# US 2026/0133884 A1 · Techniques for memory error isolation

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0133884 A1
- Document ID: US20260133884A1
- Status: Published application
- Kind: A1
- Application number: 19/430,007
- Filed: 2025-12-22
- Published: 2026-05-14
- Earliest claimed: 2020-03-20
- Assignees: NVIDIA Corporation
- Applicants: NVIDIA Corporation
- Inventors: Jonathon Stuart Ramsay Evans; Naveen Cherukuri; Jerome Francis Duluk, JR.; Shailendra Singh; Vaibhav Vyas; Wishwesh Gandhi; Arvind Gopalakrishnan; Manas Mandal
- Web page: https://patentgrep.com/patent/US20260133884A1

## CPC

- G06F 11/203: {using migration}
- G06F 11/2094: {Redundant storage or storage space (G06F11/2056 takes precedence)}
- G06F 9/45558: {Hypervisor-specific management and integration aspects}
- G06F 11/0712: {in a virtual computing platform, e.g. logically partitioned systems}
- G06F 11/0793: {Remedial or corrective actions (recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40)}
- G06F 11/2043: {where the redundant components share a common memory address space}
- G06T 1/20: Processor architectures; Processor configuration, e.g. pipelining
- G06F 2009/45579: {I/O management, e.g. providing access to device drivers or storage}
- G06F 2009/45587: {Isolation or security of virtual machine instances}
- G06F 2201/815: Virtual
- G06F 2201/82: Solving problems relating to consistency
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

Apparatuses, systems, and techniques to detect memory errors and isolate or migrate partitions on a parallel processing unit using an application programming interface to facilitate parallel computing, such as CUDA. In at least one embodiment, interrupts are intercepted and processed on a graphics processing unit indicating a memory error for one or more partitions, and a policy is applied to isolate that memory error from other partitions.

## Claims

### Claim 1

One or more processors, comprising:
circuitry to cause one or more error handlers to:
detect one or more errors in a first one or more physical memory locations accessed by one or more graphics processing unit (GPU) slices;
store an error value in the first one or more physical memory locations based, at least in part, on the detected one or more errors;
determine an access of the error value by a software program executing on a GPU slice of the one or more GPU slices; and
isolate the software program executing on the GPU slice.

### Claim 2 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is to store the error value at least by modifying a data value stored in the first one or more physical memory locations.

### Claim 3 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is to clear the error value in response to the first one or more physical memory locations being free of the detected one or more errors.

### Claim 4 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is to isolate the software program at least by triggering an interrupt for the GPU slice.

### Claim 5 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is to isolate the software program at least by physically isolating the GPU slice.

### Claim 6 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is further to attribute the software program with the detected one or more errors based, at least in part, on the determined access of the error value.

### Claim 7 (depends on 1)

The one or more processors of claim 1, wherein the circuitry is further to map the detected one or more errors to the GPU slice.

### Claim 8

A system, comprising:
one or more processors to cause one or more error handlers to:
detect one or more errors in a first one or more physical memory locations accessed by one or more graphics processing unit (GPU) slices;
store an error value in the first one or more physical memory locations based, at least in part, on the detected one or more errors;
determine an access of the error value by a software program executing on a GPU slice of the one or more GPU slices; and
isolate the software program executing on the GPU slice.

### Claim 9 (depends on 8)

The system of claim 8, wherein the one or more processors are to store the error value at least by modifying a data value stored in the first one or more physical memory locations based, at least in part, on a constant value that is to indicate that the data value is incorrect.

### Claim 10 (depends on 8)

The system of claim 8, wherein the one or more processors are to clear the error value in response to the first one or more physical memory locations being reclaimed or blacklisted.

### Claim 11 (depends on 8)

The system of claim 8, wherein the one or more processors are to isolate the software program at least by triggering an interrupt for the GPU slice that is to cause the GPU slice to be isolated.

### Claim 12 (depends on 8)

The system of claim 8, wherein the one or more processors are to isolate the software program at least by isolating the GPU slice based, at least in part, on the determined access of the error value.

### Claim 13 (depends on 8)

The system of claim 8, wherein the one or more processors are to isolate the software program at least by preventing execution of the software program on the GPU slice to reduce spread of the error value.

### Claim 14 (depends on 8)

The system of claim 8, wherein the one or more processors are further to cause a kernel-level host driver to use a mapping of the detected one or more errors to attribute the detected one or more errors to the GPU slice and/or the software program.

### Claim 15

A method, comprising:
detecting one or more errors in a first one or more physical memory locations accessed by one or more graphics processing unit (GPU) slices;
storing an error value in the first one or more physical memory locations based, at least in part, on the detected one or more errors;
determining an access of the error value by a software program executing on a GPU slice of the one or more GPU slices; and
isolating the software program executing on the GPU slice.

### Claim 16 (depends on 15)

The method of claim 15, wherein:
storing the error value comprises modifying a data value stored in the first one or more physical memory locations based, at least in part, on a specific poison value that is to indicate that the data value is associated with the detected one or more errors.

### Claim 17 (depends on 15)

The method of claim 15, wherein isolating the software program comprises blacklisting the GPU slice based, at least in part, on the error value.

### Claim 18 (depends on 15)

The method of claim 15, wherein isolating the software program comprises triggering an interrupt for the GPU slice based, at least in part, on the determined access of the error value.

### Claim 19 (depends on 15)

The method of claim 15, further comprising attributing the software program with the detected one or more errors based, at least in part, on the determined access of the error value,
wherein isolating the software program comprises, in response to the software program being attributed with the detected one or more errors, preventing execution of the software program on the GPU slice.

### Claim 20 (depends on 15)

The method of claim 15, further comprising:
generating a mapping of the detected one or more errors to the GPU slice; and
using the mapping to attribute the detected one or more errors to the GPU slice and/or the software program executing on the GPU slice.
