The Platform Controller Hub (PCH) is the single-chip Intel chipset that connects the processor to most of the system’s input/output devices, including USB ports, SATA drives, additional PCIe lanes, networking, audio and the firmware flash. On modern motherboards it sits beside the CPU socket and communicates with the processor over a dedicated DMI link, making it a central component of PC hardware platform design.
Intel introduced the PCH in 2009 with the 5 Series chipsets, when the memory controller and graphics moved into the processor. The older two-chip design, a northbridge plus a southbridge, collapsed into one hub, and the name PCH has been used for Intel desktop and mobile chipsets ever since.
How the Platform Controller Hub connects to the CPU
The PCH connects to the CPU through Direct Media Interface (DMI), a point-to-point link that is electrically similar to a PCIe x4 or x8 connection. Everything attached to the PCH, from a second NVMe drive to a USB keyboard, shares this one uplink to the processor and system memory.
The most demanding devices bypass the PCH entirely. The primary PCIe x16 slot for a graphics card, the first one or two M.2 slots and the memory channels connect directly to the CPU. The PCH handles everything else, acting as a fan-out switch that turns limited DMI bandwidth into many slower ports.
The table below shows how DMI bandwidth has grown across generations.
| DMI version | Electrical equivalent | Approximate bandwidth per direction | Typical era |
|---|---|---|---|
| DMI 2.0 | PCIe 2.0 x4 | About 2 GB/s | 2011 to 2015 |
| DMI 3.0 | PCIe 3.0 x4 | About 3.9 GB/s | 2015 to 2020 |
| DMI 4.0 x4 | PCIe 4.0 x4 | About 7.9 GB/s | Mainstream since 2021 |
| DMI 4.0 x8 | PCIe 4.0 x8 | About 15.8 GB/s | High-end since 2021 |
What the PCH provides
The PCH integrates a broad set of controllers and system functions that would otherwise need separate chips. The exact mix depends on the chipset model, which is why a Z-series board offers more ports than a B-series or H-series board.
- Chipset PCIe lanes: commonly 12 to 28 lanes for extra M.2 slots, expansion slots and onboard devices.
- USB controllers: USB 2.0, USB 3.2 Gen 1 and Gen 2, and Gen 2×2 on higher tiers.
- SATA ports: typically four to eight, often with RAID support.
- Networking: an integrated Ethernet MAC paired with an external PHY, and a CNVi interface for Wi-Fi modules.
- Audio: the HD Audio link to an onboard audio codec.
- Firmware and management: the SPI flash interface for UEFI firmware and the Intel Management Engine.
Many PCH lanes are delivered through flexible I/O (HSIO) lanes, physical lanes that can be configured as PCIe, SATA or USB 3 by the motherboard designer. This flexibility explains why installing a drive in one M.2 slot can disable certain SATA ports: both functions share the same high-speed lane.
PCH compared with related chipset designs
The PCH is best understood as the successor to the southbridge and as Intel’s specific form of the general chipset concept. The following comparison places it among older and competing designs.
| Design | Components | Memory controller location | CPU link |
|---|---|---|---|
| Northbridge and southbridge | Two chips | Northbridge | Front-side bus |
| Intel PCH | One chip | CPU | DMI |
| AMD chipset (AM4, AM5) | One or two chips | CPU | PCIe x4 link |
| Mobile SoC with integrated PCH | PCH die in CPU package | CPU | On-package interconnect |
AMD uses the plain term chipset rather than PCH, but the role is similar. One notable difference is that AMD processors expose more I/O directly from the CPU, such as USB and NVMe lanes, so the chipset carries a smaller share. In many Intel laptop processors, the PCH has moved into the CPU package as a separate die, although it still exists logically.
Choosing a board by PCH tier
For buyers, the PCH model determines how much expansion and tuning a motherboard can offer. Intel segments its desktop chipsets into tiers, and the differences matter more than the shared socket suggests.
| Tier | CPU overclocking | Chipset PCIe lanes | Best suited for |
|---|---|---|---|
| Z-series | Supported on unlocked CPUs | Highest, up to about 28 | Enthusiasts and many-drive builds |
| B-series | Not supported, memory tuning allowed | Moderate, about 10 to 14 | Mainstream gaming and work PCs |
| H-series | Not supported | Lowest | Budget and office systems |
| W-series and Q-series | Varies | Moderate to high | Workstations and managed business fleets |
A practical rule is to count the devices that will hang off the PCH: extra NVMe drives, SATA drives, capture cards and fast USB peripherals. If they exceed the chipset’s lane count or would regularly run at the same time, a higher tier, or a platform with more CPU lanes, is the safer choice.
Bottlenecks, heat and common misconceptions
The main limitation of the PCH is the shared DMI uplink. Two fast NVMe drives behind the chipset, a 10 Gb/s network adapter and several USB devices can together exceed about 7.9 GB/s on DMI 4.0 x4, so simultaneous transfers slow each other down. Individual benchmarks rarely reveal this, because they usually test one device at a time.
The PCH also produces heat, typically 6 to 15 watts on desktop parts. Most boards use a passive heatsink, but some earlier high-end chipsets required a small fan, which became a common source of noise and failure.
| Misconception | Reality |
|---|---|
| The PCH affects gaming frame rates | The GPU usually connects directly to the CPU |
| All M.2 slots are equal | Chipset slots share the DMI uplink |
| Chipset lanes add to CPU lanes freely | They are multiplexed over one narrower link |
| A newer PCH always means a new socket | Socket and chipset generations often differ |
The simple explanation
Think of the processor as a city center and the Platform Controller Hub as a large highway interchange on the edge of town. Most local roads, such as USB ports, extra storage drives and the network jack, connect to the interchange rather than directly to the center.
A single highway links the interchange to the city center. That works well most of the time, but if many cars arrive at once, traffic slows down. This is why the fastest devices, such as the graphics card, get their own direct road to the processor.
How it is used in practice
A video editor builds a workstation with a Z-series motherboard and plans four NVMe drives: one for the operating system, one for active projects, one for cache files and one for archives. Only one M.2 slot connects to the CPU; the other three are wired to the PCH.
During a test, copying footage from the archive drive to the project drive while the editing software writes cache files produces combined throughput of about 7.5 GB/s, which saturates the DMI 4.0 x4 link. To avoid the bottleneck, the editor moves the project drive to the CPU-attached slot and keeps the cache and archive drives behind the chipset, where they rarely run at full speed together.
The result is consistent timeline playback with no stalls during background copies, achieved without buying additional hardware, simply by understanding which slots share the PCH uplink.