Add ONNX Runtime Renesas EP v2.1.0 benchmarks

Latency, NPU/CPU split, node distribution and accuracy for 1 and 12 AI cores
from ORT_Renesas_EP_Test_v2.1.0 (MWMX 2.2 ENG4); update existing ORT results in place.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

README.md

@@ -66,6 +66,36 @@ TBD — not yet measured/published for this repo.
 
 ---
 
+<!-- ORT-V210:BEGIN -->
+### ONNX Runtime – Renesas EP v2.1.0 (MWMX 2.2 ENG4)
+
+Measured on **R-Car X5H** with ONNX Runtime + Renesas Execution Provider (INT8 auto-cast); nodes the NPU cannot run fall back to the CPU EP. Batch size 1, 1 NPU. Source: ORT Renesas EP test report, status 2026-09-30. This section supersedes earlier ORT figures in this card.
+
+| Component | AI Cores | Latency (ms) | Throughput (fps) | NPU Inference (ms) | NPU % | CPU + Overhead % | Portable pkg NPU-only (ms) |
+| :-- | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
+| Full model | 1 | 2.114 | 473 | 1.476 | 69.8 % | 30.2 % | 1.48359 |
+| Full model | 12 | 1.729 | 578.4 | 0.799 | 46.2 % | 53.8 % | 0.692726 |
+
+> 12-core latency is *not* always lower than 1-core: small models are dominated by CPU-side overhead.
+
+**Graph partitioning (NPU vs CPU nodes)**
+
+| Component | Total Nodes | NPU Nodes | CPU Nodes (Q/DQ inserted) | NPU % | CPU % |
+| :-- | :-: | :-: | :-: | :-: | :-: |
+| Full model | 50 | 49 | 1 (1) | 98.0 % | 2.0 % |
+
+**Accuracy (ORT)**
+
+| Runtime / Precision | Result |
+| :-- | :-- |
+| CPU EP · FP32 reference | Top-1 68.5 % · Top-5 90.1 % |
+| Renesas EP + CPU EP · INT8 · v2.1.0 | Top-1 71.4 % · Top-5 89.7 % |
+
+> Evaluated on a subset (1k images), not the full validation set.
+<!-- ORT-V210:END -->
+
+---
+
 ## Runtime Details
 
 ### MWMX Runtime

fp32/benchmarks/x5h_ort_cpu_fp32.yaml

@@ -0,0 +1,51 @@
+# FP32 reference accuracy: ONNX Runtime CPU EP (ORT_Renesas_EP_Test_v2.1.0.xlsx, 'Accuracy ( CPU fp32 )')
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: onnxruntime
+  toolchain_version: "onnxruntime CPU EP"
+  format: onnx
+  execution_provider: cpu
+  execution_precision: fp32
+
+configuration:
+  npu_instances: null
+  npu_cores_per_instance: null
+  npu_freq_mhz: null
+
+benchmark:
+  type: reference
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 224, 224]
+
+performance:
+  fps: null
+  latency: null
+
+metrics:
+  accuracy: 0.685
+  top5_accuracy: 0.901
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+reproduce:
+  steps:
+    - title: Reproduction steps coming soon
+      command: >-
+        Step-by-step ONNX Runtime reproduction instructions are coming soon.
+        In the meantime, please contact your local Renesas representative.
+      kind: note

int8/benchmarks/x5h_ort_npu_12core.yaml

@@ -0,0 +1,65 @@
+# ONNX Runtime + Renesas EP v2.1.0 (MWMX 2.2 ENG4); source: ORT_Renesas_EP_Test_v2.1.0.xlsx (status 2026-09-30)
+# model/config used for the EP run: https://github.com/onnx/models/blob/main/validated/vision/classification/resnet/model/resnet18-v1-7.onnx
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: onnxruntime      # ONNX Runtime with Renesas Execution Provider (+ CPU EP for non-NPU nodes)
+  toolchain_version: "v2.1.0 renesas"   # Renesas EP v2.1.0 (MWMX 2.2 ENG4)
+  format: onnx
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the Renesas EP
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 12
+  npu_freq_mhz: 850        # not stated in the workbook; same 850 MHz as the MWMX and earlier ORT runs
+  npu_offload_pct: 98.0   # share of graph nodes executed on NPU (read by AI-Dashboard)
+  cpu_offload_pct: 2.0   # share of graph nodes falling back to CPU EP
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 224, 224]
+
+performance:
+  latency: 1.729
+  fps: 578.4
+  npu_inference_ms: 0.799
+  portable_pkg_npu_ms: 0.6927
+
+metrics:   # accuracy is a single measurement, independent of core count
+  accuracy: 0.714
+  top5_accuracy: 0.897
+  eval_subset: "1k images"   # not the full validation set
+
+node_distribution:   # partitioning of the graph between NPU and CPU EP
+  total: 50
+  npu: 49
+  cpu: 1
+  qdq_inserted: 1
+  dispatched_to_cpu_compile: null
+  dispatched_to_cpu_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+reproduce:
+  steps:
+    - title: Reproduction steps coming soon
+      command: >-
+        Step-by-step ONNX Runtime reproduction instructions are coming soon.
+        In the meantime, please contact your local Renesas representative.
+      kind: note

int8/benchmarks/x5h_ort_npu_1core.yaml

@@ -0,0 +1,66 @@
+# ONNX Runtime + Renesas EP v2.1.0 (MWMX 2.2 ENG4); source: ORT_Renesas_EP_Test_v2.1.0.xlsx (status 2026-09-30)
+# model/config used for the EP run: https://github.com/onnx/models/blob/main/validated/vision/classification/resnet/model/resnet18-v1-7.onnx
+hardware:
+  vendor: renesas
+  chip: rcar-x5h
+  cpu: arm-cortex-a720
+  npu: npx6-48k
+  npu_count: 2
+  npu_cores: 12
+  npu_default_freq_mhz: 1066
+  accelerator:
+    - npu
+
+runtime:
+  engine: onnxruntime      # ONNX Runtime with Renesas Execution Provider (+ CPU EP for non-NPU nodes)
+  toolchain_version: "v2.1.0 renesas"   # Renesas EP v2.1.0 (MWMX 2.2 ENG4)
+  format: onnx
+  execution_provider: npu
+  execution_precision: int8  # FP32 model is auto-cast to INT8 by the Renesas EP
+
+configuration:
+  npu_instances: 1
+  npu_cores_per_instance: 1
+  npu_freq_mhz: 850        # not stated in the workbook; same 850 MHz as the MWMX and earlier ORT runs
+  npu_offload_pct: 98.0   # share of graph nodes executed on NPU (read by AI-Dashboard)
+  cpu_offload_pct: 2.0   # share of graph nodes falling back to CPU EP
+
+benchmark:
+  type: hil
+  parameters:
+    batch_size: 1
+    input_resolution: [1, 3, 224, 224]
+
+performance:
+  latency: 2.114
+  fps: 473.0
+  npu_inference_ms: 1.476
+  portable_pkg_npu_ms: 1.4836
+  cpu_inference_overhead_ms: 0.053
+
+metrics:   # accuracy is a single measurement, independent of core count
+  accuracy: 0.714
+  top5_accuracy: 0.897
+  eval_subset: "1k images"   # not the full validation set
+
+node_distribution:   # partitioning of the graph between NPU and CPU EP
+  total: 50
+  npu: 49
+  cpu: 1
+  qdq_inserted: 1
+  dispatched_to_cpu_compile: null
+  dispatched_to_cpu_accuracy: null
+
+memory:
+  peak_mb: null
+
+power:
+  avg_w: null
+
+reproduce:
+  steps:
+    - title: Reproduction steps coming soon
+      command: >-
+        Step-by-step ONNX Runtime reproduction instructions are coming soon.
+        In the meantime, please contact your local Renesas representative.
+      kind: note