Cpu pll oc voltage, Cpu-gpu temps and Power consumption(couple of questions)

[tempwatch]

Hello. I have couple of questions. Im really curious about this.

1- Why cpu pll oc voltage affects cpu temps weirdly. People generally says it skews temp readings. I checked thermal throttling and performance limit reasons tab on the hwinfo. Its always throttling at 99C doesnt matter cpu pll oc voltage. But 1.2v pll oc gives much lower temps than 1.3v pll oc. Mobo automatically sets 1.3v if i overclock or enable xmp feature of my ram. However im using default 1.2v always.

Cpu pll oc voltage doesnt affects any stability.

Prime 95 1.17v: hottest core 72C 
Prime 95 1.2v: hottest core 75C
Prime 95 1.25v: hottest core 80C
Prime 95 1.3v: hottest core 85C


Looks like temp reading offset but some cores gains more temps than others. No package power difference. Which temp is true ? 1.2v 98C works without throttling. So 1.2v should be true reading ?


Mobo: Msi z270 tomahawk arctic
Cpu: 7700k
Gpu: Vega 64 LC




2- Also i have 2 cpu package readings. One of them is on the 7700k dts tab, other one 7700k enhanced tab. Both reports little different temps.

5min recording:
Cpu pack dts:  min35 max71 avg39
Cpu pack enhanced: min35 max72 avg39 (both readings averaging same temp after couple of minute but they are always reporting different temps at the same time)
Peci: min35 max71 avg 38.6
Core max: min33 max71 avg37

Delay 0 ms.

What is core max temp btw. Which one is more reliable. I guess individual core temps more reliable but i want just one number.



3- Do you have any idea about Vega hotspot temp ? After repasting and repadding my temps improved but hotspot temp worse now.


4- Watt readings are reliable ? Cpu package power for cpu and gpu core power, gpu memory power, gpu chip power for gpu. Also i have Corsair rm1000i so i can see power consumption with hwinfo. How much reliable are these readings ? Cpu and gpu readings showing total 15-20w consumption at idle but my psu readings 60W at idle. 50-55w coming from 12v rail.  10W from 5v and 3.3v rails.

 

[Martin]

1. Sorry, I can't help with that much, you might get a better answer on dedicated OC forums like overclock.net or from the mainboard manufacturer.
2. Such differences are normal. Since the temperatures can fluctuate very quickly, they can vary depending on time of reading. The Package Temperature is s 256ms averaged value (internally by the CPU) of the maximum readout among all cores. Core Max temperature is the maximum temperature among all cores at the time of reading.
3. Sorry, I don't have more details about this value yet.
4. Power reading should be reliable, but it depends on particular component. Most of them are also utilized internally by the CPU/GPU to manage power management. But there are cases, which might affect this - for example disabling SVID support in BIOS will render the CPU power value invalid. Also note, that certain power measurements might not include whole component power. For example a GPU can measure the power of the core and memory rails, but other rails are not capable of measuring the power, they can be just estimated. This has been discussed several times on this forum.

 

[tempwatch]

Thanks for answers. Helped a lot but i have more questions

1- How throttling and performance limit reasons works ? I'm trying to estimate which value gives correct temp reading but when i reach 99C thermal throttling becomes yes also thermal event turns ''yes'' too. So i assume thermal throttling reading just checks straight temp reading. There is no algorithm behind that. Are there any feature can i check my cpu throttles because of temps ?

4- I checked power readings under load. Cpu + gpu core + gpu memory + gpu chip almost equal to my psu readings. I guess gpu chip power is wrong as you said its just estimated. So 7700k at idle should consuming 5w with balanced power plan, 7w with high performance power plan. Vega 64 LC at idle consuming 35-45W. Psu readings more reliable. If i look at the gpu power readings it should 7-8W at idle which is very low for gpu.

 

[Martin]

Of course thermal throttling depends only on temperature.
In fact, GPU Chip Power should be more accurate. This value is calculated internally by the GPU and should cover all rails - GPU Core, Memory (measured via internal/VRM telemetry) and additional minor rails (either a constant estimation or based on activity counters).