AMD
AMD · AI / ML · Filed Dec 23, 2024 · USPTO published Jun 25, 2026 · Jun 25 release
AMD Patents a System That Picks the Right AI Chip Based on Battery Mode
Not every AI task needs the fastest chip in the room. AMD's new patent describes a system that figures out which ones do, and which ones can run on a lower-power chip instead.
- Publication number
- US 2026/0178379 A1(US20260178379A1)
- Applicant
- Advanced Micro Devices, Inc.
- Filed · Published
- Dec 23, 2024 · Jun 25, 2026
- Inventors
- Virendra Pratap Arya
- Status when we published
- Waiting for an examiner
- Document
- 20 claims
- US classification
- 718/102
How AMD's patent splits AI work by power mode
Imagine your laptop is in battery-saver mode and you ask an AI assistant to do two things at once: transcribe your voice and generate a detailed image. Those two jobs are very different in terms of how much computing muscle they need. AMD's patent covers a system that knows this and routes each job accordingly.
When your device is trying to save power, the system compiles a stripped-down version of eligible AI tasks and sends them to a smaller, less power-hungry chip. But if a task is too demanding to run in a reduced form, it still gets sent to the full-power AI chip, even in battery-saver mode.
The practical effect: your device stays cooler and lasts longer on a charge, without silently failing at the tasks that genuinely need full power.
How the host circuit compiles and routes each AI task
AMD's patent describes a computing setup with at least two distinct AI processing units: a neural processing circuit (a full-featured, high-performance AI chip) and an inferencing accelerator (a smaller, less capable chip that draws less power). A host processing circuit acts as the traffic controller between them.
When the system enters a low-power operating mode, the host circuit checks each incoming AI task to see whether it can be compiled into a lighter version. If a task is flexible enough, the host prepares that low-power version and assigns it to the inferencing accelerator.
The critical rule: if a task has no low-power alternative (meaning it requires high performance no matter what), the host circuit sends it to the full neural processing circuit anyway, overriding the power-saving preference. This prevents the system from silently producing degraded results just to save energy.
The word "inference" here means running a trained AI model to get an answer (as opposed to training the model in the first place). This patent is entirely about the inference side, which is what happens on your device every time you use an AI feature.
What this means for AI on laptops and low-power devices
On laptops, phones, and edge devices where battery life is a real constraint, running every AI task at full throttle is wasteful. This patent gives AMD a formal architectural approach to splitting that workload, which is directly relevant to its Ryzen AI chip lineup and any device that mixes heavy AI features with battery-conscious use cases.
For you as a user, this could mean your device intelligently decides which AI features get a lighter treatment in power-saver mode and which ones always run at full quality, without requiring you to manage any of that manually. The approach also hints at how AMD is thinking about dedicated AI hardware as a first-class scheduling target, not just an optional accelerator.
Source: US 2026/0178379 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time AMD patent news breaks.
Add Patentlyze on Google
One click.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
More from AMD
All 119 AMD patents →
More in AI training & infrastructure
All 316 AI training & infrastructure patents →
Nvidia
Nvidia Patents a Technique That Cuts AI Training Time Without Sacrificing Accuracy
Nvidia
Nvidia's New Patent Removes a Hidden Cost That Slows Down AI Training
Qualcomm
Qualcomm Patents a Faster Way to Run a Core AI Calculation on Chips
Google
