NVIDIA Jetson Orin (T234)
Orin (T234) is NVIDIA’s 8 nm Ampere-generation Jetson SoC, sold as three module families. AGX Orin uses a 699-pin connector. Orin NX and Orin Nano are small 260-pin SO-DIMM modules. Since JetPack 6.1 the “Super” power modes raise the CPU, GPU and memory clocks on Orin NX and Orin Nano. Those boosted figures are listed in the notes. Orin Nano has no hardware video encoder and no DLA.
General Specifications
Specs
| Module | CPU | Process | Memory | GPU | NPU | Other |
|---|---|---|---|---|---|---|
| Jetson AGX Orin 64GB | 12× Cortex-A78AE @ 2.2 GHz | 8 nm | LPDDR5 256-bit Up to 64 GB 204.8 GB/s | Ampere, 2,048 CUDA cores, 64 Tensor Cores @ 1300 MHz Vendor: OpenGL ES 3.2, Vulkan 1.3 | 2× NVDLA v2, 275 TOPS (INT8) Total of GPU and DLA, sparse | Storage: 64 GB eMMC 5.1 TDP: 15–60 W |
| Jetson AGX Orin 32GB | 8× Cortex-A78AE @ 2.2 GHz | 8 nm | LPDDR5 256-bit Up to 32 GB 204.8 GB/s | Ampere, 1,792 CUDA cores, 56 Tensor Cores @ 930 MHz Vendor: OpenGL ES 3.2, Vulkan 1.3 | 2× NVDLA v2, 200 TOPS (INT8) Total of GPU and DLA, sparse | Storage: 64 GB eMMC 5.1 TDP: 15–40 W |
| Jetson Orin NX 16GB | 8× Cortex-A78AE @ 2 GHz | 8 nm | LPDDR5 128-bit Up to 16 GB 102.4 GB/s | Ampere, 1,024 CUDA cores, 32 Tensor Cores @ 918 MHz Vendor: OpenGL ES 3.2, Vulkan 1.3 | 2× NVDLA v2, 100 TOPS (INT8) 157 TOPS in Super mode | TDP: 10–25 W |
| Jetson Orin NX 8GB | 6× Cortex-A78AE @ 2 GHz | 8 nm | LPDDR5 128-bit Up to 8 GB 102.4 GB/s | Ampere, 1,024 CUDA cores, 32 Tensor Cores @ 765 MHz Vendor: OpenGL ES 3.2, Vulkan 1.3 | 1× NVDLA v2, 70 TOPS (INT8) 117 TOPS in Super mode | TDP: 10–20 W |
| Jetson Orin Nano 8GB | 6× Cortex-A78AE @ 1.7 GHz 1.5 GHz before Super mode | 8 nm | LPDDR5 128-bit Up to 8 GB 102 GB/s 68 GB/s before Super mode | Ampere, 1,024 CUDA cores, 32 Tensor Cores @ 1020 MHz Vendor: OpenGL ES 3.2, Vulkan 1.3 625 MHz before Super mode | 67 TOPS (INT8) GPU only, no DLA; 40 TOPS before Super mode | VPU: Hardware decode only, encoding runs on the CPU TDP: 7–25 W |
| Jetson Orin Nano 4GB | 6× Cortex-A78AE @ 1.7 GHz 1.5 GHz before Super mode | 8 nm | LPDDR5 64-bit Up to 4 GB 51 GB/s 34 GB/s before Super mode | Ampere, 512 CUDA cores, 16 Tensor Cores Vendor: OpenGL ES 3.2, Vulkan 1.3 | 34 TOPS (INT8) GPU only, no DLA; 20 TOPS before Super mode | VPU: Hardware decode only, encoding runs on the CPU TDP: 7–25 W |
Linux Support
Vendor BSP
- NVIDIA JetPack / Jetson Linux (Ubuntu-based) with CUDA, cuDNN and TensorRT
Mainline kernel
- Fully supported
- Works but not all features are implemented (Tegra platform support upstream; GPU acceleration needs NVIDIA’s driver stack)