The IGX T7000 uses a newer GPU architecture, more CUDA cores and twice the memory of the IGX Orin 700. Both are IGX board kits.
GPU architecture, CUDA cores and memory nvidia.com +1nvidia.com ↗“IGX Thor MicroATX board kit”nvidia.com ↗“IGX Orin MicroATX board kit”
| Attribute | IGX T7000 | IGX Orin 700 |
|---|---|---|
| Family | IGX Thor | IGX Orin |
| GPU architecture | Blackwell | Ampere |
| GPU CUDA cores | 2,560 | 2,048 |
| Tensor cores | Not listed | 64 |
| Memory | 128 GB LPDDR5X | 64 GB LPDDR5 with ECC |
| Memory bandwidth | 273 GB/s | Not listed |
What's different
- GPU architecture: The T7000's built-in GPU is Blackwell, and the Orin 700's is Ampere. Both kits can also take a separate (discrete) GPU. For the T7000, the listed options are the RTX PRO 6000 or RTX PRO 5000 Blackwell. nvidia.comnvidia.com ↗“dGPU support (RTX PRO 6000 or 5000 Blackwell)”
- CUDA cores: The T7000 has 2,560 and the Orin 700 has 2,048.
- Memory: The T7000 has 128 GB of LPDDR5X at 273 GB/s. The Orin 700 has 64 GB of LPDDR5 with ECC. I couldn't find a memory bandwidth figure for the Orin 700. Peak performance (listed in different precisions, so the numbers can't be compared directly): nvidia.comnvidia.com ↗“64GB LPDDR5 with ECC”
- IGX T7000: up to 5,581 FP4 TFLOPS (sparse). nvidia.comnvidia.com ↗“up to 5581 FP4 TFLOPS (sparse)”
- IGX Orin 700: up to 1,705 TOPS. nvidia.comnvidia.com ↗“up to 1705 TOPS”
For more detail, see the NVIDIA IGX product page and the IGX Thor datasheet.