Why Does a Computer Need a Chipset When the CPU Already Does So Much?
Or: why the most powerful processor in the world still needs help running a motherboard. The CPU does the computing, the chipset multiplies the connections, and that's why adding one more SSD can switch off a SATA port.
Or: Why the Most Powerful Processor in the World Still Needs Help Running a Motherboard
Open the specifications for a modern desktop processor and you'll find an impressive list. Multiple cores, large caches, memory controllers, PCI Express interfaces, and in many cases integrated graphics. Some add specialized hardware for AI, security and media processing.
With so much packed into one chip, an obvious question emerges: why does a computer still need a motherboard chipset? If the CPU can talk directly to memory, graphics cards and fast storage, what is the chipset doing?
The answer is that the CPU performs the computer's most important calculations, while the chipset helps connect and coordinate many of the other devices that make the computer useful. Processors have absorbed functions that once required several separate chips, but they still face practical limits in electrical connections, manufacturing cost, bandwidth and motherboard design. A computer needs somewhere to plug in storage drives, USB peripherals, networking hardware, audio and expansion cards, and the chipset provides that connectivity without giving every device its own dedicated line to the processor.
Understanding this explains why two motherboards with the same CPU can offer dramatically different features, and why installing an extra SSD can disable a SATA port or change how an expansion slot works.
How We Got Here: North and South
Early PCs used numerous separate chips for jobs that modern systems combine. The processor handled general computation, while other chips managed memory access, peripheral communication, interrupts and expansion hardware. As systems grew more powerful, manufacturers grouped those supporting functions into chipsets, and for many years PC designs were described with two parts: the northbridge and the southbridge. The names came from their positions in motherboard block diagrams, but also reflected their jobs.
Per Wikipedia's description, the northbridge was connected directly to the CPU via the front-side bus and typically handled communication among the CPU, sometimes RAM, PCI Express or AGP video cards, and the southbridge. The southbridge was the slower chip, handling communication with other motherboard parts. That arrangement defined PC motherboards for years. It also added communication steps: when the processor needed data from RAM, it might have to go through the northbridge.
As processors got faster, latency and memory bandwidth mattered more, and the industry moved northbridge functions into the processor. Wikipedia notes that AMD put the memory controller on the processor die starting with its K8 processors and Intel did so with Nehalem, and that Intel's Sandy Bridge in 2011 introduced a system agent that replaced the northbridge. Intel's southbridge was renamed the Platform Controller Hub, and AMD's equivalent is the Fusion controller hub.
The memory controller is the clearest case. It coordinates memory commands, timing and data transfers, and in mainstream desktop CPUs it now sits inside the processor, giving dedicated memory channels. That means supported memory technology depends on the CPU. A motherboard can't add DDR5 support to a processor built only for DDR4 just by using a different chipset, because the controller that has to support it is in the processor.
What the CPU Connects Directly
Processors also gained direct high-speed links for graphics cards and NVMe SSDs, using PCI Express, or PCIe. A PCIe lane is a full-duplex connection with separate transmit and receive paths, allowing information to travel in both directions simultaneously. Multiple lanes can be combined into wider links: x1 uses one lane, x4 uses four, and x16 uses sixteen. More lanes generally provide more potential bandwidth, while newer PCIe generations increase the transfer rate per lane. Synopsys explains that PCIe 5.0 doubles the signaling rate of PCIe 4.0, from 16 to 32 gigatransfers per second per lane. A PCIe 5.0 x16 connection offers approximately 63 gigabytes per second of theoretical bandwidth in each direction, or roughly 126 gigabytes per second when both directions are combined, before additional protocol overhead. Four PCIe 5.0 lanes therefore provide approximately the same theoretical bandwidth as eight PCIe 4.0 lanes.
A graphics card might use sixteen processor lanes and a primary NVMe SSD four more. Those devices get high-bandwidth paths without their traffic passing through the chipset. But the CPU can't offer a separate direct connection for every device. TechPowerUp's explainer on AMD's AM5 platform, which republishes AMD's material, says the Zen 4 processor has 28 PCIe 5.0 lanes: sixteen for a graphics slot, eight general-purpose lanes (at least four for an M.2 slot, the rest up to the motherboard maker), and four to link to the chipset. Intel's Z790 brief lists processor lane configurations of 1x16+1x4 or 2x8+1x4. The number of physical slots on a board tells you very little about how many processor lanes exist.
Adding more interfaces to the processor would raise pin counts, design complexity and cost. So motherboards connect many other devices through the chipset, which effectively expands the platform's connectivity.
The Chipset as an I/O Hub
I/O means input and output, the exchange of information between the processor and other devices. A chipset collects an assortment of interfaces, such as PCIe, USB, SATA and other platform functions, behind a single upstream link to the CPU. A keyboard doesn't need the same connection as a high-end graphics card, and neither do a SATA SSD or a printer. The chipset matches connections to needs.
Intel's brief for the Z790 gives a sense of scale: up to 20 PCIe 4.0 lanes and 8 PCIe 3.0 lanes, up to 8 SATA ports, and a long list of USB ports across several speed grades (up to ten each at 5 Gbps and 10 Gbps, for example), all behind a DMI link listed as x8 Gen4. For AMD, TechPowerUp says the B650 chipset provides eight usable PCIe 4.0 lanes, plus lanes shared between PCIe 3.0 and SATA, and six USB 10 Gbps ports; X670 boards chain two such chips together to offer 12 PCIe 4.0 lanes and more.
The chipset connects to the processor through an uplink. On AMD's AM5 that link is four PCIe lanes, and TechPowerUp notes that since the chipset supports PCIe 4.0 the link runs at Gen 4 even though the CPU supports 5.0. On Intel platforms with a separate hub, the link is called the Direct Media Interface, or DMI. The crucial point is that chipset-connected devices share the uplink's bandwidth, like several neighborhood roads feeding one highway. Each road can carry traffic, but vehicles heading to the same destination eventually pass through a shared connection.
When the Uplink Becomes a Bottleneck
Imagine an NVMe SSD, a fast USB drive and a network adapter all moving data through the chipset at once. Each has a fast local connection, but their combined demand crossing the uplink can exceed its capacity. The result is contention, and the devices may not hit their full advertised speeds simultaneously. That's why an M.2 socket's place in the board's architecture can matter: a drive on CPU lanes has a different bandwidth relationship than one behind the chipset. Neither is defective. They're different paths with different sharing.
Sharing doesn't mean slowness. Keyboards, mice, printers and audio interfaces use very little bandwidth, and they rarely all transfer at once. Even an external drive usually tops out well below the uplink's capacity. Sharing is a reasonable compromise, like a home network's switch and shared internet connection instead of a separate line for every device. The design assumes practical demand, not that everything runs flat out all the time.
It also means more chipset lanes don't equal unlimited bandwidth. A chipset may advertise many downstream lanes, but several chipset-connected SSDs could each have an x4 link while still sharing a narrower aggregate uplink. A network switch with many fast ports and a lower-capacity uplink works the same way. Local connection speed and shared upstream bandwidth are different things.
Why an SSD Can Disable a SATA Port
Modern boards often have several M.2 sockets, but they aren't wired identically. One might take four lanes directly from the processor, another four from the chipset, and a third might share lanes with another feature. The connector doesn't reveal where its bandwidth comes from. Two sockets can both support PCIe 4.0 x4, with one talking directly to the CPU and the other crossing the uplink. For everyday work both perform well. Under heavy simultaneous activity they may not.
That also explains one of the most confusing motherboard behaviors. You install a new M.2 SSD, it works, and a SATA port elsewhere stops. Chipsets and boards have limited high-speed I/O resources, and some connectors are wired to alternative uses of the same resources. A manufacturer may provide an M.2 socket and particular SATA ports that can't all work together, so enabling one disables the other. The computer isn't arbitrarily disabling hardware. The board is enforcing how its connections were allocated.
Graphics slots can behave the same way. A board may let a primary x16 slot use all sixteen lanes when alone, but divide them into two x8 links when a second device goes in a particular slot. The slots stay the same size. What changes is the number of electrical lanes. A full-length slot can run at x4, x2 or even x1, or connect through the chipset instead of the processor. For a sound card or modest network adapter, that's fine. A high-performance graphics card benefits from a wider direct link. The lesson: connector size doesn't tell you the architecture behind it, so distinguish physical slot dimensions from electrical lane configuration.
A Faster Chipset Isn't a Faster CPU
Take a Ryzen processor installed in a B650 motherboard and compare it with the same processor in a higher-tier AM5 board. With comparable memory, cooling, power delivery and settings, the processor's underlying computational capability doesn't change simply because the chipset has a different name. AMD's AM5 specifications show how the chipset tiers differ in connectivity: X670E and B650E require PCIe 5.0 support for graphics and NVMe storage, while standard X670 and B650 boards have different minimum requirements. Higher-tier chipsets may also provide more expansion lanes, USB connections and other platform features. Motherboard manufacturers sometimes pair those chipsets with stronger voltage regulation and more elaborate cooling, but those are board-design choices rather than properties of the chipset itself. A more expensive chipset can expand what the platform supports without making the CPU inherently faster.
For a typical gaming PC with one graphics card and a primary NVMe drive on CPU lanes, the chipset has relatively little direct effect on frame rates. The CPU, GPU, memory, game engine and settings matter more. A premium chipset offers more features, not a faster processor.
The Chipset Isn't the Whole Story
A motherboard is more complicated than a CPU connected to one chip. Some functions live in the processor, some in the chipset, and others in additional chips on the board. A network controller handles Ethernet, an audio codec converts between digital and analog sound, and a specialized USB controller may be extra hardware. Two boards with the same chipset can therefore offer 1 Gbps, 2.5 Gbps or 10 Gbps networking. The chipset defines some capabilities. The manufacturer decides how to implement and extend them.
USB is similar. A chipset may support a maximum number of high-speed ports, but manufacturers choose which to wire to rear connectors or internal headers, so a chipset specification describes potential, not the ports on a given product. (Monday's USB article covered what happens after a cable plugs in. This is the plumbing behind the socket.)
Power delivery is separate again. Voltage-regulator modules convert power from the supply into the voltages the CPU needs, using transistors, inductors and capacitors, the same technologies behind our earlier articles on cooling and coil whine. Stronger regulation and VRM cooling can help a board sustain demanding workloads, but that comes from the power design, not the chipset label. A premium chipset and a premium VRM often appear together because of product positioning, but they do different jobs.
The chipset also takes part in platform management, such as initialization, low-speed buses, peripheral configuration, security facilities and power coordination, with the division between processor, chipset and embedded controllers changing across generations. It isn't a second general-purpose CPU running your applications. Its contribution is organizational and electrical.
Why Not Put Everything in the CPU?
Some systems do. Smartphones, tablets, compact computers and embedded devices often use a system-on-chip, or SoC, which combines CPU cores, graphics, memory controllers, storage interfaces and wireless logic in one package, saving space and power. Some laptop and compact desktop platforms integrate what used to be chipset functions.
But integration involves tradeoffs. Every interface needs silicon, power, design effort, external pins and a share of the heat budget, and high-speed links need careful signal engineering. Adding every possible USB controller, SATA connection and expansion option to a desktop CPU would burden every buyer with features many never use. A separate chipset lets manufacturers serve very different users around a common processor family. A home-office machine with one SSD and integrated graphics doesn't need what a workstation with several NVMe drives, expansion cards and fast networking does.
So the chipset has shrunk in importance but not in purpose. Memory controllers, graphics links and NVMe access moved to the CPU, and the traditional northbridge has largely vanished from mainstream desktops. The need for additional peripheral connectivity remains, and chipsets are the descendants of the old support architecture.
Reading the Manual
Chipset choice matters most when a computer needs lots of connectivity: a workstation with several fast SSDs, a system with multiple network adapters or capture cards, a home server with many drives. What matters isn't the advertised tier but the layout on the specific board. The manual and block diagram reveal which M.2 sockets connect to the CPU and which to the chipset, whether an SSD disables SATA ports, whether secondary slots share processor lanes with the graphics slot, what electrical widths each slot supports, and which features can't run together. A board with more visible connectors isn't necessarily more flexible if they share limited resources in ways that conflict with your plans. Check the block diagram before you populate every slot, because a little research can prevent frustrating surprises.
The Bard's Take
Modern processors are extraordinary. They contain billions of transistors, multiple cores, sophisticated caches, memory controllers and high-speed interfaces, and many functions that once needed separate motherboard chips now live inside them.
But a processor isn't the entire computer. It still has to talk to storage, USB devices, networking hardware, expansion cards and audio systems, and giving every possible device a dedicated high-speed connection would be needlessly expensive and complicated.
That's where the chipset comes in. The CPU provides direct connections for the most demanding devices: memory, the main graphics card, selected NVMe drives. The chipset expands the platform with additional interfaces, routed through a shared link to the processor. That lets manufacturers build boards with different capabilities around compatible CPUs.
But the extra connections aren't necessarily independent. Chipset-connected devices may share uplink bandwidth, and some connectors compete for the same underlying lanes. That's why installing another SSD can disable a SATA port, why a full-length slot may run electrically at x4, and why two boards with the same CPU can have very different expansion without very different processor performance.
The chipset isn't a second brain competing with the CPU. It's part of the infrastructure that lets the CPU communicate with the rest of the machine. As architectures evolve, more of those functions may move into the processor or SoC, but the requirement remains: processing information is only useful when the system can receive it, store it and send it where it needs to go. The CPU does the computing. The chipset helps make sure the rest of the computer has a way to participate.