise you will not be using all the capabilities of your CPU. It would be like disabling the turbo charger of an engine before stress-testing it.

Jacob

Thank you for your help. I totally agree with you disabling AVX should not be done for stress testing. A person I was posting with says if your not using a AVX programs why test. I don't agree with that thinking.

There are also a lot of people in forums that if prime95 does not pass large FFTs just use small FFTs so the CPU will pass the test.

All this haphazard testing is giving Pirme95 a bad name, they blame prime95 for failing when overclocking or just testing stock and some say you don't need a 100% stable CPU depending on what you are doing?

I have been using prime95 torture testing since it came out, I still think it is better than the other stability testing programs for CPU and memory. What is the opinion here?

[QUOTE=wingman99;438659]What is the opinion here?[/QUOTE]

My personal opinion (one I think is widely shared here) is if a computer doesn't pass all the various Prime95/mprime torture tests, then the machine should not be trusted. It isn't stable. In fact, it was Prime95/mprime which was used by very diligent testers to prove earlier this year that Intel's new Skylake processors had a bug.

I run mprime on all machines I'm responsible for, doing Double Checks. If a machine starts returning bad results, it gets the "evil eye". I can't even guess how much money this has saved me in the past; when mprime starts returning bad results that machine should be retired / repaired ASAP.

For those who don't use machines for serious work, instability might be acceptable. A gaming machine, for example. Not my world.

But any of you must consider next thing: on forums like this forum ( but even on this forum) there is two type of persons: one type is: buy new PC, start it and forget it.
Those kind of persons, dont know, wont know, or dont need to know how PC works, and how to make PC better.
On the second hand other , second type of person will squeeze last Mhz from CPU, will take huge CPU coolers, will take fastest RAM ... and so on ....
There is problem when person of first kind have question, and person of second king answer it.
Since knowledge person of first type is too small and not adequate , problem arise :)

I personally build PC from parts I select, put it together and start prime95 / mprime for at least day or two. When I got stable machine, then I can do rest.

I see arguments for stress testing with and without AVX. The voltage increases when AVX is in use. The extra voltage may be what is making it stable also the turbo will be higher without AVX.

In my experience a PC will fail P95 faster with AVX/FMA enabled, due to more stress, higher voltage and higher temperatures.

If any stresstest fails in P95, than the machine is unstable and I wouldn't trust the results from that PC. You might still be able to run games and Internet browsing on the PC without it crashing. But I would try o find the issue before running any form of compute on it.

[QUOTE=VictordeHolland;438708]In my experience a PC will fail P95 faster with AVX/FMA enabled, due to more stress, higher voltage and higher temperatures.

If any stresstest fails in P95, than the machine is unstable and I wouldn't trust the results from that PC. You might still be able to run games and Internet browsing on the PC without it crashing. But I would try o find the issue before running any form of compute on it.[/QUOTE]I'm half-way in this dispute.

Naturally I prefer my machines to be reliable but I also like my algorithms to be fail-soft. In my line of work, integer factorization, the principal algorithms have that behaviour. If an ECM curve very occasionally gives the wrong result, there are many more curves to try. Any one curve has only a tiny chance of finding a curve any way and losing (say) one curve in ten thousand is quite unimportant. Likewise, a NFS siever may throw out bad relations on occasion but as long as the great majority are correct, a failure rate even as high as 0.1% is quite unimportant.

Horses for courses.

[QUOTE=xilman;438712]I'm half-way in this dispute.

Naturally I prefer my machines to be reliable but I also like my algorithms to be fail-soft. In my line of work, integer factorization, the principal algorithms have that behaviour. If an ECM curve very occasionally gives the wrong result, there are many more curves to try. Any one curve has only a tiny chance of finding a curve any way and losing (say) one curve in ten thousand is quite unimportant. Likewise, a NFS siever may throw out bad relations on occasion but as long as the great majority are correct, a failure rate even as high as 0.1% is quite unimportant.

Horses for courses.[/QUOTE]

Yes but NFS post-processing needs rock stable memory.