訂閱電子報，掌握最新科技與產業趨勢
NVIDIA
NVIDIA® IGX Thor™ Module and Boardkit
NVIDIA IGX Thor™ Module and Boardkit
Enterprise-ready platform for physical AI.
The NVIDIA IGX Thor™ Platform unlocks real-time sensor processing and AI reasoning for physical AI agents with industrial-grade hardware, enterprise software, and functional safety. It’s powered by the NVIDIA Blackwell architecture iGPU and an optional dGPU, delivering up to 5581 FP4 TFLOPS of AI compute to effortlessly run multiple generative AI models at the edge. Compared to NVIDIA IGX Orin™, it provides up to 8x higher AI compute on iGPU, 2.5x higher AI compute on dGPU, and 2x better connectivity.
NVIDIA IGX Thor offers two production-ready systems—the NVIDIA IGX™ T5000 module and the NVIDIA IGX T7000 board kit—built for diverse physical AI applications with functional safety. The IGX T5000 module features a Blackwellpowered iGPU, a 14-core Arm® Neoverse®-V3AE CPU, a suite of accelerators, and flexible I/O for customization. The NVIDIA IGX™ T7000 board kit is a complete system, featuring a module, a Board Management Controller (BMC), and an NVIDIA ConnectX®-7 SmartNIC with 400 GbE networking. You can also add an NVIDIA RTX PRO™ Blackwell dGPU—designed to accelerate customers’ time-to-market. IGX Thor production systems use Transformer Engine and Multi-Instance GPU (MIG) support in the Blackwell architecture to deliver the performance to run multiple AI models with robust workload isolation. Functional Safety Island (FSI) inside the Thor SoC also enables advanced safety functions through the IGX software stack, ensuring reliable operation in mission-critical environments.
IGX Thor represents a new class of enterprise edge and robotics computers, purpose-built to power the next generation of industrial and medical edge applications. Taking advantage of the NVIDIA Blackwell GPU architecture, it excels at generative reasoning and multimodal sensor processing, supporting a broad range of generative AI models—from Vision Language Action (VLA) models like NVIDIA Isaac™ GR00T N to popular LLMs and VLMs such as NVIDIA Cosmos™ Reason. It also delivers a seamless cloud-to-edge experience, running the NVIDIA AI Enterprise software stack and NVIDIA NIM for physical AI applications. These include NVIDIA Isaac for robotics, NVIDIA Metropolis for visual AI agents, and NVIDIA Holoscan for sensor processing. Industrial safety AI agents can also be deployed at the edge using the NVIDIA Halos Outside-in Safety Agent Blueprint.
Our ecosystem of partners provides a complete range of carrier boards, design services, cameras, and sensors, along with AI and system software to speed solution development. They also support accelerating medical certifications (IEC 60601, 62304) and industrial functional safety certifications (ISO 26262, IEC 61508), helping you bring products to market faster and with confidence.
With enterprise-level software, massive AI compute, and network security, IGX Thor is ideal for advancing medical imaging, surgical robotics, humanoid robotics, industrial AI automation, high-performance computing, and beyond. As an enterprise-ready platform, IGX Thor enables companies to focus on application development and accelerate the realization of AI’s benefits.
NVIDIA IGX Thor offers two production-ready systems—the NVIDIA IGX™ T5000 module and the NVIDIA IGX T7000 board kit—built for diverse physical AI applications with functional safety. The IGX T5000 module features a Blackwellpowered iGPU, a 14-core Arm® Neoverse®-V3AE CPU, a suite of accelerators, and flexible I/O for customization. The NVIDIA IGX™ T7000 board kit is a complete system, featuring a module, a Board Management Controller (BMC), and an NVIDIA ConnectX®-7 SmartNIC with 400 GbE networking. You can also add an NVIDIA RTX PRO™ Blackwell dGPU—designed to accelerate customers’ time-to-market. IGX Thor production systems use Transformer Engine and Multi-Instance GPU (MIG) support in the Blackwell architecture to deliver the performance to run multiple AI models with robust workload isolation. Functional Safety Island (FSI) inside the Thor SoC also enables advanced safety functions through the IGX software stack, ensuring reliable operation in mission-critical environments.
IGX Thor represents a new class of enterprise edge and robotics computers, purpose-built to power the next generation of industrial and medical edge applications. Taking advantage of the NVIDIA Blackwell GPU architecture, it excels at generative reasoning and multimodal sensor processing, supporting a broad range of generative AI models—from Vision Language Action (VLA) models like NVIDIA Isaac™ GR00T N to popular LLMs and VLMs such as NVIDIA Cosmos™ Reason. It also delivers a seamless cloud-to-edge experience, running the NVIDIA AI Enterprise software stack and NVIDIA NIM for physical AI applications. These include NVIDIA Isaac for robotics, NVIDIA Metropolis for visual AI agents, and NVIDIA Holoscan for sensor processing. Industrial safety AI agents can also be deployed at the edge using the NVIDIA Halos Outside-in Safety Agent Blueprint.
Our ecosystem of partners provides a complete range of carrier boards, design services, cameras, and sensors, along with AI and system software to speed solution development. They also support accelerating medical certifications (IEC 60601, 62304) and industrial functional safety certifications (ISO 26262, IEC 61508), helping you bring products to market faster and with confidence.
With enterprise-level software, massive AI compute, and network security, IGX Thor is ideal for advancing medical imaging, surgical robotics, humanoid robotics, industrial AI automation, high-performance computing, and beyond. As an enterprise-ready platform, IGX Thor enables companies to focus on application development and accelerate the realization of AI’s benefits.
Technical Specifications
| NVIDIA IGX T5000 | NVIDIA IGX™ T7000 | ||
| dGPU Option | ‒ | With NVIDIA RTX PRO™ 6000 Blackwell Max-Q Workstation Edition | With NVIDIA RTX PRO™ 5000 Blackwell |
| AI Performance | 2,070 TFLOPS (FP4-Sparse) | Up to 5,581 TFLOPS (FP4-Sparse) | Up to 4,293 TFLOPS (FP4-Sparse) |
| iGPU | 2,560-core NVIDIA Blackwell architecture GPU with 96 fifth generation Tensor Cores Multi-Instance GPU with 10 TPCs |
2,560-core NVIDIA Blackwell architecture GPU with 96 fifth generation Tensor Cores Multi-Instance GPU with 10 TPCs |
2,560-core NVIDIA Blackwell architecture GPU with 96 fifth generation Tensor Cores Multi-Instance GPU with 10 TPCs |
| iGPU Max Frequency | 1.57 GHz | 1.57 GHz | 1.57 GHz |
| dGPU | 24,064-core Blackwell architecture GPU with fifth-generation Tensor Cores | 14,080-core Blackwell architecture GPU with fifth-generation Tensor Cores | |
| CPU | 14-core Arm Neoverse-V3AE64-bit CPU 64 KB I-cache, 64 KB dCache 1 MB L2 cache per core 16 MB shared system L3 cache |
14-core Arm Neoverse-V3AE64-bit CPU 64 KB I-cache, 64 KB dCache 1 MB L2 cache per core 16 MB shared system L3 cache |
14-core Arm Neoverse-V3AE64-bit CPU 64 KB I-cache, 64 KB dCache 1 MB L2 cache per core 16 MB shared system L3 cache |
| CPU Max Frequency | 2.6 GHz | 2.6 GHz | 2.6 GHz |
| Vision Accelerator | 1x PVA v3 | 1x PVA v3 | 1x PVA v3 |
| Memory | 128 GB 256-bit LPDDR5X 273 GB/s |
128 GB 256-bit LPDDR5X | 273 GB/s 96 GB dGPU memory | 1,792 GB/s |
128 GB 256-bit LPDDR5X | 273 GB/s 48 GB dGPU memory | 1,344 GB/s |
| Storage | Supports NVMe through PCIe Supports SSD through USB 3.2 |
4x SATA connectorsM.2 key M connector (PCIe Gen5 x2) |
4x SATA connectorsM.2 key M connector (PCIe Gen5 x2) |
| Video Encode | 2x NVEncode | 2x NVEncode (iGPU) 4x NVEncode (dGPU |
2x NVEncode (iGPU) 3x NVEncode (dGPU |
| Video Decode | 2x NVDecode | 2x NVDecode (iGPU) 4x NVDecode (dGPU) |
2x NVDecode (iGPU) 3x NVDecode (dGPU) |
| PCIe | Up to 8 lanes: Gen5 Root port only: C1 (x1) and C3 (x2) Root point or endpoint: C2 (x1), C4 (x8), and C5 (x4) |
2x PCIe Gen5 (x8, x16) | 2x PCIe Gen5 (x8, x16) |
| USB | xHCI host controller with integrated PHY (up to) 3x USB 3.2 4x USB 2.0 |
1x USB 3.2 Type-C 4x USB 3.2 Type A |
1x USB 3.2 Type-C 4x USB 3.2 Type A |
| Networking | 4x 25GbE | 2x RJ45 (1GbE each) 2x QSFP28 (200GbE each) |
2x RJ45 (1GbE each) 2x QSFP28 (200GbE each) |
| Integrated ConnectX | No | Yes | Yes |
| Display | 1x shared HDMI 2.1 1x VESA DisplayPort 1.4a: HBR2, MST |
1x VESA DisplayPort 1.4a | 1x VESA DisplayPort 1.4a |
| Other I/O | 2x 13-pin CAN header 2x 6-pin automation header 2x 5-pin header JTAG connector 1x 4-pin fan connector: 12V, PWM, and tach 2x 5-pin audio panel header 2-pin RTC backup battery connector |
2x USB header for optional connection via cable 4x COM Ports 1x LPT 1x TPM 1x CAN 1x GPIO Audio Line-out, Mic In |
2x USB header for optional connection via cable 4x COM Ports 1x LPT 1x TPM 1x CAN 1x GPIO Audio Line-out, Mic In |
| BMC | ‒ | Yes | Yes |
| Safety Support | Functional Safety Island on SoC | Functional Safety Island on SoC Safety MCU on the carrier board |
Functional Safety Island on SoC Safety MCU on the carrier board |
| NVDIA AI Enterprise Support | Yes | Yes | Yes |
| Power | 40 W–130 W | 40 W–130 W TMP Up to 300 W for dGPU |
40 W–130 W TMP Up to 300 W for dGPU |
| Mechanical | 100 mm x 87 mm 699 pin B2B connector Integrated thermal transfer plate (TTP) with heat pipe |
243.84 mm x 198.98 mm x 31.05 mm | 243.84 mm x 198.98 mm x 31.05 mm |
*NVIDIA Authorized Distributor
