Here's how the two IGX board kits compare on those dimensions:
| Attribute | IGX Orin 700 | IGX T7000 |
|---|---|---|
| GPU architecture | Ampere | Blackwell |
| CUDA cores | 2048 | 2560 |
| Memory | 64 GB LPDDR5 (with ECC) | 128 GB LPDDR5X |
| Memory bandwidth | not listed in the catalog | 273 GB/s |
| CPU cores | 12 | 14 |
| Peak AI performance | 1705 TOPS (INT8) | 5581 FP4 TFLOPS (sparse) |
Key differences on the three dimensions you asked about:
- GPU architecture: The IGX Orin 700 uses the Ampere architecture, while the IGX T7000 moves to NVIDIA's newer Blackwell architecture with 2560 CUDA cores versus 2048 on the Orin 700. - Memory: The IGX T7000 doubles the memory to 128 GB of LPDDR5X, compared with 64 GB LPDDR5 (with ECC) on the IGX Orin 700. nvidia.com +1nvidia.com ↗“memory.memory_gb=128 GB, memory.memory_type=LPDDR5X”nvidia.com ↗“64GB LPDDR5 with ECC” nvidia.com +3nvidia.com ↗“compute.gpu_architecture=Ampere”nvidia.com ↗“compute.gpu_architecture=Blackwell”nvidia.com ↗“compute.gpu_cuda_cores=2560”nvidia.com ↗“compute.gpu_cuda_cores=2048”
- CPU cores: The IGX T7000 has 14 CPU cores; the IGX Orin 700 has 12. One note on the performance figures: the 1705 TOPS figure for the IGX Orin 700 is an INT8 measure, while the 5581 figure for the IGX T7000 is FP4 (sparse) — these are different precisions and should not be directly compared. nvidia.com +1nvidia.com ↗“compute.cpu_cores=14”nvidia.com ↗“compute.cpu_cores=12”