How Intel's Memory Hub Manages Graphics Data Across Different Memory Types

A hardware design for a computer memory hub that lets a processor treat different types of memory as one unified space for graphics tasks.

What it covers

This patent describes a memory control hub that acts as a traffic cop between a computer's main processor and its memory. When the processor needs to access graphics data, the hub checks if the data is a graphics operand. If it is, the hub uses a translation table to map the processor's virtual address to the actual physical location of the data, which might be in main system memory or dedicated graphics memory. This allows the system to seamlessly move or store graphics data across two different memory pools while keeping the processor's job simple.

What it doesn't cover

• —Does not cover software-only methods for memory management that lack a hardware memory control hub.
• —Does not cover systems that only use a single unified memory pool without a translation table mechanism.
• —Does not cover methods of graphics rendering or image processing algorithms themselves.

The clever bit

The invention uses a translation table and fence registers to reorder memory addresses on the fly, effectively hiding the complexity of physical memory location from the CPU while optimizing access patterns for the graphics device.

Why it matters

This technology was crucial during the transition to integrated graphics architectures where the line between system RAM and dedicated video memory began to blur. By allowing the CPU to manage graphics memory dynamically, it helped improve performance in early 2000s PC architectures, specifically those using Intel's chipset designs.

Real-world examples

• 1.Intel integrated graphics chipsets from the early 2000s
• 2.Unified Memory Architecture (UMA) implementations in desktop PCs