NVIDIA DRIVE AGX Thor
NVIDIA DRIVE AGX Thor is a centralized automotive computing platform built on the Blackwell GPU architecture, delivering up to 2,000 FP4 TOPS on a single ASIL-D SoC to unify autonomous driving, cockpit, and infotainment functions in software-defined vehicles.
- CPU
- Arm Neoverse V3AE
- Security
- ISO 21434
- Integration
- Single-SoC AV + cockpit + IVI
- Process Node
- TSMC 4N (5nm-class)
Where to buy in Singapore
Automotive-grade sensing and compute is supplied to vehicle manufacturers under design-win programmes, in production volumes and with multi-year validation — it is not sold as individual units through resellers. Approach the manufacturer directly if you are specifying for a production programme.
We do not supply this product. It already has strong distribution in Singapore, and buying locally gets you local warranty and service that we could not match. We keep the page because the comparison is still useful.
Highlights
- ✓Data-center-class Blackwell architecture in-vehicle
- ✓Single-chip consolidation of AV and cockpit domains
- ✓ISO 26262 ASIL-D functional safety
- ✓Headroom for generative-AI and transformer models
Overview
Overview
NVIDIA DRIVE AGX Thor is a centralized in-vehicle computing platform built on the NVIDIA Blackwell GPU architecture and next-generation Arm Neoverse V3AE CPUs. Its core design goal is consolidation: rather than relying on multiple separate electronic control units (ECUs) for different vehicle functions, Thor combines automated driving, parking, driver and occupant monitoring, digital cluster, and infotainment onto a single System-on-Chip (SoC). This architectural approach is foundational to what the automotive industry calls the "software-defined vehicle" — a vehicle whose capabilities are primarily defined and updated through software rather than fixed hardware.
Performance
Thor delivers up to 1,000 INT8 TOPS (Tera Operations Per Second) and 2,000 FP4 TOPS of compute throughput. These figures reflect its design orientation toward modern AI workloads, specifically transformer-based neural networks and generative AI models that underpin advanced perception, path planning, and in-cabin intelligence. The chip is fabricated on TSMC's 4N process node, which provides the transistor density and power efficiency needed to support driving automation stacks ranging from L2++ (advanced driver assistance) through L4 (high automation) — all on a single piece of silicon.
Safety and Security
Automotive deployments impose strict functional safety and cybersecurity requirements. DRIVE AGX Thor is developed to comply with ISO 26262 ASIL-D, the highest integrity level in the automotive functional safety standard, and ISO 21434, the standard governing automotive cybersecurity engineering. These certifications are prerequisites for deployment in production vehicles, where failures in compute systems can have direct safety consequences.
Who Uses It and How
DRIVE AGX Thor is used by automotive OEMs and Tier-1 suppliers building next-generation vehicle platforms, as well as by developers and researchers working on autonomous driving stacks. By centralizing compute that was previously distributed across many separate ECUs, the platform reduces hardware complexity and wiring, while enabling over-the-air software updates to add or improve vehicle capabilities post-production. The DRIVE AGX Thor Developer Kit began shipping in September 2025, and the platform has entered production vehicles, confirming genuine commercial availability for both development and deployment use cases.
Specifications
| CPU | Arm Neoverse V3AE |
| Security | ISO 21434 |
| Integration | Single-SoC AV + cockpit + IVI |
| Process Node | TSMC 4N (5nm-class) |
| AI Performance | 1,000 INT8 TOPS / 2,000 FP4 TOPS |
| GPU Architecture | NVIDIA Blackwell |
| Functional Safety | ISO 26262 ASIL-D |