[Tutorial] Graphics Cards Voltage Regulator Modules (VRM) Explained
2 – VRM: Theory
The Voltage Regulator Module or VRM is a device that performs DC-DC conversion (DC = Direct Current). This conversion is fundamental because many chips, like the GPU in our case, do not operate at 12V or 5V but at lower voltages like 1V. Then devices to reduce the voltage are required and these devices are the famous VRMs. So a VRM is a DC-DC converter. The other goal of a VRM is to provide a constant DC output voltage as well as providing a lot of current (amperes) to the GPU.
A VRM is also called a buck converter.
Here is the basic scheme of a VRM:
The SW1 is a transistor that commutates the input voltage, L1 is an inductor that stores and releases energy into a capacitor Cout allowing to smooth the voltage signal. SW2 is a freewheeling transistor that is closed when SW1 is opened. In low power buck converters, a diode is used in place of the transistor.
A VRM can be divided in three parts: the logic device, the power device and the filtering device.
The logic device, also called voltage controller, drives the power device in order to supply the GPU with the correct voltage and current. The voltage controller is the brain of VRM.
The power device is the VRM muscles and are made up to power MOSFET (Metal-Oxyd Semiconductor Field Effect Transistor). A MOSFET is power switching element and is one of the most common transistor type in electronics.
The filtering device allows to smooth the MOSFET output current and is made up of inductors and capacitors.
A phase is made up of a power device and a filtering device. The phase allows to simplify the description of a VRM: a VRM is made up of a voltage controller and several phases.
The following images show the VRM of Asus’s EAH 5830 DirectCU:
Asus EAH 5830 DirectCU PCB with VRM phases and voltage controller (VC) – source
Asus EAH 5830 DirectCU PCB – VRM details
Asus EAH 5830 DirectCU – voltage controller details
According to the global view of the PCB, this Radeon HD 5830 has 8-phase VRM. There are one or two phases for feeding the memory (yes, graphics memory is also fed by the VRM) and the other phases are used to feed the GPU (6 or 7 phases).
Why a VRM is made up of several phases ?
Because modern GPUs require a lot of current: up to 80A for some GPUs. Yes, dear readers, 80 amperes! So much current can only be provided by a VRM with several phases. Such kind of VRM is called a multiphase VRM or multiphase converter. Multiphase can be replaced by n-phase.
Actually all modern graphics cards come with a multiphase VRM.
In a multiphase VRM, each MOSFET (actually there are two MOSFETs per phase) is driven by a PWM signal generated by the voltage controller. The PWM command signal drives the opening and closing of these power MOSFETs.
The output of a n-phase VRM is the sum of each phase.
PWM means Pulse Width Modulation and is simply a signal with a square pulse whose the width can vary.
The following image shows the PWM signals for a 4-phase VRM:
4-phase VRM – PWM signals
The PWM principle is rather simple: the nature does not support brutal variations (of voltage, speed, whatever quantity you want) and prefers smooth ones. Just drop a rectangular pulse wave (the voltage), whose pulse width is modulated, in a filtering device that accepts only smooth signal variation and the result will be an average value of the waveform. And this average value is linked to the pulse width.
In a n-phase VRM, each couple of MOSFET delivers only 1/n of the total output power, allowing smaller MOSFET and inductors. What’s more, the power dissipation is spread over all MOSFETs, avoiding an unique hot spot that exists in a single phase VRM.
There are two kind of PWM generators: analog and digital. Quickly said, analog PWM generators use simple electronic compoments (continuous signal generator, comparator, etc.) to generate the PWM signal.
Digital PWM generators are based on a micro-controller and offer programming capabilities.
So the amount of power that a VRM can deliver is proportional to the number of phases. MSI has understood this simple principle and has equipped its high-end Radeon HD 5870 Lightning with a 15-phase VRM (12 phases for the GPU and 3 for the memory)!
The filtering device in each phase is actually a way to store (when MOSFETs are closed) and transfert (when MOSFETs are opened) the energy required by the GPU. There two main types of VRMs:
- Inductor-based VRMs
- Capacitor-based VRMs
Capacitor-based VRMs are suitable for handled devices (low power) while inductor-based VRMs are suitable in presence of high current with low voltage. In this case, the inductors are called high current inductors. These high current inductors feature a low inductance (typically 2.0 uH and below) and a high current rating (15A up to 100A).
A choke is the common name given to a power inductor used in a filter element.
Keep in mind that the GPU operates with very low voltage and high current signal (the current per phase can reach 20A or more). In this situation the inductor acts as a reservoir for the current avoiding brutal variations of current (inductors chokes are passive components that are designed to resist changes in current). The inductor stores the energy in the form of a magnetic field. This filter can be also seen as a low-pass filter for the current.