Sandy Bridge Safe Voltage Confusion

[soccerballtux]

Not really. My friend's E5200 @ 3.75 degraded with 1.425v (BIOS) / 1.4v (CPU-Z load), in an IP35-E. Now it only runs at 3.625Ghz stable.

pretty certain that's the fault of the ip35-e. My e2180 was 3.4ghz stable at 1.48v. Over time I started getting BSODs, I ran prime95 again and noticed it was 1.47 dipping to 1.46v at times full load-- used to be 1.48 full load no problem.

Notched it up one setting and was fine.

 

[soccerballtux]

As an Intel engineer (but not in any way am I a company spokesperson), I have to say that I appreciate the implied compliment, but the real world is not so clear. Quality and Reliability Engineering (QRE) is a difficult statistical job and Intel does the best to set limits which will ensure a long and reliable life for it's products, but even so every CPU is different. While you could do a statistical analysis and determine a median value down to a microvolt, in the real world it's unclear where any individual part would fall in that broad statistical spectrum beyond a probability.

All that said, I look at people who claim to know the maximum safe voltage for a processor who have no large dataset of information or long term analysis and I personally think these "experts" have no idea what they are talking about.

* Not speaking for Intel Corporation *

meh, we know from the past that 65nm chips can handle 1.5v 24/7, and 45nm 1.4v 24/7; so I'd be confident enough to run 1.3v 24/7 on a 32nm, simply because of node changes. I'm just guestimating in an attempt to keep the potential difference / nanometer of silicon roughly the same. Degrading happens either from electromigration, or electrons embedding in dielectric in the chip. You really can't protect against electromigration unless you minimize the current (so don't fold), but you can control how much energy the electrons have: by minimizing heat and voltage.

 

[RussianSensation]

One point of reference we can use for Intel's 32nm process are 1st generation 32nm processors such as the Core i7 980X. The maximum is 1.375V.

Therefore, 1.38V at this time sounds fairly reasonable until official information is released by Intel.

 

[pm]

Isn't this variation the reason there are different VID's for each chip?

Yes. It's similar anyway... the factors that affect reliability are similar to those that set the speed bin but are not the same. But yeah, it's all manufacturing variation.

Degrading happens either from electromigration, or electrons embedding in dielectric in the chip. You really can't protect against electromigration unless you minimize the current (so don't fold), but you can control how much energy the electrons have: by minimizing heat and voltage.

Electromigration hasn't been a problem for a long time unless a designer messes up - that more or less ceased to be a problem back around 130/90nm when the switch to filled vias and copper happened. Nowadays it's all Hot-e NMOS degradation, PMOS BTI (bias-temperature instability), and TDDB (time dependent dielectric breakdown).

I'm not posting to spread FUD (fear, uncertainty, doubt) - well not too much anyway... more just a note of caution. People can do what they want with the parts that they bought - but I don't think that the way that people on AT and other forums come up with "safe" voltage values is scientifically justified. Particularly since a lot of these same people seem to think it's a great idea to turn off the load-line calibration while boosting the voltage.

* still not acting as a spokesperson for Intel *

 

[soccerballtux]

Yes. It's similar anyway... the factors that affect reliability are similar to those that set the speed bin but are not the same. But yeah, it's all manufacturing variation.

Electromigration hasn't been a problem for a long time unless a designer messes up - that more or less ceased to be a problem back around 130/90nm when the switch to filled vias and copper happened. Nowadays it's all Hot-e NMOS degradation, PMOS BTI (bias-temperature instability), and TDDB (time dependent dielectric breakdown).

I'm not posting to spread FUD (fear, uncertainty, doubt) - well not too much anyway... more just a note of caution. People can do what they want with the parts that they bought - but I don't think that the way that people on AT and other forums come up with "safe" voltage values is scientifically justified. Particularly since a lot of these same people seem to think it's a great idea to turn off the load-line calibration while boosting the voltage.

* still not acting as a spokesperson for Intel *

both BTI and TDDB are because of (either in part with BTI or hole with TDDB) electrons tunneling and embedding in the dielectric. The more I've been thinking about this the more I'm convinced if you run your chip at low voltage all day and night (one of the reasons I like CoolnQuiet more than speedstep is that you can control the voltage directly on ALL AMD's processors) then you don't need to worry about the few hours that you're running at higher voltages. Now that said, I could get to 3.8ghz+ if my cooler were capable of holding my chip below 55C. The second I hit 56C I reboot or prime fails.