What do you need to run?

For FAEs, integrators and engineers choosing edge-AI hardware (GPU, NPU, FPGA, MPSoC, Versal), for vision and custom models first and local LLMs second. Every number is cited and dated. About

Ranked result

Latest full feed Atom JSON Feed

What's coming all upcoming hardware →

Show data — full results table, advanced recommender, customer export, methodology

show superseded older results

Track Task Vendor Confidence ≥ Search with power only

All published results. Activate a column header to sort.
Select row	Hardware	Task	Model	Prec.	Stack	Metric	Value	Unit	≈FPS	W	Perf/W	Conf.	Date	Source

Workload-based recommender

Task Target FPS / tok/s ≥ Power budget (W) ≤ Hardware budget ($) ≤ Min confidence Model-size class Batch allow mixed (fill gaps from other size/batch, flagged)

Ranks the best published result per platform within one like-for-like group (same model-size class and batch size). Model, precision, stack and power basis can still differ — check before quoting. Unknown price/power never disqualifies — it is flagged.

Customer Prepared by

Revisions

Data revisions
Rev	Description	Date
A	MemryX, Axelera and MLPerf Tiny datasets added; 8,698 rows in total	2026-10-16
B	Workload field added to every row; 7,149 rows map to the tracked workloads	2026-10-16