Does this model fit in this Jetson's memory?
Method v1.0 · dataset 2026-09-07 · verified 2026-09-07 · last updated September 2026
22 LLM, VLM and ASR models across 7 Jetson modules. Weights come from a published quantised artefact size when one exists, else parameters × bytes-per-parameter; KV cache from the model's own architecture; runtime overhead, OS headroom and reserve from the memory budget shared with Camera Stream Capacity. Every number traces to a source. Download the whole registry as JSON with provenance, or read the methodology.
Models × modules at a glance
Verdict and total memory at Q4 quantisation, 4096-token context, 1 sequence, llama.cpp, headless. FITS is under 85% of module memory, TIGHT is 85–100%, DOES NOT FIT is at or over capacity. Click a cell for the full breakdown; click a model name for its facts and every quantisation.
Module memory
Installed memory per module, from the NVIDIA technical specifications table (class A). Jetson modules share one memory pool between CPU and GPU: OS, runtime and model weights all draw on it.
| Module | Memory | Source |
|---|---|---|
| Jetson Orin Nano 8GB | 8 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson Orin NX 8GB | 8 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson Orin NX 16GB | 16 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson AGX Orin 32GB | 32 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson AGX Orin 64GB | 64 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson Thor T5000 | 128 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
| Jetson T4000 | 64 GB | NVIDIA · nvidia.com:jetson-modules-spec-table · verified 2026-09-07 |
Method and data
- Engine: Model Memory Fit — pick any model, module, quantisation, context and concurrency and get the live breakdown.
- Methodology: formulas, evidence classes, verdict thresholds and confidence rule.
- Dataset: model-memory.json — every model row and module capacity with its provenance record.
Check your exact combination.
Pick the model, module, quantisation, context length and concurrency — the engine builds the breakdown live and gives you a permanent link.