ThunderX

The AIBOX Series is an independent high-performance in-vehicle AI computing hardware platform launched by ThunderX, combining a plug-and-play hardware and fully open software stack architecture to give automakers a fast track to deploy and continuously upgrade intelligent driving and smart cockpit functions. Powered by the Driodo AIOS, the platform deeply integrates three core capabilities - compute allocation, model inference, scheduling, and AI scenario adaptation - and provides standardized interfaces and a one-stop development platform, enabling OEMs to rapidly build multi-agent AI function matrices such as welcome Agent, enhanced sentinel Agent, parking-memory Agent, and GUI Agent.

Hardware platform: NVIDIA ORIN Y (AIDC-N1) / Qualcomm QCS9075 (AIDC-Q1); Software: powered by Driodo AIOS; Cooling: air-cooled / liquid-cooled modes; Communication: standardized Ethernet protocol; Features: plug-and-play, independent development unaffected by the cockpit system, supports multi-image processing and concurrent models; Cost: lower than upgrading the cockpit chip platform

• Smart cockpit AI
• Welcome Agent
• Enhanced sentinel Agent
• Parking-memory Agent
• GUI Agent
• In-vehicle multi-agent AI function matrix

RazorDCX is a centralized intelligent driving domain controller from ThunderX that integrates environmental perception, information fusion, decision planning, and actuation functions, serving as a high-performance in-vehicle real-time computing platform for the development and implementation of intelligent driving strategies. The series covers multiple computing platforms ranging from entry-level to flagship, from integrated parking-and-driving to cockpit-driving fusion, supporting all-time integrated parking-and-driving and up to L2.9+ intelligent driving capability.

AI compute: 50 TOPS - 2000 TOPS (by model); CPU compute: 200-500 kDMIPS; Functional safety: ASIL-D; Sensors: up to 12 camera inputs (up to 8M pixels); NOA: no high-precision map or positioning required; Chip platform: Qualcomm SA8540P / SA8620P / SA8650P / SA8255P / SA8775P and NVIDIA ORIN Y, etc.

• L2+ assisted driving
• Integrated parking-and-driving
• Highway NOA
• Urban NOA
• Cockpit-driving fusion
• Closed-park L4 autonomous driving

RazorDCX Takla is a domain controller solution from ThunderX for integrated parking-and-driving scenarios, using Qualcomm's advanced 7nm chip SA8540P to deliver 60 TOPS / 200 kDMIPS of computing power to the vehicle. It can be applied to low-speed automated parking, highway automated driving, and urban automated driving, and can be extended to closed-park autonomous driving scenarios such as unmanned vending vehicles, mines, and automated cleaning vehicles.

SoC: Qualcomm SA8540P (7 nm); AI compute: 60 TOPS (INT8), expandable to 720 TOPS (INT8); SoC compute: 200K-400kDMIPS; ASIL-D compute: 4 kDMIPS; Camera: 12 channels up to 8M pixel (1x8M@30fps + 7x3M + 4x2M); Perception: millimeter-wave radar x5 / ultrasonic x12 / LiDAR x3; Ethernet: 2x1Gb/s + 6x100Mb/s; CAN: 12-channel CAN-FD; Power: whole unit <50W; Functional safety: ASIL-D (safety chip); Grade: parking-and-driving integrated, closed road L2.9+ / others L2

• Parking-and-driving integrated
• Low-speed automated parking
• Highway automated driving
• Urban automated driving
• Unmanned vending vehicle
• Mine / automated cleaning vehicle closed-park L4

RazorDCX Pantanal is ThunderX's latest production-grade intelligent driving domain controller solution, offering multiple stepped configurations to meet the varying needs of entry-level to flagship models for intelligent driving domain control. Using Qualcomm's new-generation Snapdragon Ride platform production-grade 4nm process intelligent driving chip, it can integrate 29 ADAS functions across the three modules of safety, comfort, and parking, achieving memory parking, L2.9 highway autonomous driving, and L3 urban autonomous driving.

SoC: Qualcomm SA8650P (4 nm Snapdragon Ride); AI compute: 50-100 TOPS / 230 kDMIPS per chip, expandable to 400 TOPS; Camera: 12x3M@30fps; Sensors: 7V5R12USS base / 11V5R12USS high-tier (free from high-precision map); Ethernet: 8 automotive-grade channels; CAN/CANFD: multiple reserved; Functions: 29 ADAS (safety / comfort / parking), HAVP / L2.9 highway / L3 urban; Software: QNX / Linux (Gunyah) + SafetyRTOS; Power: ~60W; Functional safety: ASIL-D; Dimensions: 254x188x44 mm

• L2.9 highway autonomous driving
• L3 urban autonomous driving
• Memory parking HAVP
• Closed-park L4 autonomous driving
• Unmanned vending vehicle
• Mine / automated cleaning vehicle

RazorDCX Congo is ThunderX's latest-generation ultimate cost-performance intelligent driving domain controller solution, meeting the need for low-cost intelligent driving domain control. Using Qualcomm's new-generation Snapdragon Ride platform production-grade intelligent driving chip, it supports highway NOA + automated parking (parking-and-driving integrated), enabling L2+ basic low-speed parking and highway automated driving, and can be extended to closed-park autonomous driving scenarios.

SoC: Qualcomm SA8620P (Snapdragon Ride); AI compute: 36 TOPS / 80 kDMIPS (ASIL-B); ASIL-D compute: 9k + 2.7kDMIPS; Camera: 2x8M@30fps + 2x3M (SVC) + 1x3M (RVC) + 4x2M (AVM); Perception: front radar x1 / corner radar x4 / USS x12; Ethernet: 4 automotive-grade channels; CAN: CANFD x6 (1 wake-up channel); Cooling: air-cooled; Grade: parking-and-driving integrated, closed road L2.9 / L2+; Functional safety: board-level ASIL-B, functional-level up to ASIL-D (with TC397); Dimensions: ≤255x167x70 mm

• Highway NOA
• Automated parking
• Parking-and-driving integrated
• L2+ assisted driving
• Closed-park autonomous driving
• Unmanned vending vehicle / mine

RazorDCX Tongass is a new-generation production-grade cockpit-parking integrated domain controller solution based on Qualcomm SA8255P, delivering smooth and realistic graphics, intelligent interaction experience, and excellent image processing capability. Supported by ThunderSoft's deep understanding of the cockpit domain, it provides automakers with a cockpit-parking integrated domain control solution that combines high performance and cost advantages, integrating parking functions such as APA and RPA to deliver a brand-new driving experience of cockpit-parking fusion.

SoC: Qualcomm SA8255P (5 nm); AI compute: 20 TOPS; GPU: 1.5 TFLOPS; CPU: 170 kDMIPS; Cockpit: one-chip multi-screen up to 12 screens max 8K; Multi-OS: cluster (Linux/QNX) + other screens (Android); Smart vision: AVM + DMS / FACE ID + DVR + OMS + CMS; Parking: APA / RPA / HPA; Interfaces: MIPI 4x4lane CSI-2, USB 2x3.1 + 1x2.0, PCIe 2xGen4, LPDDR5 36GB; Functional safety: ASIL-D; Software: Driodo OS Pre 1.0 8255 cockpit edition

• Smart cockpit
• Cockpit-parking integrated
• Automated parking APA/RPA/HPA
• Multi-screen interaction
• AVM surround view
• DMS/OMS driver monitoring

RazorDCX Tarkine is the industry's first single-SoC cockpit-driving fusion domain controller solution based on Qualcomm SA8775P, using the 4nm Snapdragon Ride Flex platform automotive-grade chip, where a single chip simultaneously satisfies both cockpit-domain and intelligent-driving-domain needs. It enables cockpit 3D HMI, in-cabin vision, gaming and infotainment, connectivity, and more, while also covering intelligent-driving assistance such as automated parking and highway NOA, making it a current-market cockpit-driving fusion domain control solution that combines cost-effectiveness and technical leadership.

SoC: Qualcomm SA8775P (4 nm Snapdragon Ride Flex); Compute: 72 TOPS + 230 kDMIPS (ASIL-B); Camera: 11V5R12U (2 front 8M + 4 surround-view + 4 around-view + 1 rear); Perception: millimeter-wave radar x5 / ultrasonic x12 / LiDAR x1; Ethernet: 4x1000BASE-T1 + 4x100BASE-T1; CAN: 12-channel CAN-FD; Software: QNX + Android, SAIL SafetyRTOS; Power: whole unit <60W (~75W); Functional safety: ASIL-D (safety chip); Grade: L2++ highway / urban; Cooling: cavity liquid-cooled; Volume: 203x182x60.9 mm

• Cockpit-driving fusion
• Smart cockpit 3D HMI
• Highway NOA
• Urban intelligent driving
• Automated parking
• Unmanned mining truck / embodied robot

RazorDCX Sylvania is ThunderX's flagship new-generation AI cockpit domain controller solution unveiled at CES 2026, based on Qualcomm Snapdragon Cockpit Platform Supreme Edition (SA8397P), focusing on central computing architecture upgrades for high-end models. Driven by the single-chip intelligent fusion philosophy, it promotes deep integration of AI technology with the cockpit system, deeply integrating with Driodo OS 2.0 Pre to provide OEMs with a complete development framework and toolchain.

SoC: Qualcomm SA8397P (Snapdragon Cockpit Platform Supreme Edition); CPU: specint2017rate 80; AI dense compute: 320 TOPS; Memory bandwidth: 272 GB/s (raw); Architecture: single chip simultaneously carries complex large AI models and traditional cockpit functions; Software: Driodo OS 2.0 Pre; Interaction: multimodal interaction with millisecond-level response, intent recognition accuracy >95%; Interfaces: rich interfaces reserve ample upgrade headroom

• High-end AI cockpit
• Central computing architecture
• In-vehicle multimodal interaction
• Large-model AI cockpit
• Intelligent voice / vision Agent
• High-end model mass production

RazorWareX is ThunderX's platform software for intelligent driving application algorithms, configuring a high-real-time, high-bandwidth communication backbone among the domain controller's chips and the vehicle's systems, fully allocating hardware compute, sharing sensor data, and providing an AI algorithm runtime environment to deliver a highly reliable open software platform for intelligent driving. RazorWareX 1.0 is a high-functional-safety, high-real-time, deterministic middleware platform dedicated to intelligent driving scenarios.

Features: high real-time / high determinism / high functional safety / end-to-end latency guarantee / lowest latency for sensitive data; Architecture: time-triggered communication backbone, SOA architecture; Interfaces: heterogeneous RTE interface supporting multi-chip multi-domain; Toolchain: compliant with SOA architecture, supports automatic code generation; Functional safety: ASIL-D; Applications: cross-chip / cross-system communication, AI algorithm runtime environment

• Intelligent driving middleware
• Cross-domain communication backbone
• AI algorithm runtime environment
• Cockpit-driving fusion software platform
• Toolchain automatic code generation
• SOA service-oriented architecture