 | This article needs to be updated.(January 2017) |
"Uncore" is a term used by Intel to describe the functions of a microprocessor that are not in the core, but which must be closely connected to the core to achieve high performance. [1] It has been called "system agent" since the release of the Sandy Bridge microarchitecture. [2]
Typical processor cores contains the components of the processor involved in executing instructions, including the ALU, FPU, L1 and L2 cache. In contrast, Uncore functions include QPI controllers, L3 cache, snoop agent pipeline, on-die memory controller, on-die PCI Express Root Complex, and Thunderbolt controller. [3] Other bus controllers such as SPI and LPC are part of the chipset. [4]
The Intel uncore design stems from its origin as the northbridge. The design of the Intel uncore reorganizes the functions critical to the core, making them physically closer to the core on-die, thereby reducing their access latency.
Specifically, the microarchitecture of the Intel uncore is broken down into a number of modular units. The main uncore interface to the core is the so-called cache box (CBox), which interfaces with the last level cache (LLC) and is responsible for managing cache coherency. Multiple internal and external QPI links are managed by physical-layer units, referred to as PBox. Connections between the PBox, CBox, and one or more iMCs (MBox) are managed by the system configuration controller (UBox) and a router (RBox). [5]
Removal of serial bus controllers from the Intel uncore further enables increased performance by allowing the uncore clock (UCLK) to run at a base of 2.66 GHz, with overclocking limits in excess of 3.44 GHz. [6] This increased clock rate allows the core to access critical functions (such as the IMC) with significantly less latency, typically reducing core access times to DRAM by 10 ns or more.
Updated scope and components
Modern Intel microarchitectures (from Ice Lake and Tiger Lake to Alder Lake, Raptor Lake, and beyond) integrate an ever-increasing number of functions into the uncore. These include:
Cache and memory subsystems:
The last-level cache (LLC) and integrated memory controller now form the core of the uncore, supporting high-bandwidth, low-latency access to DRAM. Dynamic uncore frequency scaling is used to balance performance and power consumption.
Interconnects and I/O controllers:
High-speed interconnects—such as QuickPath Interconnect (QPI), Ultra Path Interconnect (UPI), and Direct Media Interface (DMI)—connect the core to other components. In newer processors, the on-die PCI Express root complex, Thunderbolt controller, and other peripheral interfaces are also integrated into the uncore.
Advanced system functions:
Modern uncore designs include additional functions such as integrated graphics support, wireless connectivity (e.g., Wi-Fi 6/6E and even Wi-Fi 7 in upcoming models), and dedicated security and power management engines. Some recent architectures also integrate features that accelerate AI workloads by combining traditional uncore functions with neural processing components.
Modular architecture and clocking
To achieve high performance, Intel's uncores are organized into modular units:
The cache box (CBox) interfaces with the LLC and manages cache coherency.
Physical-layer units (PBox) manage multiple high-speed links.
Controllers such as the System Configuration Controller (UBox) and Router (RBox) coordinate data transfers between the cores and various uncore components.
A key feature of modern designs is the high operating frequency of the uncore clock (UCLK). For example, many current processors run the uncore at a base frequency of around 2.66GHz (overclocking limits upwards of 3.44GHz), reducing access latency between the core and critical functions such as the memory controller. Advanced power management – including dynamic uncore frequency scaling – is now essential, especially for modern workloads (including heterogeneous CPU-GPU applications) requiring both energy efficiency and fast response.
Evolving role in modern microarchitectures
In recent years, the distinction between "core" and "uncore" has blurred. Advances in process integration have enabled functions traditionally separated in older designs to be more tightly combined:
In hybrid architectures (e.g., Alder Lake and later), the system agent now incorporates elements such as integrated graphics and additional I/O logic.
The integration of AI acceleration features and advanced connectivity controllers into the uncore reflects the growing demands of modern applications.
These enhancements not only reduce latency by placing critical functions physically closer to the cores but also improve energy efficiency. With dynamic scaling and sophisticated management of power states, modern uncore designs help optimize performance across diverse workloads while keeping overall power consumption under control.
References
"Thunderbolt™ Technology for Developers." Intel.com. Retrieved March 30, 2025.
Anand Lal Shimpi, "Intel's Sandy Bridge Architecture Exposed," AnandTech, September 14, 2010; updated information available in later reviews of later architectures.
"Intel® Xeon® Processor Uncore Programming Guide." Intel.com. Retrieved March 2025.