Safe Voltages, Load vs Idle?

[Strawberry]

Hey.

If I'm running on a board with a fair amount of Vdroop (drops of 0.06 V during load are not uncommon), is the maximum safe voltage dictated by load voltage or idle voltage?

If my idle voltages are higher than 1.36 but it drops down to that during load, is it still safe for 24/7 use or are idle voltages dangerous as well?

 

[JonJ678]

Voltage at idle matters a lot less as the processor is generally cooler, and more importantly not actually doing very much.When you see something like "stay under 1.3V" it is almost always referring to the voltage displayed by cpu-z under load.

Vdroop isn't a particularly bad thing incidentally, it's just annoying when after many hours of prime the voltage dips just a little too low and the thing crashes.

 

[Corasik]

High voltage can kill a chip regardless of temperature, and load.

Too much vdroop is definatly a bad thing though as you might have far too much idle voltage to have stability under load.

Idle voltages will get more and more of an issue as the process gets smaller, there is a post on here from earlier about someone who fried one of the new 32nm processors with 1.46v or so.. and the chip didnt fry while under load.. it was fried during a virus scan... so the cpu was pretty much idle and the voltage would have been at its highest.

The heatsinks do a good job at keeping the cpu's overall temperature down, but during idle only very limited parts of the cpu are doing anything, and its very easy for hotspots to develop which may damage the cpu.

I believe intel recommends a maximum of 1.46v for 45nm processors and 1.4v for 32nm processors.

Interestingly speedstep can help out here, as during idle speedstep will cut the clockspeed and lower the voltage, but when the cpu's needed it will up the voltage and clock speed. Hence a dangerously high idle voltage can be mitigated somewhat.

 

[weird_dave]

Where is core V actually measured these days, is it external to the CPU or internal now?
I'm guessing that the voltage feedback contol for the regulator isn't being taken from the point of load inside the CPU but from the Vcore plane under the CPU and that the Vdroop is due to the connections between the power plane and the die? Maybe tomorrow I'll dig out the pdfs....

 

[Strawberry]

I saw that post about frying the 32nm, but that was during a Virus Scan Benchmark, and not an actual Virus Scan. I guess that's still quite a bit of load.

Either way though, my chip is 45nm quad. So if I set my bios settings to 1.4v, and under load it drops to 1.36 (max safe 24/7 voltage by intel), it should be fine for 24/7?

 

[JonJ678]

Very interesting post there Corasik, thank you.

That would be good Dave, I've looked through a few whitepapers but don't know enough electronics to make much progress.

 

[.walls]

but during idle only very limited parts of the cpu are doing anything, and its very easy for hotspots to develop which may damage the cpu.

Sorry... but you don't ever use 1% of your CPU, or 10% or 53%, etc. Your pc is measuring % of CPU utilisation time. Therefore these hotspots you talk of....

The rest of your post was very interesting and something that I had not considered. I expect that it is the extended "overvolting" of the non-vdroop that you get when you are not stress-testing your rig and therefore when your peecee is doing the least that it can do is going to be the highest voltage running through the CPU... perhaps it's a good idea to always have some load running on a big overclock (f@h, seti, etc )

That said however - everywhere that I have read so far (including ocuk) recommends heavily against going over 1.4V for 45 nm process chips, let alone 1.45.

 

[weird_dave]

That would be good Dave, I've looked through a few whitepapers but don't know enough electronics to make much progress.

I couldn't see a voltage feedback pin in the list, but I may have missed it, the one pdf I found wasn't too great. There is a Vcc sense pin but I'm not yet sure if that's where it gets measured by the CPU. What I really need to find is the CPU register list, hopefully that will give a little more info.

If anyone can point me in the direction....

 

[Corasik]

Well, I agree that windows is only displaying utilisation, but you know, the millions of transistors inside your CPU arnt randomly switching on and off.

For each x86 or x64 instruction that is run, different groups of transistors will be operated until the cpu comes up with the desired result. If you were to code an application which did nothing but add AX to BX and leave the result in CX, and then loop, you could easily get core utilisation to 100%, but the FPU, MMX/SSE units would be completely idle, and even most of the ALU would be inactive, with just very limited parts of the CPU actually performing "work".

Thats why testing CPU's is so difficult, you cant just say oh it works, and its 60 degrees, because you can be pretty sure that unless you run every single possible x86, x87, x86-64, SSE1/2/3/4, MMX instruction (plus any other I forgot about) then you have probably not checked that every transistor on the cpu is operating properly.

Its also why different applications heat the cpu to different temperatures even though windows might report 100% utilisation. Yes, the cpu is 100% busy, because every execution unit is in use and busy.... but not every transistor.

Hotspots on the cpu die can occur under load, or at idle, so high voltages really should be avoided. (PS, the 1.46 I quoted came from someone else on the forums here... I just assumed it was accurate )