SEP 8, 2026 · PREPRINT
A Measurement Study of LLM Inference Trade-offs Across Edge Continuum Hardware
arXiv
A controlled measurement study of LLM inference across hardware platforms with multiple metrics, but no comparative intervention, randomization, or clinical outcome; findings are descriptive and exploratory rather than hypothesis-testing.
Reported
Hardware configurations testedNVIDIA Jetson AGX Orin; near-edge server wi…
Accuracy referenceGPT-4o (cloud-hosted)
Metrics reportedaccuracy, model footprint, per-token decodi…