Understanding CPU voltages (Intel)

[r00tb33r]

I've got an i9-11900KF on a Z590 Master board.

 

I'm monitoring voltages in HWMonitor and CPU-Z and I see a maximum VID of 1.72V in HWMonitor, but core voltage in CPU-Z between 0.8 and 0.9V at idle.

 

While running benchmarks I observe the core voltage go up to 1.44V in CPU-Z during multi-core tests, and reach 1.668-1.680V during single threaded tests.

 

Is that too much for this chip?

[RONOTHAN##]

1 minute ago, r00tb33r said:

I'm monitoring voltages in HWMonitor and CPU-Z

Don't, neither are reliable reading voltage on modern hardware. CPU-Z is good with clock speeds, that's about it. Use HWINFO64 for this testing. 

 

And yes, 1.68V is insane for that chip, but that reading is also likely just wrong. 

[r00tb33r]

15 minutes ago, RONOTHAN## said:

Don't, neither are reliable reading voltage on modern hardware. CPU-Z is good with clock speeds, that's about it. Use HWINFO64 for this testing. 

 

And yes, 1.68V is insane for that chip, but that reading is also likely just wrong. 

Okay, got HWINFO64.  Ran the same tests.

 

I'm guessing Vcore is the number of interest, it is 0.8-0.9V at idle, ~1.4V during multi-threaded tests, and 1.650-1.661V during single-threaded tests.  1.661V being the maximum recorded.  Lowest 0.770V, ~1.0V average.

 

So...  How bad is it?

[RONOTHAN##]

2 minutes ago, r00tb33r said:

~1.4V during multi-threaded tests

That's high but not completely terrible. 

 

3 minutes ago, r00tb33r said:

1.650-1.661V during single-threaded tests.

That's a fair bit higher than I'd like it. I'd want it below 1.5V at all times for daily usage. 

 

What are your temps looking like? That's voltages are high enough that I'd be expecting it to be thermal throttling like crazy. 

[r00tb33r]

1 minute ago, RONOTHAN## said:

That's high but not completely terrible. 

 

That's a fair bit higher than I'd like it. I'd want it below 1.5V at all times for daily usage. 

 

What are your temps looking like? That's voltages are high enough that I'd be expecting it to be thermal throttling like crazy. 

Didn't throttle during benchmarks, package peaked at 94°C with hottest core at 96°C.  Yeah it got close, but didn't throttle.  It'll be lower once I get the new fans. It's idling at 32C.

 

Peak power draw on the package was 291W.

 

I'm gonna see how much I can back off the offset without completely wiping out the gains.

 

 

[RONOTHAN##]

10 minutes ago, r00tb33r said:

Didn't throttle during benchmarks, package peaked at 94°C with hottest core at 96°C.  Yeah it got close, but didn't throttle.  It'll be lower once I get the new fans. It's idling at 32C.

 

Peak power draw on the package was 291W.

 

I'm gonna see how much I can back off the offset without completely wiping out the gains.

 

 

That makes it sound like the VCore sensor is in the worst possible location. Does your board have a VRVOUT sensor reported in Hwinfo? That should be die sense if you have it and be the most reliable voltage sensor. 

[r00tb33r]

So these are the results with Adaptive Vcore set for Vcore Voltage Mode.  I completely removed manual offsets, everything is on auto.

 

That dropped Vcore by ~0.050V on average, but still momentarily peaked at 1.650V during single-threaded tests.

[r00tb33r]

7 minutes ago, RONOTHAN## said:

That makes it sound like the VCore sensor is in the worst possible location. Does your board have a VRVOUT sensor reported in Hwinfo? That should be die sense if you have it and be the most reliable voltage sensor. 

Yes, VR VOUT reading is very close to Vcore, within a few mV.

[RONOTHAN##]

22 minutes ago, r00tb33r said:

Yes, VR VOUT reading is very close to Vcore, within a few mV.

Even under load? At idle they should be pretty similar, though under load they should be a bit different. 

[r00tb33r]

With Vcore Voltage Mode set to auto the peak is down to 1.620V, that's about 20-30mV reduction across the board.  It stays well under 1.6V most of the time.

 

Setting TVB Frequency Clipping from disabled to auto didn't produce a noticeable difference.

 

I've reset the BIOS to defaults.  It would appear that wiping out the per-core ratio limits (where the "favorite" cores are normally listed) is what brought the Vcore down to 1.25V during multi-threaded tests and to ~1.40V during single-threaded tests, with a momentarily peak of 1.428V.  Peak power draw is down to 209W, and peak package temperature down to 75C.

 

It sounds like I'd have to undervolt it at same clocks I got before to keep Vcore reasonable.  It's operating at the limit out of the box it seems.

[r00tb33r]

I've hit similar benchmark results as before with -0.100V Vcore offset.  During multi-threaded tests it stayed ~1.35V, and ~1.45V during single-threaded, with a momentary peak of 1.512V.

 

I guess I should try going lower on Vcore offset.

[r00tb33r]

Lowest stable Vcore offset I tested is -0.200V.  Crashed at -0.225V.

 

Completes benchmarks with a maximum recorded Vcore of 1.397V.  I'm quite happy with this number.

 

I'm going to try raising the boost ratios table more.  We're already reaching 5.4GHz I was peaking at with higher voltages.

[r00tb33r]

I was crashing on Blender Benchmark so had to raise Vcore offset to -0.150V and go less radical on the multi-core boost table.  I was able to regain stability without losing benchmark scores.

 

Having run two types of benchmarks the highest recorded Vcore right now is 1.368V, which I think is excellent.

 

  

2 hours ago, r00tb33r said:

It sounds like I'd have to undervolt it at same clocks I got before to keep Vcore reasonable.  It's operating at the limit out of the box it seems.

Or rather offset Gigabyte's overvolt.

 

This is another instance for the subject of board manufacturers significantly overvolting processors out of the box without users explicitly telling the board to do so.