How Graphics Processors Switch Between Different Tasks Efficiently

A method for graphics chips to pause and swap between different programs or tasks without waiting for every single part of the processor to finish its current job.

What it covers

This patent describes a way to pause a graphics processing pipeline so the computer can switch to a different task. Normally, a processor might have to wait for every single unit to finish its work before it can switch tasks, which is slow. This method sends a 'request-to-halt' signal to all units. Each unit reports back whether it is busy, idle, or has successfully paused. Once all units are either idle or paused, a 'stay-halted' signal locks them in place so their current state can be saved to memory and replaced with the state of a new task.

What it doesn't cover

• —Does not cover general-purpose CPU context switching that does not involve a graphics-specific pipeline.
• —Does not cover systems that require every unit to be completely idle before initiating a switch.
• —Does not cover software-only task switching that lacks the specific hardware 'stay-halted' signal architecture.

The clever bit

Instead of waiting for a total system stop, it allows units to enter a 'halted' state—where they stop mid-task but save their progress—allowing the system to switch tasks even when some units are still technically in the middle of an operation.

Why it matters

Graphics processors (GPUs) are designed to do many small tasks in parallel. Before this, switching between these tasks was inefficient because one slow unit could hold up the entire pipeline. This patent helped enable modern multitasking on GPUs, allowing the computer to switch between different applications or processes without a noticeable stutter in graphics performance.

Real-world examples

• 1.Nvidia GeForce graphics cards
• 2.GPU-accelerated multitasking in modern operating systems
• 3.Graphics drivers managing multiple concurrent applications