AI compute specifications
Numerical precision, workload definitions and system requirements in published performance figures.
Units and numerical precision
NVIDIA lists 67 INT8 TOPS for Jetson Orin Nano Super. For DGX Spark it lists up to one PFLOP at FP4 with sparsity. These figures use different arithmetic and assumptions; their ratio does not establish relative agent-task performance.
Sources: NVIDIA Jetson Orin Nano Super specifications ↗ · NVIDIA DGX Spark specifications and footnotes ↗
System requirements
The products have different memory capacities and deployment roles. Model fit, software support, available power and input devices affect workload suitability alongside the published compute figure.
Sources: NVIDIA Jetson Orin Nano Super specifications ↗ · NVIDIA DGX Spark specifications and footnotes ↗
Benchmark configuration
A reproducible evaluation specifies the model, numerical precision, input size, software version and power setting. Interactive assistants and batch image-processing workloads have different performance requirements.
Comparison methodology
The comparison table preserves manufacturer units without converting them into a score. This project has not independently benchmarked either product. The listed figures are manufacturer specifications.
This is our interpretation of the linked documentation. No hands-on testing or independent benchmarking was performed.
Primary sources
Sources checked November 4, 2026. Product claims and documentation can change.