A reproducible playground for edge AI research
EdgeLab enables researchers to evaluate AI workloads on real edge hardware, generate defensible performance results, analyse model–hardware interactions and compare architectures across vendors without owning, managing, or optimising for specific devices.
Expand the experimental scope of your research
Edge AI research beyond simulations and specs
Edge AI research introduces challenges that are difficult to address within typical academic project constraints. While model benchmarking is routine, evaluating how workloads behave across different edge hardware platforms is rarely systematic. As a result, experiments are often conducted on the most readily available or dominant platforms, rather than explored across the broader hardware landscape.
Hardware access, engineering overhead, and limited observability make it difficult to generate results that are both scientifically interesting and credible.
Hardware benchmarking is ad‑hoc and rarely spans multiple vendors
Device availability limits experimental scope and comparability
Engineering overhead discourages systematic evaluation
Performance metrics like latency and power are difficult to measure reliably
Results based on theoretical TOPS lack credibility
Reproducing results across labs and projects is challenging
What EdgeLab changes in your workflow
Rather than relying on:
Solely on simulations or emulations
Ad‑hoc, project‑specific setups with large engineering overhead
Defaulting to the same hardware-platform
You get:
Real‑hardware measurements
Standardized, repeatable and simple to set up evaluation pipelines
Cross‑vendor comparability
EdgeLab complements your research workflow. It does not require you to optimise for deployment or manage hardware.
Supporting research across the full AI lifecycle
Generate credible performance results
Establish trustworthy performance baselines for your workloads using real edge hardware.
Reproducible and comparable
EdgeLab is an initiative by imec. Results are transparent, reproducible and vendor‑agnostic.
Access hardware you don't own
Benchmark and compare models across a wide range of edge devices, including those you would normally not have access to, without the need to own or maintain it.
Reduce engineering overhead
Spend less time on infrastructure and more time on scientifically meaningful questions. When ready, migrate your EdgeLab config with zero effort.
Understand model–hardware interactions
Measure latency, memory usage, power consumption, and stability under controlled conditions and gain insight into how models behave across devices