AMD 45nm median voltage limits:

[BeepBeep2]

Some theoretical voltage limits:


Thuban (6 core) *
- 24/7 voltage = 1.475-1.52v
- benchmarking voltage = 1.55v
- suicide voltage = 1.575v
- unsafe voltage, will degrade = 1.6v+

Deneb (4 core)*
- 24/7 voltage = 1.50-1.55v
- benchmarking voltage = 1.55-1.60v
- suicide voltage = 1.625v
- unsafe voltage, will degrade = 1.65-1.7v+

Heka (3 core)*
- 24/7 voltage = 1.50v-1.55v
- benchmarking voltage = 1.55v-1.60v
- suicide voltage = 1.625v
- unsafe voltage, will degrade = 1.65-1.7v+

Callisto (2 core)*
- 24/7 voltage = 1.50v-1.55v
- benchmarking voltage = 1.55v-1.60v
- suicide voltage = 1.625v
- unsafe voltage, will degrade = 1.65v-1.70v+

Regor (2 core Athlon II)*
- 24/7 voltage = 1.50v-1.55v
- benchmarking voltage = 1.55-1.60v
- suicide voltage = 1.60v-1.625v
- unsafe voltage, will degrade = 1.625-1.675v +

Propus (4 core Athlon II)*
- 24/7 voltage = 1.50-1.55v
- benchmarking voltage = 1.55-1.60v
- suicide voltage = 1.60-1.625v
- unsafe voltage, will degrade = 1.625-1.675v+

* Voltages moderately safe when CPU temperature is under 55-60c 100% load.

[Mavhenz]

This is lovely!

If my Deneb need 1,48 volts @ 4ghz but I read 1,45 volts in Cpu-Z (Voltage drop).

Do I consider the real voltage as 1.48 or 1,45?

1.48?

[BeepBeep2]

This is lovely!

If my Deneb need 1,48 volts @ 4ghz but I read 1,45 volts in Cpu-Z (Voltage drop).

Do I consider the real voltage as 1.48 or 1,45?

1.48?

Yep the idea just popped into my head a while ago.

Consider what CPU-Z says.

[luf1c3r]

Very good BeepBeep2, many will thank you this summary

[massman]

It sounds really weird that there's only a 0.1V delta between "24/7 safe" and "will degrade, totally unsafe"

[slaveondope]

You cant always go by what your bios or cpuz tells you.
I have a 955c3 that runs at 1.5v 4ghz on 2 boards(Biostar TA785G3 and BFI 790gx M3H5) while on a third (MSI 760GM-E51) it uses 1.42v at 4ghz.
This of course is a lie as the thermal output at 1.42c is the same as the boards running 1.5v. If I use 1.5v on the later board(MSI) for 4ghz the heat produced far excedes that of 1.5v on my other 2 boards.
Same .08v variation applies to all clocks as reported by bios and cpuz at stable voltages/clocks.

[Zeus]

Deneb 24/7 voltage 1.575V? I sincerly hope that is a typo. 1.5-1.525V sounds a lot more suitable to me. I'd say 1.5V for C3 revision, 1.525V for C2 revision.

[chris.y2k.r1]

Deneb 24/7 voltage 1.575V? I sincerly hope that is a typo. 1.5-1.525V sounds a lot more suitable to me. I'd say 1.5V for C3 revision, 1.525V for C2 revision.

Really? I was thinking the opposite but those voltages seem about right to me. My C2 gains until 1.5V then after that it is pointless... so that seems to roughly match my findings. Although I am on an M4A79T Deluxe Random Vcore Generator Edition..

[Zeus]

It depends on whether the chip tolerates the voltage or not. Some chips like voltage where others don't. From what i've seen C2 is a little more voltage intolerant beyond certain values than C3 but seemed to need a tad more.

The OP states "with good air or watercooling". Since "good" is highly debatable, it's better to put the ballpark number a tad or two lower. I for one never go over 1.48V for 24/7 operation, despite good watercooling.

[Frogeye]

I wouldn't go any higher for 24/7 then 1.53v with good cooling. 1.47v seems like a real sweet spot for my chip. I'm now convinced that my ram is the only thing that's stopping me from 4.2ghz.

[BeepBeep2]

Deneb 24/7 voltage 1.575V? I sincerly hope that is a typo. 1.5-1.525V sounds a lot more suitable to me. I'd say 1.5V for C3 revision, 1.525V for C2 revision.

Well gotvoltage has been running is 555BE @ 4.1 on 1.625v for four months without degradation...

I've been running...1.575v for half a year.

Both are C3. It's pretty safe.

[Manicdan]

i think you need to add a little footnote with something like *these safe voltages only apply to CPUs that are kept under ~60C
since we know how temperature sensitive they are with stability

[BeepBeep2]

i think you need to add a little footnote with something like *these safe voltages only apply to CPUs that are kept under ~60C
since we know how temperature sensitive they are with stability

+1 thanks


EDIT: List tweaked, keep commenting, I'll change it.

[Manicdan]

+1 thanks


EDIT: List tweaked, keep commenting, I'll change it.

perfect, cause for me running my cpu off the stock heatsink using a resistor to run the fan at about 2000rpms, makes the safe 24/7 voltage about 1.35, lol. its about the lowest extreme cooling ever.

[Particle]

It's much more simple than that. AMD stated that 1.55V was the maximum that their 45nm SOI process should be subjected to. It hardly matters which core you've got. 1.55V should be safe provided the silicon is kept in check thermally (at or below 62C) and 1.55V is what your board is actually giving it. Load line calibration can cause major spikes, so it's worth being mindful of that. Numbers above that are just speculation. Numbers below that are just preference.

[BeepBeep2]

It's much more simple than that. AMD stated that 1.55V was the maximum that their 45nm SOI process should be subjected to. It hardly matters which core you've got. 1.55V should be safe provided the silicon is kept in check thermally (at or below 62C) and 1.55V is what your board is actually giving it. Load line calibration can cause major spikes, so it's worth being mindful of that. Numbers above that are just speculation. Numbers below that are just preference.

That was for the AM2 Phenom II 940.

I mean, yes that's true but chew has already seen degradation of Thuban around 1.52v.

That's partially why I've changed the chart so all non-Thuban CPU's are 1.55v for 24/7.

[Vapor]

I guess I shouldn't be running my Thuban with 1.62v with 23/7 load

Eh, if it dies, it dies....it's served its purpose

[crazydiamond]

nice, what mhz?

[Particle]

That was for the AM2 Phenom II 940.

That's pretty much the entire point I was making. That doesn't just apply to one core. It's valid for all cores made on that process unless the process itself changes. It might have been tweaked over time, but I haven't seen any evidence suggesting there there has been a modification causing a significant change in maximum safe voltage.

I always take claims of "degradation" with a mountain of salt. While I know the intentions are always good, informing others of where a CPU is physically damaged, it's very difficult to objectively isolate. It's just speculation.

[demonkevy666]

That's pretty much the entire point I was making. That doesn't just apply to one core. It's valid for all cores made on that process unless the process itself changes. It might have been tweaked over time, but I haven't seen any evidence suggesting there there has been a modification causing a significant change in maximum safe voltage.

I always take claims of "degradation" with a mountain of salt. While I know the intentions are always good, informing others of where a CPU is physically damaged, it's very difficult to objectively isolate. It's just speculation.

I believe a slight change in process did happen Thuban uses a low K dielectric layer now and Deneb did not.

[Leopold Butters]

I run 1.57 24/7

[Particle]

I believe a slight change in process did happen Thuban uses a low K dielectric layer now and Deneb did not.

Sure, but it would be incorrect for any of us to just assume that means less voltage is safe. We need to either establish that through testing with many samples (like our overclocking tracker threads) or get it from an authoritative source on the subject like an actual engineer. Until then, the proper course of action is to assume business is as usual with extra caution being an optional, personal preference.

In terms of how that applies here, I don't want those who don't know much about what voltages are safe to reference this information and perhaps limit their results below where they could have been because they were afraid of using a higher voltage that was probably still just fine/safe. Overclockers are already generally aware that there is risk involved, even more certainly so if they're referencing information about what voltages are "safe" to use. For this audience, I think conservatism should be a note rather than the primarily advertised information.

[FlanK3r]

maybe u can add average OC potencial for chips, good work. We can waiting for great correct from master Chew

[Chumbucket843]

there are variables that you have not taken into consideration. different lots (batches), wafers, and dies can have different characteristics. also there are inter-die variations. engineers at amd must deal with these factors. they have to have a reliable product but performance is still important. maybe they could help overclockers out some more by giving some guidelines.

[BeepBeep2]

That's pretty much the entire point I was making. That doesn't just apply to one core. It's valid for all cores made on that process unless the process itself changes. It might have been tweaked over time, but I haven't seen any evidence suggesting there there has been a modification causing a significant change in maximum safe voltage.

I always take claims of "degradation" with a mountain of salt. While I know the intentions are always good, informing others of where a CPU is physically damaged, it's very difficult to objectively isolate. It's just speculation.

Adding of the low-k dielectric layer changed things up quite a bit for Thuban.

chew* advised staying under 1.52v with Thuban because he was seeing chips degrading...

But thats all part of why I started this thread...to find out this stuff. Not to state it.