7700k Prime 95

[chris cooper]

Hi Guys,

 

Been tweaking my overclock.  I am currently running 4.9ghz at 1.3v on my 7700k.  

 

If I use IBT the system stable, Real Bench the system stable, cinebench the system is again stable.  Gaming and productivity wise, I haven't seen any crashes, lock ups, or instability with the above settings, until I test with Prime 95.

 

So, if I use Prime 95 version 26.6 the system is stable for long runs and temps well around 70 degress under 100% load, however as soon as I start testing a on Prime 95 version 29.3, I have several cores fail instantly and hwinfo reporting temps of well over 85 degrees.

 

Any ideas what is going on here, would you consider the system stable?  Why does a newer version of prime cause issues, should I be gauging the "health" of my overclock on version 26.6 or 29.3?

 

I am lost here.

 

Cheers

 

Chris.

 

[just_dave]

Newer Prime is AVX version, it stresses Intel CPUs to extreme levels.

I wouldnt use it for stress testing, just something like Aida64

[chris cooper]

Quote

4 minutes ago, dave_k said:

Newer Prime is AVX version, it stresses Intel CPUs to extreme levels.

I wouldnt use it for stress testing, just something like Aida64

 

Thanks for your reply man...

 

I thought it would have something to do with AVX.  So what are the benefits of testing with this AVX or not.

 

My concern here, is although my rig is stable with everything I can throw at it except the newer version of prime.  Am I going to get down the line something that uses the AVX extension cause my rig to lock up.

 

Should, I ultimately be looking to get it stable, with a stress test that uses the AVX extensions.

 

To be fair I had only used Realbench, Aida and IBT to test my overclocks until now.  But I dont like there being some "underlying" instability.  If that makes sense.

 

[Valkyrie Lenneth]

15 minutes ago, dave_k said:

Newer Prime is AVX version, it stresses Intel CPUs to extreme levels.

I wouldnt use it for stress testing, just something like Aida64

prime not stable = pc not stable, stock is also stable on prime and all it does is calculations really...

 

aida64 is not reliable for any proper testing because it passes through aida64 quite easily, yet get bsods when gaming after once ignored a prime error, never again

[DocSwag]

8 minutes ago, chris cooper said:

Thanks for your reply man...

 

I thought it would have something to do with AVX.  So what are the benefits of testing with this AVX or not.

 

My concern here, is although my rig is stable with everything I can throw at it except the newer version of prime.  Am I going to get down the line something that uses the AVX extension cause my rig to lock up.

 

Should, I ultimately be looking to get it stable, with a stress test that uses the AVX extensions.

 

To be fair I had only used Realbench, Aida and IBT to test my overclocks until now.  But I dont like there being some "underlying" instability.  If that makes sense.

 

Benefits of testing with AVX:

-you'll know if it's stable under AVX

 

Disadvantages:

-An OC stable in workloads you normally encounter won't be stable under AVX

 

You could try seeing if your mobo has an AVX offset. Some mobos let you make it so that your CPU will run AVX workloads at a slightly lower frequency.

 

You don't usually encounter AVX much though, so I think something like realbench will be fine.

1 minute ago, Valkyrie Lenneth said:

prime not stable = pc not stable, stock is also stable on prime and all it does is calculations really...

 

aida64 is not reliable for any proper testing because it passes through aida64 quite easily, yet get bsods when gaming after once ignored a prime error, never again

IMO Prime is too extreme, realbench is better.

[Valkyrie Lenneth]

9 minutes ago, chris cooper said:

Thanks for your reply man...

 

I thought it would have something to do with AVX.  So what are the benefits of testing with this AVX or not.

 

My concern here, is although my rig is stable with everything I can throw at it except the newer version of prime.  Am I going to get down the line something that uses the AVX extension cause my rig to lock up.

 

Should, I ultimately be looking to get it stable, with a stress test that uses the AVX extensions.

 

To be fair I had only used Realbench, Aida and IBT to test my overclocks until now.  But I dont like there being some "underlying" instability.  If that makes sense.

 

up to 90*C is safe with 1.3v really , ive ran chips on 1.35v with a 24/7 load to 85-90*C and the chips never degraded or anything

 

i dont recommend it stressing it without AVX because windows uses AVX by default aswell if u get it stable without avx that would mean the chip might not be fully stable after all if u dont want any underlying instability make sure to run a 12hour minimum small fft prime95 run avx enabled ,  and a custom prime95 run filling up all ur ram , disable all programs and check for ur remaining free ram, fill prime95 in to use that amount of memory and stress atleast 24hours

 

aswell 3passes in memtest89 and if u wanna make sure its 12hours+

 

^ if u dont really do that its not guaranteed to be stable coz prime can still crash after 15-17hours of stressing ( happened quite alot )

[Valkyrie Lenneth]

2 minutes ago, DocSwag said:

Benefits of testing with AVX:

-you'll know if it's stable under AVX

 

Disadvantages:

-An OC stable in workloads you normally encounter won't be stable under AVX

 

You could try seeing if your mobo has an AVX offset. Some mobos let you make it so that your CPU will run AVX workloads at a slightly lower frequency.

 

You don't usually encounter AVX much though, so I think something like realbench will be fine.

IMO Prime is too extreme, realbench is better.

prime95 = reliable, other things so far have not been reliable at all for me, all of them stable > one error in prime... decided to ignore it coz it drove me nuts, in the end cpu was not stable   if prime95 crashes its just not stable and thats a fact, coz prime just does calculations and all stock chips are stable on prime95 ^^

[chris cooper]

Some intresting comments here.  And defo food for thought.

 

So i did a very quick run with the latest version of Prime with an AVX offset of 3 and it didn't crash instantly like it was before...  Maybe that has given me some more headroom... 

 

I'll keep ya posted.

 

[porina]

12 minutes ago, DocSwag said:

You could try seeing if your mobo has an AVX offset. Some mobos let you make it so that your CPU will run AVX workloads at a slightly lower frequency.

Agreed. Look for AVX offset if bios offers it. It might also show up in XTU if the system supports it at all.

 

On most modern Intel CPUs, recent versions of Prime95 are a powerful stress test, but even then I wouldn't rely on it by itself. On AVX-weak systems (like non-k OC, or Ryzen) the stress isn't that bad and other things might provoke instability faster. When I OC'd i3-6100, I found realbench threw up errors faster than P95. On R7 1700, I initially verified using P95 at 3.6 GHz 1.20v. I can run that as long as you like without error. Other things didn't agree and I got random reboots without applying a bit more voltage.

[DocSwag]

57 minutes ago, Valkyrie Lenneth said:

prime95 = reliable, other things so far have not been reliable at all for me, all of them stable > one error in prime... decided to ignore it coz it drove me nuts, in the end cpu was not stable   if prime95 crashes its just not stable and thats a fact, coz prime just does calculations and all stock chips are stable on prime95 ^^

Realbench has worked fine for me.

 

shrug