GPUs |

Flexible Software-Only Diverse Redundancy on COTS GPUs for Safety-Related Kernels

Wed, 01 Jan 2025 00:00:00 +0000

A Software-Only Approach to Enable Diverse Redundancy on Intel GPUs for Safety-Related Kernels

Sun, 01 Jan 2023 00:00:00 +0000

Analysis of Kernel Redundancy for Soft Error Mitigation on Embedded GPUs

Sun, 01 Jan 2023 00:00:00 +0000

Achieving Diverse Redundancy for GPU Kernels

Sat, 01 Jan 2022 00:00:00 +0000

GPU4S: Major Project Outcomes, Lessons Learnt and Way Forward

Mon, 01 Feb 2021 00:00:00 +0000

GPU4S: Embedded GPUs in space - Latest project updates

Wed, 01 Jan 2020 00:00:00 +0000

Software-only triple diverse redundancy on GPUs for autonomous driving platforms

Wed, 01 Jan 2020 00:00:00 +0000

Software-only Diverse Redundancy on GPUs for Autonomous Driving Platforms

Mon, 01 Jul 2019 00:00:00 +0000

High-Integrity GPU Designs for Critical Real-Time Automotive Systems

Fri, 01 Mar 2019 00:00:00 +0000

Safety-Related Challenges and Opportunities for GPUs in the Automotive Domain

Thu, 01 Nov 2018 00:00:00 +0000

GPU4S

Mon, 01 Jan 2018 00:00:00 +0000

GPU4S (Embedded GPUs for Space) is a research initiative funded by the European Space Agency (ESA) evaluating commercial off-the-shelf (COTS) embedded graphics processing units for space payload processing and autonomous navigation.

🎯 Objectives & Scope

Space applications increasingly require high-throughput parallel computing for satellite image processing, SAR, and autonomous navigation. GPU4S investigated the feasibility and reliability of deploying embedded GPUs in harsh radiation space environments.

💡 Key Research Innovations

Space GPU Benchmark Suite: Developed a standardized open-source benchmark suite representative of space payload processing workloads.
SafeSoftDR (Software-Based Diverse Redundancy): Created a software-only fault tolerance library providing kernel-level diverse redundancy on GPUs to mitigate soft errors without hardware modification.
COTS Architecture Evaluation: Evaluated low-power embedded GPU architectures (ARM Mali, NVIDIA Jetson, AMD) for radiation tolerance and performance-per-watt.

🏛️ Key Collaborators

Barcelona Supercomputing Center (BSC)
European Space Agency (ESA)
Airbus Defence and Space