PC Processors

[shinkojiro]

https://www.hwinfo.com/forum/threads/explaining-the-amd-ryzen-power-reporting-deviation-metric-in-hw... 

This read by the stilt explains some boards have been fooling the processor into believing its running at a lower power than it really is. By doing so, the processor will use more power than the default PPT limit of 88w for 65w processors, causing them to run hotter than one would expect from 88w.

The new Beta of HW info shows power deviation % now. You only need to worry about what it reads under a full load. Higher than 100% means your power reporting is being over reported, which will hinder performance, and lower than that means it is being under reported, which would improve performance at the cost of more heat.

Not every board/processor combo is out of the ~100% deviation. I know Mine isnt

It would be a good idea, if you are experiencing a hot 65w processor like 90C during cenebench or similar, to check this out for yourself.

[qbtheslayer]

I just checked my ASUS X570 TUF... And with my BIOS set on "Auto" for everything CPU related it was at 100.6%.  So my board is being a good boy!!  But with my manual PBO settings this is what I get:

So my take away is that boards that don't report 100% under load are basically applying a form of PBO as a default/Auto setting and that is not very good.  I am glad to know that if I should choose to stop using my settings that the X570 TUF  actually does things right with regards to Power Reporting...  although I still think ASUS have pushed the Voltages too high (but that is another discussion).

For now I am happy with my manual PBO combined with my Undervolt on the Vcore...  it's the perfect balance of performance and temps.

QB

[shinkojiro]

Asus has been good with auto settings not doing stupid things for the most part.

[pokester]

Many reviewers have brought up this issue with some boards. That the manufacturers in an attempt to have their boards look faster are not following reference specs. Obviously if overdone this can cause issues. Nice article. Thanks for the share!

[shinkojiro]

It actually cant be overdone to a damaging extent. They could make the processor think its drawing only 1W, but it will still be limited by FIT and thermals. a 3900x for example no matter how hard you fake it won't draw more than around 195w in a CBR20 run no matter how good your cooling is since it will stop around 1.45v, just the same as making the PPT limit 10000w. The only issues it would cause is over-powering air and some AIO coolers when one would not expect because of the unseen higher power draw.

They would have to find a way to misreport power, voltage, and temperature to the processor simultaneously to get damaging results.

It seems that somewhat prior to this HWiNfo update, most boards had corrected the issue to some extent with the latest or near-latest bios revisions. ASUS even just now has a preview bios where the reference value can be changed so you can cheat manually.

[qbtheslayer]

Some more info from Anand on the subject:

Electromigration: Why AMD Ryzen Current Boosting Won't Kill Your CPU 

QB

[shinkojiro]

Yea, Its definitely not a trick that will kill your CPU. Its has the same effect as raising the PPT limit from the default 88w for 65w processors and 142w for 95/105w processors, but the FIT voltage limits and thermal throttle thresholds remain for all non-manual clock operations. The processor isnt allowed to cook itself to death.