↓ Skip to main content

Winbond W83627THF Super I/O: Registers, LPC, and Fan Monitoring

Winbond W83627THF Super I/O: Registers, LPC, and Fan Monitoring

A motherboard can report CPU temperature, fan RPM, and supply voltages without routing every measurement through the main processor. Much of that work happens inside a dedicated hardware controller: the Super I/O chip.

The Winbond W83627THF is a useful example of this architecture. It combines several legacy peripheral controllers with a hardware-monitoring subsystem, exposing their configuration through a shared register interface.

For firmware developers and hardware engineers, the interesting part is not just what the chip monitors. It is how software reaches those registers, selects individual Logical Devices, configures the LPC-connected controller, and converts raw sensor readings into meaningful values.

This guide examines the W83627THF architecture and its register-level programming model, including configuration mode, Hardware Monitor initialization, temperature banks, fan divisors, and voltage conversion.

🧩 1. Super I/O Architecture: One Chip, Multiple Logical Devices
#

The original purpose of Super I/O was straightforward: consolidate several low-speed peripheral controllers that previously required separate chips into one package.

The Winbond W83627THF follows this approach by implementing multiple Logical Devices (LDNs). Each LDN represents a functional subsystem with its own configuration registers and hardware resources.

The relevant Logical Device assignments are:

LDN Function Description
0x00 FDC Floppy Disk Controller
0x01 PRT Parallel Printer Port
0x02 UART1 Serial Port 1
0x03 UART2 Serial Port 2
0x05 KBC Keyboard Controller
0x07 GPIO5 General-Purpose I/O
0x08 GPIO2 General-Purpose I/O
0x09 GPIO3/4 General-Purpose I/O
0x0A ACPI Power Management
0x0B Hardware Monitor Sensor monitoring

Logical Device 4 is reserved in this assignment and is not exposed as a normal functional device.

The important architectural detail is that these functions share a host-facing configuration interface while retaining separate logical register spaces.

For example, configuring UART1 and enabling the Hardware Monitor use the same Index/Data configuration mechanism. Software changes the selected LDN before accessing the registers associated with the target function.

This arrangement keeps the external interface relatively simple without requiring a separate host configuration mechanism for every internal controller.

Figure 1: Internal Super I/O Logical Devices connected through the shared LPC host interface.

The LPC interface connects the controller to the chipset, while individual logical functions provide the peripheral signals and services required by the motherboard.

🔌 2. LPC Bus: Replacing the Legacy ISA Connection
#

By the late 1990s, the ISA bus had become an unnecessarily expensive way to connect low-speed motherboard peripherals.

A typical motherboard still needed floppy controllers, keyboard interfaces, serial ports, and other legacy functions. However, these devices did not justify the pin count and routing complexity of a full ISA interface.

Intel introduced Low Pin Count (LPC) to address this problem.

LPC provides a reduced-pin communication interface between the chipset and low-speed devices such as Super I/O controllers and BIOS Flash memory.

Several characteristics define its operation:

  1. Variable wait states: Slower peripherals can extend a transaction instead of being forced to complete at the host’s preferred speed.
  2. Retry support: Supported I/O and memory transactions can be retried when a device cannot complete an operation immediately.
  3. Power-management events: Wake events and power-state transitions can be handled through the interface.
  4. Legacy I/O compatibility: Existing peripheral functions can retain familiar software interfaces despite the change in physical bus architecture.

The transition from ISA to LPC offered several practical benefits.

Reduced pin count and motherboard complexity
#

LPC uses fewer physical signals than a conventional ISA connection. This reduces routing requirements and simplifies motherboard layout.

Expanded addressable space
#

The LPC address model supports a 4 GB address space, compared with the 16 MB limitation associated with the older X-bus architecture.

This removed an earlier restriction on the addressable space available to BIOS Flash and related devices.

Compatibility with existing software
#

The operating system generally interacts with the peripheral functions through familiar interfaces. BIOS firmware initializes the relevant hardware during startup, allowing software to continue using legacy-compatible resources without needing to understand every detail of the underlying bus.

Figure 2: LPC-connected Super I/O controller with keyboard, mouse, floppy, parallel-port, and serial-port interfaces.

The I/O clock routing retains aspects of the traditional ISA-era topology. LPC changes the physical communication mechanism while preserving the peripheral functions required by existing firmware and operating systems.

That separation between bus implementation and software-visible functionality was a major reason for LPC’s adoption.

⚙️ 3. W83627THF Configuration Mode and Register Access
#

The W83627THF uses an Index/Data port mechanism to access its configuration registers.

The default configuration port pair is:

Port Function
0x4E Address / Index Port
0x4F Data Port

Software first writes a register index to the Address Port. It then reads or writes the corresponding register value through the Data Port.

This is a common register-access pattern in legacy Super I/O controllers.

Entering Configuration Mode
#

Before accessing the Extended Function configuration registers, the controller must enter Configuration Mode.

For the configuration sequence described here, write 0x87 to port 0x4E twice consecutively:

MOV DX, 04EH       ; Configuration address port
MOV AL, 87H
OUT DX, AL         ; First 87h write
OUT DX, AL         ; Second 87h write: enter Configuration Mode

The two-write sequence acts as an activation safeguard. A single accidental write is insufficient to enter the configuration interface.

Once the controller accepts the sequence, software can access the Extended Function registers.

Selecting a Logical Device
#

Register CR07 selects the Logical Device Number.

To access the Hardware Monitor, select Logical Device 0x0B:

MOV DX, 04EH
MOV AL, 07H
OUT DX, AL         ; Select Logical Device Select register CR07

MOV DX, 04FH
MOV AL, 0BH
OUT DX, AL         ; Select Logical Device B

The selected LDN determines which logical device’s configuration registers are accessed.

Selecting the Hardware Monitor does not automatically enable it. Its activation register must also be programmed.

Enabling the Hardware Monitor
#

Register CR30 is the Device Activation register. Bit 0 controls whether the selected logical device is enabled.

Write 0x01 to activate the Hardware Monitor:

MOV DX, 04EH
MOV AL, 30H
OUT DX, AL         ; Select CR30

MOV DX, 04FH
MOV AL, 01H
OUT DX, AL         ; Set activation bit 0

For more complex implementations, preserve unrelated register bits when modifying configuration values rather than assuming every register can safely be overwritten.

Exiting Configuration Mode
#

After completing configuration, write 0xAA to the configuration address port:

MOV DX, 04EH
MOV AL, 0AAH
OUT DX, AL         ; Exit Configuration Mode

The complete sequence is:

  1. Enter Configuration Mode.
  2. Select the target LDN.
  3. Read or modify the required configuration registers.
  4. Exit Configuration Mode.

The order matters. Accessing a register before selecting the intended logical device can modify an entirely different hardware function.

Alternate configuration ports
#

The HEFRAS bit, located at CR26 bit 6, can select an alternate configuration port pair.

Depending on the configuration, software can use:

  • Default: 0x4E and 0x4F
  • Alternate: 0x2E and 0x2F

This provides flexibility when the default configuration ports conflict with another device.

🌡️ 4. Hardware Monitor: Temperature, Fan, and Voltage Registers
#

The Hardware Monitor is one of the most useful subsystems in the W83627THF.

It exposes measurements from motherboard temperature sensors, fan tachometers, and voltage-monitoring inputs. Firmware and monitoring applications can use these values to evaluate system conditions and implement cooling or alarm policies.

However, reading the registers correctly requires more than knowing their addresses.

The monitoring module has its own I/O base address, initialization requirements, register banks, and conversion formulas. These details must be handled consistently.

Hardware Monitor initialization
#

The initialization procedure has four main stages:

  1. Select Logical Device B and program CR60 and CR61 with the Hardware Monitor base address.
  2. Set bit 7 of CR4D.
  3. Configure the clock input source using bits 3–2 of CR4B.
  4. Program the appropriate high and low alarm thresholds for voltage, temperature, and fan measurements.

For example, the following values configure a base address of 0x0290:

CR60 = 0x02
CR61 = 0x90

The resulting runtime Index/Data ports are:

Port Function
0x295 Hardware Monitor Address / Index Port
0x296 Hardware Monitor Data Port

The runtime register-access procedure is similar to the Super I/O configuration interface:

  1. Write the target register index to 0x295.
  2. Read or write the corresponding value through 0x296.

The distinction is important: 0x4E and 0x4F access the Super I/O configuration registers, whereas 0x295 and 0x296 access the Hardware Monitor runtime registers in this example.

Temperature sensors and register banks
#

The W83627THF organizes its Hardware Monitor registers into multiple banks.

The Bank Select register is CR4E. Bank 0 is represented by 0x80 in the bank-selection encoding described here.

The three temperature channels are distributed across different banks:

Sensor Description Bank Register
SYSTIN System Temperature 0 CR27
CPUTIN CPU Temperature 1 CR50
AUXTIN Auxiliary Temperature 2 CR50

Software must select the correct bank before reading the relevant temperature register.

For example, select Bank 1:

CR4E = 0x81

Then read the CPU temperature from CR50.

Similarly, Bank 2 uses:

CR4E = 0x82

The same register address can expose different data depending on the selected bank.

This is a common source of errors in hardware-monitoring utilities. A program that reads CR50 without first selecting the correct bank may return an auxiliary temperature instead of the CPU temperature.

When a reading appears implausible, verify the bank-selection sequence before investigating the physical sensor or its wiring.

Fan monitoring: Converting pulse counts into RPM
#

The Hardware Monitor does not store fan speed as a directly readable RPM value.

Instead, it measures tachometer pulses and stores a count. Software combines that count with a configurable divisor to calculate the approximate rotational speed.

The conversion formula is:

\[ \text{RPM} = \frac{1,350,000}{\text{Count} \times \text{Divisor}} \]

The three fan-count registers are located in Bank 0:

Register Function
CR28 Fan 1 pulse count
CR29 Fan 2 pulse count
CR2A Fan 3 pulse count

The divisor is a power of two:

\[ \text{Divisor} = 2^X \]

The three-bit exponent \(X\) is split across two registers:

  • Bits 5–4 of CR47 contain the lower two bits.
  • Bit 5 of CR5D contains the most significant bit.

Figure 3: CR47h bit definitions. Bits 5–4 contain the lower two bits of the SYSFANIN divisor.

The fields must be combined before calculating the divisor.

The following C example reconstructs the exponent:

/* Example for SYSFANIN */

unsigned int divisor_bits10;
unsigned int divisor_bit2;
unsigned int X;
unsigned int Divisor;

divisor_bits10 = (val_CR47 & 0x30) >> 4;
divisor_bit2   = (val_CR5D & 0x20) >> 5;

X = (divisor_bit2 << 2) | divisor_bits10;

Divisor = 1U << X;

The exponent ranges from 0 to 7, giving divisor values from 1 to 128.

Figure 4: CR5Dh bit definitions. Bit 5 contains the third divisor bit, while other bits control temperature and VBAT enables.

Figure 5: Fan divisor lookup table showing the relationship between the three-bit exponent and the resulting divisor.

The important detail is that the divisor encoding is bit-packed. The raw fields do not directly represent the divisor.

Software must reconstruct the three-bit exponent and calculate \(2^X\).

For example, if the reconstructed exponent is 3:

\[ \text{Divisor} = 2^3 = 8 \]

The RPM calculation then uses that value together with the measured count.

Also account for zero or invalid counter values in the monitoring code. A zero count should not be passed directly into the division formula. Depending on the controller’s measurement behavior and configuration, it may indicate a stopped fan, a disconnected tachometer signal, or an unavailable measurement.

Voltage monitoring and ADC scaling
#

Voltage monitoring requires additional conversion because the Hardware Monitor’s ADC does not necessarily measure the full supply voltage directly.

Motherboards commonly use resistor dividers to scale higher voltages into the ADC’s supported input range.

Software must reverse that scaling to recover the original rail voltage.

The CPUVCORE channel supports two detection modes.

VRM8 mode:

\[ V = \text{Reading} \times 0.016\text{ V} \]

VRM9 mode:

\[ V = (\text{Reading} \times 0.00488) + 0.69\text{ V} \]

VRM9 is the default mode described here and is intended for low-voltage CPU supply measurements.

Other voltage channels use their own conversion formulas.

+12V rail — Pin 99
#

The +12V input uses a resistor divider:

\[ V_{12V} = \text{Reg}[21h] \times 0.016 \times \frac{R_1+R_2}{R_2} \]

Recommended resistor values:

  • \(R_1 = 28\,\text{k}\Omega\)
  • \(R_2 = 10\,\text{k}\Omega\)

The resistor ratio reconstructs the voltage present before the divider.

+3.3V rail — Pin 98
#

\[ V_{3.3V} = \text{Reg}[22h] \times 0.016 \]

This channel uses the direct conversion formula specified for the input.

-12V rail — Pin 97
#

The negative voltage input requires an offset correction:

\[ V_{-12V} = \left(\text{Reg}[24h] \times 0.016 - 3.6\right) \times \frac{R_3+R_4}{R_4} + 3.6 \]

Recommended resistor values:

  • \(R_3 = 232\,\text{k}\Omega\)
  • \(R_4 = 56\,\text{k}\Omega\)

The 3.6V offset is part of the measurement circuit. Removing it from the formula or using the wrong resistor ratio produces an incorrect result.

VBAT battery voltage — Pin 74
#

The battery measurement is located in Bank 5.

First select the bank:

CR4E = 0x05

Then calculate the voltage:

\[ V_{\text{BAT}} = \text{Reg}[51h] \times 0.016 \]

The general rule is that every voltage channel must use the scaling formula appropriate to its electrical input circuit.

A monitoring application should not apply one universal multiplier to all ADC readings. The resistor-divider ratios, bias voltages, and detection modes determine the correct conversion.

🔍 5. Verifying Sensor Readings with Direct I/O Access
#

Once the Hardware Monitor is initialized, direct register inspection is a practical way to verify the implementation.

A hardware debugging utility such as RW-Everything can expose I/O port values on supported systems. This makes it possible to compare the chip’s raw register contents with the values calculated by custom monitoring software.

The basic procedure is:

  1. Write the target register index to 0x295.
  2. Read the register value through 0x296.
  3. Select another bank when the required sensor resides outside Bank 0.
  4. Read the relevant temperature, fan, or voltage register.
  5. Apply the corresponding conversion formula.

For example, use the appropriate Bank Select value, such as 0x81 or 0x82, when accessing the temperature registers described earlier.

Figure 6: Real-time Hardware Monitor values inspected directly through I/O ports using a hardware debugging utility.

The useful part of this exercise is comparing raw register values against independently calculated results.

If the values agree, the register selection, bit masks, divisor reconstruction, and voltage-scaling calculations are likely consistent.

If they do not, check the following:

  • Is the Hardware Monitor logical device enabled?
  • Was the correct runtime I/O base address programmed?
  • Are the Index and Data ports correct?
  • Is the expected register bank selected?
  • Are the fan-divisor fields assembled in the correct bit order?
  • Are the voltage-divider resistor values correct?
  • Is CPUVCORE using the expected VRM detection mode?

Bank-selection errors deserve particular attention because a valid register read can still return data from the wrong sensor.

Safe access and debugging considerations
#

Direct I/O access is useful for hardware investigation, but it should be treated as privileged hardware programming rather than ordinary application-level diagnostics.

On modern operating systems, access to I/O ports is restricted. A debugging utility may require administrator privileges, a supported driver, or a compatible operating-system configuration.

Start with read-only inspection.

Writing arbitrary values to monitoring registers can change alarm thresholds, sensor behavior, or device configuration. Modifying configuration registers without understanding their side effects can also interfere with firmware initialization and other monitoring software.

For repeatable testing, record the original register values before making changes and avoid modifying registers controlled by the BIOS or another active hardware-monitoring driver.

🛠️ 6. Super I/O in Modern Motherboard Architecture
#

The W83627THF illustrates why Super I/O became an important part of PC motherboard design.

It consolidates multiple low-speed peripheral controllers into one device, maintains compatibility with legacy interfaces, and provides hardware-monitoring facilities through a relatively compact register interface.

Its Logical Devices cover functions including:

  • Floppy disk control
  • Parallel and serial ports
  • Keyboard controller support
  • GPIO
  • ACPI power management
  • Hardware monitoring

The LPC bus provides the host connection, while the Index/Data mechanism exposes the controller’s configuration and monitoring registers.

Understanding this architecture also helps when working with newer embedded controllers. Although their internal implementations differ, many still handle similar responsibilities: keyboard scanning, fan control, temperature acquisition, GPIO management, power sequencing, and communication with the host processor.

For a practical comparison with a more integrated motherboard controller, see IT8528E Embedded Controller Firmware Architecture and Code Flow. It covers the controller’s LPC interface, peripheral modules, firmware initialization, and OEM service structure.

For the firmware configuration side, UEFI Setup Callback Tutorial: Build an Interactive AMI Aptio 4 Module provides a related look at implementing motherboard setup behavior in AMI firmware.

The next time a monitoring utility reports an unexpected fan speed or motherboard temperature, the displayed number is only the final result of several hardware operations: an I/O transaction, a register-bank selection, a raw sensor measurement, and a conversion formula. Understanding each stage makes it much easier to identify whether the problem lies in the sensor, the register configuration, or the software reading it.

Related