{ "type": "excalidraw", "version": 2, "source": "https://excalidraw.com", "elements": [ { "id": "title", "type": "text", "x": 210, "y": 25, "width": 980, "height": 35, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9070, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "NVIDIA AI Infra 面试考察地图：今天重点复习版", "fontSize": 28, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "NVIDIA AI Infra 面试考察地图：今天重点复习版", "autoResize": false, "lineHeight": 1.25 }, { "id": "subtitle", "type": "text", "x": 140, "y": 72, "width": 1120, "height": 24, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9071, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "岗位关键词：GPU kernel、LLM inference serving、TensorRT-LLM / Triton / Dynamo、分布式通信、性能诊断、系统 tradeoff", "fontSize": 18, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "岗位关键词：GPU kernel、LLM inference serving、TensorRT-LLM / Triton / Dynamo、分布式通信、性能诊断、系统 tradeoff", "autoResize": false, "lineHeight": 1.25 }, { "id": "k0", "type": "rectangle", "x": 40, "y": 125, "width": 410, "height": 355, "angle": 0, "strokeColor": "#748ffc", "backgroundColor": "#dbe4ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 30, "groupIds": [], "roundness": { "type": 3 }, "seed": 9072, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "k0_label", "type": "text", "x": 58, "y": 137, "width": 374, "height": 26, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9073, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "一、CUDA / Kernel 基础", "fontSize": 20, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "一、CUDA / Kernel 基础", "autoResize": false, "lineHeight": 1.25 }, { "id": "k00", "type": "rectangle", "x": 70, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#a5d8ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9074, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "k00_text", "type": "text", "x": 82, "y": 197, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9075, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "线程映射\ngrid/block/warp\none block per row", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "线程映射\ngrid/block/warp\none block per row", "autoResize": false, "lineHeight": 1.25 }, { "id": "k01", "type": "rectangle", "x": 255, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#b2f2bb", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9076, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "k01_text", "type": "text", "x": 267, "y": 197, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9077, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "访存优化\ncoalescing\nsmem / bank conflict", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "访存优化\ncoalescing\nsmem / bank conflict", "autoResize": false, "lineHeight": 1.25 }, { "id": "k02", "type": "rectangle", "x": 70, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffd8a8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9078, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "k02_text", "type": "text", "x": 82, "y": 297, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9079, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Reduce 模板\nwarp shuffle\nblock reduce", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Reduce 模板\nwarp shuffle\nblock reduce", "autoResize": false, "lineHeight": 1.25 }, { "id": "k03", "type": "rectangle", "x": 255, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#d0bfff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9080, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "k03_text", "type": "text", "x": 267, "y": 297, "width": 141, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9081, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "常写算子\nsoftmax\nlayernorm\ngemv/gemm", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "常写算子\nsoftmax\nlayernorm\ngemv/gemm", "autoResize": false, "lineHeight": 1.25 }, { "id": "knote", "type": "text", "x": 70, "y": 390, "width": 340, "height": 22, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9082, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "现场高频：写 row-wise softmax / layernorm；解释 O(N^2) 重复扫描；判断瓶颈是 memory-bound 还是 compute-bound。", "fontSize": 16, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "现场高频：写 row-wise softmax / layernorm；解释 O(N^2) 重复扫描；判断瓶颈是 memory-bound 还是 compute-bound。", "autoResize": false, "lineHeight": 1.25 }, { "id": "l0", "type": "rectangle", "x": 490, "y": 125, "width": 410, "height": 355, "angle": 0, "strokeColor": "#748ffc", "backgroundColor": "#e5dbff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 30, "groupIds": [], "roundness": { "type": 3 }, "seed": 9083, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "l0_label", "type": "text", "x": 508, "y": 137, "width": 374, "height": 26, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9084, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "二、LLM Inference 栈", "fontSize": 20, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "二、LLM Inference 栈", "autoResize": false, "lineHeight": 1.25 }, { "id": "l00", "type": "rectangle", "x": 520, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#a5d8ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9085, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "l00_text", "type": "text", "x": 532, "y": 197, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9086, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Prefill vs Decode\nprefill compute-heavy\ndecode memory/latency-heavy", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Prefill vs Decode\nprefill compute-heavy\ndecode memory/latency-heavy", "autoResize": false, "lineHeight": 1.25 }, { "id": "l01", "type": "rectangle", "x": 705, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#b2f2bb", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9087, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "l01_text", "type": "text", "x": 717, "y": 197, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9088, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "KV Cache\npaged KV\nreuse/offload", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "KV Cache\npaged KV\nreuse/offload", "autoResize": false, "lineHeight": 1.25 }, { "id": "l02", "type": "rectangle", "x": 520, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffd8a8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9089, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "l02_text", "type": "text", "x": 532, "y": 297, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9090, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Batching\ncontinuous / in-flight\ntail latency", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Batching\ncontinuous / in-flight\ntail latency", "autoResize": false, "lineHeight": 1.25 }, { "id": "l03", "type": "rectangle", "x": 705, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#d0bfff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9091, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "l03_text", "type": "text", "x": 717, "y": 297, "width": 141, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9092, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Optimizations\nquantization\nspeculative decoding\ncustom attention", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Optimizations\nquantization\nspeculative decoding\ncustom attention", "autoResize": false, "lineHeight": 1.25 }, { "id": "lnote", "type": "text", "x": 520, "y": 390, "width": 340, "height": 22, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9093, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "NVIDIA 语境优先讲 TensorRT-LLM：custom attention、in-flight batching、paged KV cache、FP8/FP4/INT4、spec decode。", "fontSize": 16, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "NVIDIA 语境优先讲 TensorRT-LLM：custom attention、in-flight batching、paged KV cache、FP8/FP4/INT4、spec decode。", "autoResize": false, "lineHeight": 1.25 }, { "id": "d0", "type": "rectangle", "x": 940, "y": 125, "width": 410, "height": 355, "angle": 0, "strokeColor": "#748ffc", "backgroundColor": "#d3f9d8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 30, "groupIds": [], "roundness": { "type": 3 }, "seed": 9094, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "d0_label", "type": "text", "x": 958, "y": 137, "width": 374, "height": 26, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9095, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "三、分布式与部署", "fontSize": 20, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "三、分布式与部署", "autoResize": false, "lineHeight": 1.25 }, { "id": "d00", "type": "rectangle", "x": 970, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#a5d8ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9096, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "d00_text", "type": "text", "x": 982, "y": 197, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9097, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Parallelism\nTP / PP / DP / EP\nMoE routing", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Parallelism\nTP / PP / DP / EP\nMoE routing", "autoResize": false, "lineHeight": 1.25 }, { "id": "d01", "type": "rectangle", "x": 1155, "y": 185, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#b2f2bb", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9098, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "d01_text", "type": "text", "x": 1167, "y": 197, "width": 141, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9099, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "NCCL\nall-reduce\nall-gather\nreduce-scatter", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "NCCL\nall-reduce\nall-gather\nreduce-scatter", "autoResize": false, "lineHeight": 1.25 }, { "id": "d02", "type": "rectangle", "x": 970, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffd8a8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9100, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "d02_text", "type": "text", "x": 982, "y": 297, "width": 141, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9101, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Serving System\nTriton Server\nNIM / TensorRT-LLM", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Serving System\nTriton Server\nNIM / TensorRT-LLM", "autoResize": false, "lineHeight": 1.25 }, { "id": "d03", "type": "rectangle", "x": 1155, "y": 285, "width": 165, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#d0bfff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9102, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "d03_text", "type": "text", "x": 1167, "y": 297, "width": 141, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9103, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Dynamo\ndisaggregated serving\nsmart routing\nKV cache tiers", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Dynamo\ndisaggregated serving\nsmart routing\nKV cache tiers", "autoResize": false, "lineHeight": 1.25 }, { "id": "dnote", "type": "text", "x": 970, "y": 390, "width": 340, "height": 22, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9104, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "系统题常问：吞吐和延迟如何权衡；prefill/decode 分离为什么有效；多 GPU 通信如何影响 batch 和并发。", "fontSize": 16, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "系统题常问：吞吐和延迟如何权衡；prefill/decode 分离为什么有效；多 GPU 通信如何影响 batch 和并发。", "autoResize": false, "lineHeight": 1.25 }, { "id": "p0", "type": "rectangle", "x": 40, "y": 525, "width": 640, "height": 315, "angle": 0, "strokeColor": "#748ffc", "backgroundColor": "#fff3bf", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 30, "groupIds": [], "roundness": { "type": 3 }, "seed": 9105, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p0_label", "type": "text", "x": 58, "y": 537, "width": 604, "height": 26, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9106, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "四、性能分析与 Debug", "fontSize": 20, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "四、性能分析与 Debug", "autoResize": false, "lineHeight": 1.25 }, { "id": "p00", "type": "rectangle", "x": 80, "y": 590, "width": 180, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#a5d8ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9107, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p00_text", "type": "text", "x": 92, "y": 602, "width": 156, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9108, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Nsight Systems\n看端到端 timeline\nCPU/GPU overlap", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Nsight Systems\n看端到端 timeline\nCPU/GPU overlap", "autoResize": false, "lineHeight": 1.25 }, { "id": "p01", "type": "rectangle", "x": 290, "y": 590, "width": 180, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#b2f2bb", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9109, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p01_text", "type": "text", "x": 302, "y": 602, "width": 156, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9110, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Nsight Compute\n看单 kernel\nSM/DRAM/L2/occupancy", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Nsight Compute\n看单 kernel\nSM/DRAM/L2/occupancy", "autoResize": false, "lineHeight": 1.25 }, { "id": "p02", "type": "rectangle", "x": 500, "y": 590, "width": 140, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffd8a8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9111, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p02_text", "type": "text", "x": 512, "y": 602, "width": 116, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9112, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "Roofline\n算术强度\n瓶颈定位", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "Roofline\n算术强度\n瓶颈定位", "autoResize": false, "lineHeight": 1.25 }, { "id": "p03", "type": "rectangle", "x": 80, "y": 705, "width": 180, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#d0bfff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9113, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p03_text", "type": "text", "x": 92, "y": 717, "width": 156, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9114, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "指标解释\nachieved occupancy\nwarp stall\nmemory throughput", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "指标解释\nachieved occupancy\nwarp stall\nmemory throughput", "autoResize": false, "lineHeight": 1.25 }, { "id": "p04", "type": "rectangle", "x": 290, "y": 705, "width": 180, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#c3fae8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9115, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p04_text", "type": "text", "x": 302, "y": 717, "width": 156, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9116, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "优化动作\nfuse kernels\n减少 H2D/D2H\n改 layout", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "优化动作\nfuse kernels\n减少 H2D/D2H\n改 layout", "autoResize": false, "lineHeight": 1.25 }, { "id": "p05", "type": "rectangle", "x": 500, "y": 705, "width": 140, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffc9c9", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9117, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "p05_text", "type": "text", "x": 512, "y": 717, "width": 116, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9118, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "验证\nprofile 前后对比\n正确性测试", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "验证\nprofile 前后对比\n正确性测试", "autoResize": false, "lineHeight": 1.25 }, { "id": "pa0", "type": "arrow", "x": 260, "y": 625, "width": 30, "height": 0, "angle": 0, "strokeColor": "#495057", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9119, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "endArrowhead": "arrow", "startArrowhead": null, "points": [ [ 0, 0 ], [ 30, 0 ] ], "lastCommittedPoint": null, "startBinding": null, "endBinding": null }, { "id": "pa1", "type": "arrow", "x": 470, "y": 625, "width": 30, "height": 0, "angle": 0, "strokeColor": "#495057", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9120, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "endArrowhead": "arrow", "startArrowhead": null, "points": [ [ 0, 0 ], [ 30, 0 ] ], "lastCommittedPoint": null, "startBinding": null, "endBinding": null }, { "id": "pa2", "type": "arrow", "x": 260, "y": 740, "width": 30, "height": 0, "angle": 0, "strokeColor": "#495057", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9121, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "endArrowhead": "arrow", "startArrowhead": null, "points": [ [ 0, 0 ], [ 30, 0 ] ], "lastCommittedPoint": null, "startBinding": null, "endBinding": null }, { "id": "pa3", "type": "arrow", "x": 470, "y": 740, "width": 30, "height": 0, "angle": 0, "strokeColor": "#495057", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9122, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "endArrowhead": "arrow", "startArrowhead": null, "points": [ [ 0, 0 ], [ 30, 0 ] ], "lastCommittedPoint": null, "startBinding": null, "endBinding": null }, { "id": "q0", "type": "rectangle", "x": 720, "y": 525, "width": 630, "height": 315, "angle": 0, "strokeColor": "#748ffc", "backgroundColor": "#ffe3e3", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 30, "groupIds": [], "roundness": { "type": 3 }, "seed": 9123, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "q0_label", "type": "text", "x": 738, "y": 537, "width": 594, "height": 26, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9124, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "五、今天面试快问快答准备", "fontSize": 20, "fontFamily": 5, "textAlign": "left", "verticalAlign": "middle", "containerId": null, "originalText": "五、今天面试快问快答准备", "autoResize": false, "lineHeight": 1.25 }, { "id": "q00", "type": "rectangle", "x": 760, "y": 590, "width": 260, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#a5d8ff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9125, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "q00_text", "type": "text", "x": 772, "y": 602, "width": 236, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9126, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "1. 写代码\nsoftmax / layernorm / reduce\n讲线程映射和同步点", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "1. 写代码\nsoftmax / layernorm / reduce\n讲线程映射和同步点", "autoResize": false, "lineHeight": 1.25 }, { "id": "q01", "type": "rectangle", "x": 1050, "y": 590, "width": 260, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#b2f2bb", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9127, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "q01_text", "type": "text", "x": 1062, "y": 602, "width": 236, "height": 60, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9128, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "2. 讲系统\nLLM serving 请求生命周期\nprefill -> decode -> stream tokens", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "2. 讲系统\nLLM serving 请求生命周期\nprefill -> decode -> stream tokens", "autoResize": false, "lineHeight": 1.25 }, { "id": "q02", "type": "rectangle", "x": 760, "y": 705, "width": 260, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#ffd8a8", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9129, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "q02_text", "type": "text", "x": 772, "y": 717, "width": 236, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9130, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "3. 讲 tradeoff\n吞吐 vs 延迟\n显存 vs 复用\n编译优化 vs 灵活性", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "3. 讲 tradeoff\n吞吐 vs 延迟\n显存 vs 复用\n编译优化 vs 灵活性", "autoResize": false, "lineHeight": 1.25 }, { "id": "q03", "type": "rectangle", "x": 1050, "y": 705, "width": 260, "height": 70, "angle": 0, "strokeColor": "#1e1e1e", "backgroundColor": "#d0bfff", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": { "type": 3 }, "seed": 9131, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false }, { "id": "q03_text", "type": "text", "x": 1062, "y": 717, "width": 236, "height": 80, "angle": 0, "strokeColor": "#374151", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9132, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "4. 讲工程\n如何定位慢请求\n如何设计 benchmark\n如何灰度上线", "fontSize": 16, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "4. 讲工程\n如何定位慢请求\n如何设计 benchmark\n如何灰度上线", "autoResize": false, "lineHeight": 1.25 }, { "id": "bottom", "type": "text", "x": 130, "y": 885, "width": 1100, "height": 24, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "roundness": null, "seed": 9133, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "临场回答模板：先定义目标和约束 -> 给出线程/系统映射 -> 判断瓶颈 -> 说明优化动作 -> 说验证指标。不要只背名词，要把 latency、throughput、memory footprint、correctness 连起来。", "fontSize": 18, "fontFamily": 5, "textAlign": "center", "verticalAlign": "middle", "containerId": null, "originalText": "临场回答模板：先定义目标和约束 -> 给出线程/系统映射 -> 判断瓶颈 -> 说明优化动作 -> 说验证指标。不要只背名词，要把 latency、throughput、memory footprint、correctness 连起来。", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-bg", "type": "rectangle", "x": 80, "y": 989.0, "width": 1280, "height": 300, "angle": 0, "strokeColor": "#64748b", "backgroundColor": "#e2e8f0", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 70, "groupIds": [], "seed": 93001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-title", "type": "text", "x": 108, "y": 1011.0, "width": 1224, "height": 38, "angle": 0, "strokeColor": "#1e40af", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "练习题与面试题补充：GPU/AI Infra 主题", "fontSize": 24, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "练习题与面试题补充：GPU/AI Infra 主题", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-practice-box", "type": "rectangle", "x": 108, "y": 1067.0, "width": 598.0, "height": 190, "angle": 0, "strokeColor": "#2563eb", "backgroundColor": "#bfdbfe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 93002, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-practice-text", "type": "text", "x": 130, "y": 1087.0, "width": 575.0, "height": 150, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "练习题\n1. 用自己的话重画本图的数据流或控制流\n2. 补一个最小可验证实验, 明确输入、输出和指标\n3. 列出 3 个容易误解的边界条件并给出反例", "fontSize": 17, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "练习题\n1. 用自己的话重画本图的数据流或控制流\n2. 补一个最小可验证实验, 明确输入、输出和指标\n3. 列出 3 个容易误解的边界条件并给出反例", "autoResize": false, "lineHeight": 1.25 }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-interview-box", "type": "rectangle", "x": 734.0, "y": 1067.0, "width": 598.0, "height": 190, "angle": 0, "strokeColor": "#7c3aed", "backgroundColor": "#ddd6fe", "fillStyle": "solid", "strokeWidth": 2, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 93003, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "roundness": { "type": 3 } }, { "id": "practice-interview-panel-nvidia-ai-infra面试考察地图-今日复习-interview-text", "type": "text", "x": 756.0, "y": 1087.0, "width": 575.0, "height": 150, "angle": 0, "strokeColor": "#243042", "backgroundColor": "transparent", "fillStyle": "solid", "strokeWidth": 1, "strokeStyle": "solid", "roughness": 1, "opacity": 100, "groupIds": [], "seed": 91001, "version": 1, "isDeleted": false, "boundElements": null, "updated": 1, "link": null, "locked": false, "text": "面试题\n1. 这个主题解决的核心瓶颈是什么\n2. 关键抽象如何映射到硬件或系统组件\n3. 如果线上结果不符合预期, 你会先看哪些证据", "fontSize": 17, "fontFamily": 5, "textAlign": "left", "verticalAlign": "top", "containerId": null, "originalText": "面试题\n1. 这个主题解决的核心瓶颈是什么\n2. 关键抽象如何映射到硬件或系统组件\n3. 如果线上结果不符合预期, 你会先看哪些证据", "autoResize": false, "lineHeight": 1.25 } ], "appState": { "gridSize": null, "viewBackgroundColor": "#ffffff" }, "files": {} }