I have an idea!

[f688xt6]

So, I just watched the new "Which video card to buy Late 2016" episode on Vessel. And this gave me an idea for what could be interesting content.

 

Premise: What does it take to make a CPU fail? (And how much can you get from them at the absolute limit)

 

Actual content: Through either stock in CPU's in LMG or by procuring them otherwise, a logical, organized (spreadsheets?) bench test of many of the popular CPU's dating back to say LGA775 or so and push them as hard as possible under normal conditions (meaning, water cooling but no exotic stuff like Liquid Nitrogen).

 

Start with basic overclocking. Look for the thermal limits with voltage and multiplier, then once those are established try to make the CPU fail by either removing the heat management incrementally, or by pushing voltages far beyond reasonable for the given CPU.

 

The idea is to show how much abuse these things can take, and to also show that benchmarks can give you a number, but the real-world robust-ness of a CPU package compared from generation to generation.

 

No I don't expect the most current CPU's to be destroyed, because that would be expensive unless Intel and AMD were willing to allow it. And yes, I know this would be a long-term setup, or multi-part series. And it would be highly complicated. Kind of like a video version (highly simplified) of Passmark's CPU benchmarks, but pushing past those levels.

 

Basically it would be like dyno-testing a car engine until it fails (or explodes). It's fun, shows weaknesses, and can provide useful data to improve later versions.

[GDRRiley]

it sounds cool 

 

[Mi26]

It gonna be alot of hardwork 

[RadiatingLight]

sounds cool. also, some videos they work on take much more work than this, like luke's failing testing of edge battery life on laptops which has no doubt consumed at least a dozen hours.

[fjdhdhcyfhf]

I don't know what you would gain by simulating cooling failure. Iv overheated my AMD bulldozer many times while experimenting with silent cooling and she's still running. My lesson was a silent pc built in winter needs to be rebuilt in summer. The motherboard shuts down the CPU when it gets above spec temperature limits so i don't think you will find anything useful.

 

But don't let me put you off your experiments, you might find something interesting, like how fast can you cool a computer? i.e. how much temperature/second change can you mess with before the CPU dies. This is useful because my PWM fans currently go to 100% speed at 70 degrees, is this appropriate or is this sudden change of temperature messing with the linear expansivity of the metals inside the CPU die?

 

[f688xt6]

28 minutes ago, SCHISCHKA said:

I don't know what you would gain by simulating cooling failure. Iv overheated my AMD bulldozer many times while experimenting with silent cooling and she's still running. My lesson was a silent pc built in winter needs to be rebuilt in summer. The motherboard shuts down the CPU when it gets above spec temperature limits so i don't think you will find anything useful.

 

But don't let me put you off your experiments, you might find something interesting, like how fast can you cool a computer? i.e. how much temperature/second change can you mess with before the CPU dies. This is useful because my PWM fans currently go to 100% speed at 70 degrees, is this appropriate or is this sudden change of temperature messing with the linear expansivity of the metals inside the CPU die?

 

It's not only about cooling failure, but total CPU capability and durability relative to prior generations. An example would be, a few years ago I mistakenly sent an i7-870 into the 4.2 Ghz range, and managed to not only boot the system, but shut it down successfully without a hard power switch shutdown. I know it wouldn't have been stable under all loads, but it survived and is still running today. There's no way I could pull that percentage of overclock on my i7-4790k and not cause damage. Or could I? That would be the point. Not so much how to fail, but what could you really push to without lasting effects. This would of course require a failure point. Since most OS's will throttle the CPU, it may require extra measures to create a failure. Just spitballing ideas at this point though.

[fjdhdhcyfhf]

9 minutes ago, f688xt6 said:

It's not only about cooling failure, but total CPU capability and durability relative to prior generations. An example would be, a few years ago I mistakenly sent an i7-870 into the 4.2 Ghz range, and managed to not only boot the system, but shut it down successfully without a hard power switch shutdown. I know it wouldn't have been stable under all loads, but it survived and is still running today. There's no way I could pull that percentage of overclock on my i7-4790k and not cause damage. Or could I? That would be the point. Not so much how to fail, but what could you really push to without lasting effects. This would of course require a failure point. Since most OS's will throttle the CPU, it may require extra measures to create a failure. Just spitballing ideas at this point though.

 you might find something on the overclockers forum - if that forum is still active? every CPU and motherboard combo is going to get different results. my old core2duo was reported by others to reach 3.9-4GHz but i could only get mine stable at around 3.6GHz. Over the last decade iv been overclocking iv had burnt out motherboards but no CPU failures, but thats because i tend to buy cheap motherboards

[f688xt6]

1 minute ago, SCHISCHKA said:

 you might find something on the overclockers forum - if that forum is still active? every CPU and motherboard combo is going to get different results. my old core2duo was reported by others to reach 3.9-4GHz but i could only get mine stable at around 3.6GHz. Over the last decade iv been overclocking iv had burnt out motherboards but no CPU failures, but thats because i tend to buy cheap motherboards

 I've seen some of their stuff, but a Linus video, or video series is why we're all here right? Or make Luke do it.