Intel Announces New Prime95 Suckage in Skylake

Mercutio

Fatwah on Western Digital
Joined
Jan 17, 2002
Messages
22,275
Location
I am omnipresent
Now we just have to hope that all our Skylake CPUs aren't destined to turn up as paving materials likth FDIV-bug impacted P54C Pentiums were.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,926
Location
USA
They claim a BIOS fix can solve this. I'll be interested in what followup tests show.
 

Buck

Storage? I am Storage!
Joined
Feb 22, 2002
Messages
4,514
Location
Blurry.
Website
www.hlmcompany.com
I'm actually a little surprised that Intel don't test internally with Prime95 or a similar application?

I was thinking the same thing. Then I wondered, am I reading the situation correctly? Maybe I am wrong, but it seems that not every Prime95 instance produces the same results with Skylake. So, could Prime95 be running on a Skylake CPU for 24 hours, and the problem is never encountered?
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,728
Location
Horsens, Denmark
Interesting. I wonder if this explains some of my issues with doing an overclock; I use Prime95 and was having inconsistent results. Looking forward to trying again after the fix.
 

DrunkenBastard

Storage is cool
Joined
Jan 21, 2002
Messages
775
Location
on the floor
Interesting. I wonder if this explains some of my issues with doing an overclock; I use Prime95 and was having inconsistent results. Looking forward to trying again after the fix.

My understanding is that you need to be careful of what version of Prime95 you are running. Newer versions will push AVX instructions in the fpu in a way that really diverges from actual real world scenarios. IIRC version 27 or something like that is best to avoid AVX. If you are using an AVX version make sure you have adaptive/offset vcore turned off.
 

sdbardwick

Storage is cool
Joined
Mar 12, 2004
Messages
609
Location
North San Diego County
My understanding is that you need to be careful of what version of Prime95 you are running. Newer versions will push AVX instructions in the fpu in a way that really diverges from actual real world scenarios. IIRC version 27 or something like that is best to avoid AVX. If you are using an AVX version make sure you have adaptive/offset vcore turned off.
Horseshit.
AVX and AVX2 (FMA) instructions are used by Prime95 exactly as they are designed to be used. Idiots who didn't/don't understand the response of the CPU to AVX instructions spread misinformation - usually like "Prime95 caused my overclocked and overvolted CPU to die!" because the additional (and documented) voltage rise triggered by AVX instructions destabilized their CPU.
 
Top