Intel iGPU Inference Deep Dive: Xe2 Architecture, oneDNN & OpenVINO

From Xe2 microarchitecture to oneDNN primitives, from SPIR-V compilation to OpenVINO graph optimization, from performance analysis to GPU+NPU co-inference — a systematic deep dive into AI inference on Intel iGPU.

1. 1

   Xe2 GPU Architecture

   Advanced

   #intel#xe2#gpu-architecture#igpu#lunar-lake#panther-lake
2. 2

   Xe2 Execution Model and Programming Abstractions

   Advanced

   #intel#xe2#simd#sycl#execution-model#workgroup
3. 3

   SPIR-V Compilation and Level Zero Runtime

   Advanced

   #intel#spirv#level-zero#compiler#runtime#jit#aot
4. 4

   oneDNN Primitive System

   Advanced

   #intel#onednn#primitive#memory-format#operator-library
5. 5

   oneDNN GPU Kernel Optimization

   Advanced

   #intel#onednn#kernel-optimization#gemm#xmx#mixed-precision
6. 6

   OpenVINO Graph Optimization Pipeline

   Advanced

   #intel#openvino#graph-optimization#model-compilation#plugin
7. 7

   Intel Model Optimization Stack: Choosing Between Optimum Intel, NNCF, and OpenVINO

   Intermediate

   #intel#optimum#nncf#openvino#quantization#model-conversion
8. 8

   Performance Analysis and Bottleneck Diagnosis

   Advanced

   #intel#performance#profiling#roofline#vtune#bottleneck
9. 9

   NPU Architecture and GPU+NPU Co-Inference

   Advanced

   #intel#npu#openvino#hetero#multi-device#co-inference
10. 10

    LLM Inference on NPU: KV Cache and the Software Stack

    Advanced

    #intel#npu#llm#kv-cache#openvino#npuw#static-shape
11. 11

    NPU Execution Model and the Boundaries of Its Programming Model

    Advanced

    #intel#npu#execution-model#dma#tiling#attention#programming-model#cute
12. 12

    Hands-On: HF → GGUF / ONNX / OpenVINO — Three End-to-End Paths

    Intermediate

    #quantization#model-conversion#hands-on#llama-cpp#onnx#openvino#intel-igpu