Ciru Lab / XDNA2 benchmark board

NPU token telemetry

Static AMD Ryzen AI NPU measurements from the Ciru benchmark bundle. The current matrix tracks FastFlowLM-NPU prefill and decode through 32k context.

Lab dashboard Race pages Catalog Sources

NPU / FastFlowLM-NPU / context matrix

NPU PP/TG Scatter

Fastest NPU Decode Rows

Published Qwen3.5 Decode Ladder

Published Qwen3.5 Prefill Ladder

Readout

NPU Context Rows

Model	Backend	Context	PP tok/s	TG tok/s	Timestamp

Reference Benchmarks And Architecture Notes