Some Motherboard Manufacturers Apparently Tweak Power Measurements to Boost AMD CPUs

[Guest_Jim_*]

The newest HWiNFO beta has added a metric to its collection of measurements called "Power Reporting Deviation" that can reveal whether your motherboard manufacturer has tweaked its BIOS to boost AMD CPUs. As explained by The Stilt on the HWiNFO forums, modern AMD APUs base some of its measurements on values reported by the motherboard, including the current it is drawing. The product of current and voltage is the power consumed by the CPU and factors into the boosting algorithm, so by making the processor think the current is lower than it actually is, it will boost more, and some manufacturers are doing exactly that.

The way this works is the VRM controller on the motherboard is to report an integer between 0 and 255, but this value is on an undefined scale. To determine what the actual current is, a look-up table is used to find the value, but if the motherboard has a bias applied, the value returned can be significantly lower. This will then trick the AMD CPU into thinking it is operating under its PPT, package power tracking limiter and so can push itself a bit farther.

The "Power Reporting Deviation" measurement is to identify what the bias is and can be tested by having the latest HWiNFO beta (v6.27-4185) open and running a stable load that will take the CPU to its limits, such as CineBench R20, multi-thread. If the measurement returns a value of 60%, per the example The Stilt uses, then instead of a Ryzen 7 3700X having a PPT of 88 W, the actual power limit will be 147 W, or 88 W / 0.6.

The reason a motherboard manufacturer would do this is fairly obvious and hardly uncommon either; it is a way to boost performance compared to competitors, without it being apparent. The problem is this bias is something likely buried in the BIOS code where the user cannot affect it, though there are some exceptions. The Stilt states there are certain MSI motherboards that do allow it to be controlled, but for those that do not allow control and the bias is clear, he suggests contacting the manufacturer to remove the bias. He also states this kind of alteration to the BIOS is something AMD does not desire.

This discovery has prompted many to voice concerns that this allowance for greater power draws could ultimately harm the CPU, which has prompted Dr. Ian Cutress of Anandtech to respond. His piece also covers some of the previously discovered methods motherboard manufacturers have used to boost Intel CPUs, which provides some context for the situation. He also states very clearly that no, our AMD CPUs are not in danger, at least not for regular users. The short version is the possible sources of damage to a CPU already have mechanisms in place to protect them, so as long as your cooling is adequate, there should not be a problem during the expected life-time of the part.

Getting a bit more into it, Dr. Cutress talks about electromigration, which is the most likely way a CPU will break, outside of physical damage or disabling thermal limits. Within a CPU you have the various wires carrying electrons from one point to another. These electrons are moving and have some momentum to them, and while it happens rarely, they will occasionally strike an atom that comprises the wire. This impact will move the atom from its place in the crystal lattice and impact the cross-section of the wire, reducing it at the point of impact. Reducing cross-sectional area increases the resistance along the connection, which will require more voltage. The thing is, electromigration is impacted by factors such as voltage, temperature and current, so we can enter into a feedback loop that shortens the life of the process. However, the changes in voltage that come with this bias are not necessarily that great.

With more current being allowed by these biases, and thus more electrons moving and potentially striking atoms, this would appear to be a bad situation for avoiding electromigration. However, as Dr. Cutress continues to explain, manufacturers have been working on the electromigration problem for a while now. There are active countermeasures in place to reduce the amount of electromigration, or to reduce the impacts of it, and it also becomes less of an issue as process nodes and voltages shrink. To further support how unlikely electromigration is to be an issue, he describes the one time it actually significantly damaged a processor of his. It was an Intel Core i7-2600K and he used it for overclocking competitions, as it would reach 5.1 GHz under extreme cooling. After a couple years, it needed more voltage to run at stock, but it was also spending those years being pushed to its limits.

In the end, it looks like some motherboard manufacturers are trying to give themselves advantages over their competition by tricking AMD CPUs into push their current draw limits. Outside of overclocking and assuming adequate cooling is applied, this should not be an issue for most. What could be more an issue is not being able to trust the values reported by software, if this is something you monitor.

Source: HWiNFO Forums and Anandtech

Back to original news post