Using 32 nm chip failure data to predict safe SB voltage

[JamesWatt1]

There is a lot of speculation as to what the voltage threshold is, beyond which, you are seriously diminishing the life of your CPU.

One method I was thinking about was looking for trends in failure data for the 32 nm Nehalem chips and trying to extrapolate to how the Sandy Bridge architecture will do. I appreciate it won't be a perfect predictor, but hopefully it will give a better idea than I have now.

For folks who have had 32 nm Nehalem chips (Arrandale/Clarkdale/Gulftown) that have died after over-clocking, what voltages and temperatures were you pushing them and how long were you using them before they died? I know these chips are just a year old, but we may be able to get some idea this way.

 

[Diogenes2]

Really not a valid comparison, since SB is a new fab .. It seems to be tolerant of higher volts...

Besides, I don't think you are going to find anyone here who has killed a 32nm, except maybe aigomorla ...

You might want to pose your question atXtremeSystems ..

 

[JamesWatt1]

Thanks.

I tried to register and was denied. Then, I saw that registration was closed: http://www.xtremesystems.org/forums/showthread.php?t=266354

Do you know when they will re-open registration?

 

[smakme7757]

There have been a few people who have killed their chips at Xtremesystems. but at the same time there are people running with 1.55v and their chips still work well, but the overall consensus has been that running at high volts has been causing the chips to ever so slightly degrade.

By degrade i mean requireing slightly more voltage to reach the same overlcocks. So nothing major at the moment, but no one expects a chip rated for 1.23v to run 24/7 @ 1.40v+ to not degrade or die over a period of time.

 

[Acanthus]

I have always used the "intel spec sheet +10%" method for overvolting while OCing and ive never had a problem with a chip.

 

[aigomorla]

32nm chips are fragile.

Anyone thats told you that there not is bullshitting you up the butt.

45nm themselves were more fragile then 65nm.

22nm will be more fragile then 32nm.


45nm tollerances to my own tests were about 1.45vcore at most.
45nm represented a lot of C2Q and the i7's.

When the i7's first came out, they were actually tanks.
The C0/C1's have a safe voltage of up to 1.5vcore. The D0 revision brought that back down to 1.45

I have killed 2 gulfys.
1. was 1.47 /w 1.45VTT.. killed it from VTT.
2. 1.4 .w 1.4 VTT killed it once again because of VTT.

I think 32nm are more fragile on VTT if anything then Vcore.

So you need to be more careful on the VTT, as the memory controller is the first to pop on these cpu's.

 

[Diogenes2]

Was this killing over period of time ?

I have 2600K at 1.44 ( CPUZ ) folding 24/7 -since 1/28 ...

I don't know what VTT is - there is not a setting in BIOS for VTT ...

 

[aigomorla]

Was this killing over period of time ?

I have 2600K at 1.44 ( CPUZ ) folding 24/7 -since 1/28 ...

I don't know what VTT is - there is not a setting in BIOS for VTT ...

VTT might be listed as QPI/DRAM voltage.. but not DIMM voltage.

And first gulftown didnt last 2 weeks.
The second one lasted a month.

The third is still going strong, and the 4th i am using right now.
But i was very conservative with voltages on the last 2 rarely exceeding 1.375VTT and 1.375Vcore.

You learn your lesson after killing 2.