[{"data":1,"prerenderedAt":225},["ShallowReactive",2],{"pageDataBenchmark":3},{"unique_url":4,"data":5,"content":29,"title":221,"meta_description":222,"page_image":223,"top_id":224,"to_index":22},"benchmark",{"page_name":4,"sections":6},{"hero":7,"top_texts":11,"table_section":13,"architecture_efficiency":138,"solutions_section":170,"blad_posts":188},{"sub_title":8,"title":9,"mobile_title":10},"对比","EyeQ™6H vs. NVIDIA Jetson AGX Orin","EyeQ™6H vs.\u003Cbr \u002F>NVIDIA Jetson AGX Orin",{"text":12},"EyeQ™6H 在边缘侧实现高性能 AI 运算，支持部署深度神经网络（DNN）、 Transformer 模型及多种主流机器学习框架，能够实时完成图像分类、目标检测与图像分割等任务。 作为我们迄今为止最先进的系统集成芯片（SoC）， EyeQ™6H 可将原始传感器数据高效转化为精准且可执行的洞察，赋能驾驶自动化系统做出快速、可靠的决策。",{"table_btn_mobile_img_1":14,"table_btn_mobile_img_2":14,"bottom_text":15,"asterisk_1":16,"asterisk_2":17,"asterisk_3":18,"table_1_title":19,"table_list_1":20,"table_2_title":83,"table_list_2":84},"https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fbenchmark\u002Ftable_mobile_btn_3.png","Superior efficiency and lower latency, at a fraction of the cost.","* Nvidia 与 MLPerf 公布的公开数据。","*EyeQ™6H 支持 Int8 精度计算，并通过先进的 QAT（量化感知训练）和 PTQ（训练后量化）技术，在性能、能效与精度之间实现理想平衡。","** Orin 数据基于 FP16。","图像分类",[21,39,50,59,72],{"id":22,"type":23,"cells":24},1,"header",[25,31,35],{"id":22,"background_color":26,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},"#BAC0D5","#D4D6E5","25","","center",{"id":32,"background_color":33,"border_color":27,"width":28,"bold_text":34,"small_text":29,"regular_text":29,"text_align":30},2,"#C8CDFF","EyeQ™6H",{"id":36,"background_color":26,"border_color":27,"width":37,"bold_text":38,"small_text":29,"regular_text":29,"text_align":30},3,"50","NVIDIA Jetson AGX Orin 64GB*",{"id":32,"type":40,"cells":41},"sub_header",[42,45,48],{"id":22,"width":28,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":44,"text_align":30},"#9EA6C252","TOPS",{"id":32,"width":28,"background_color":46,"border_color":27,"bold_text":47,"small_text":29,"regular_text":29,"text_align":30},"#BBC1FF","34 TOPS",{"id":36,"width":37,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":49,"text_align":30},"275 TOPS",{"id":36,"type":51,"cells":52},"sub_sub_header",[53,55,56],{"id":22,"width":28,"background_color":54,"border_color":27,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},"#9EA6C229",{"id":32,"width":28,"background_color":46,"border_color":27,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},{"id":36,"width":37,"background_color":54,"border_color":27,"bold_text":57,"small_text":29,"regular_text":29,"text_align":58},"单帧延迟(msec)","left",{"id":60,"type":61,"cells":62},4,"simple_cell",[63,66,68,70],{"id":22,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":65,"text_align":30},"#F0F2FF","模式",{"id":32,"background_color":46,"border_color":27,"width":28,"bold_text":67,"small_text":29,"regular_text":29,"text_align":30},"Power",{"id":36,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":69,"text_align":30},"Power (MaxQ)",{"id":60,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":71,"text_align":30},"Unconstrained (MAXN)",{"id":73,"type":61,"cells":74},5,[75,77,79,81],{"id":22,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":76,"text_align":30},"ResNet-50",{"id":32,"background_color":46,"border_color":27,"width":28,"bold_text":78,"small_text":29,"regular_text":29,"text_align":30},"0.5 msec",{"id":36,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":80,"text_align":30},"1.64 msec",{"id":60,"background_color":64,"border_color":27,"width":28,"bold_text":29,"small_text":29,"regular_text":82,"text_align":30},"0.64 msec","Vision Transformer",[85,102,111,125],{"id":22,"type":23,"cells":86},[87,90,92,94,96,98],{"id":22,"background_color":26,"border_color":27,"width":88,"bold_text":89,"small_text":29,"regular_text":29,"text_align":30},"16.666","模型",{"id":32,"background_color":26,"border_color":27,"width":88,"bold_text":91,"small_text":29,"regular_text":29,"text_align":30},"分辨率",{"id":36,"background_color":26,"border_color":27,"width":88,"bold_text":93,"small_text":29,"regular_text":29,"text_align":30},"参数",{"id":60,"background_color":26,"border_color":27,"width":88,"bold_text":95,"small_text":29,"regular_text":29,"text_align":30},"MACs",{"id":73,"background_color":33,"border_color":27,"width":88,"bold_text":34,"small_text":97,"regular_text":29,"text_align":30},"(Int8)*",{"id":99,"background_color":26,"border_color":27,"width":88,"bold_text":100,"small_text":101,"regular_text":29,"text_align":30},6,"NVIDIA Jetson AGX Orin 32GB","(FP16)**",{"id":32,"type":40,"cells":103},[104,105,106,107,108,109],{"id":22,"width":88,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":65,"text_align":30},{"id":32,"width":88,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},{"id":36,"width":88,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},{"id":60,"width":88,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":29,"text_align":30},{"id":73,"width":88,"background_color":46,"border_color":27,"bold_text":67,"small_text":29,"regular_text":29,"text_align":30},{"id":99,"width":88,"background_color":43,"border_color":27,"bold_text":29,"small_text":29,"regular_text":110,"text_align":30},"Unconstrained",{"id":36,"type":61,"cells":112},[113,115,117,119,121,123],{"id":22,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":114,"text_align":30},"EfficientViT-B1",{"id":32,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":116,"text_align":30},"224x224",{"id":36,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":118,"text_align":30},"9.1M",{"id":60,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":120,"text_align":30},"0.52G",{"id":73,"background_color":46,"border_color":27,"width":88,"bold_text":122,"small_text":29,"regular_text":29,"text_align":30},"0.564 msec",{"id":99,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":124,"text_align":30},"1.48 msec",{"id":60,"type":61,"cells":126},[127,129,130,132,134,136],{"id":22,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":128,"text_align":30},"EfficientViT-B2",{"id":32,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":116,"text_align":30},{"id":36,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":131,"text_align":30},"24M",{"id":60,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":133,"text_align":30},"1.6G",{"id":73,"background_color":46,"border_color":27,"width":88,"bold_text":135,"small_text":29,"regular_text":29,"text_align":30},"0.932 msec",{"id":99,"background_color":64,"border_color":27,"width":88,"bold_text":29,"small_text":29,"regular_text":137,"text_align":30},"2.63 msec",{"title":139,"text":140,"sub_title":141,"sub_title_bold":142,"sub_text":143,"top_arrow":144,"bottom_arrow":145,"left_arrow":146,"left_arrow_text":147,"right_arrow":148,"right_arrow_text":149,"img_list":150},"高效架构设计","凭借独特且高度高效的多元加速器架构， EyeQ™ 在低功耗的前提下，实现了业界领先的计算机视觉性能。","异构计算","为每项任务匹配最合适的计算内核","从通用 CPU 到高计算密度的加速器，包括深度学习神经网络， 能灵活调度不同类型的处理单元，最大化计算效率与性能表现。","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Feyeq_arch_top_arrow2.svg","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Feyeq_arch_bottom_arrow2.svg","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Feyeq_arch_mobile_left_arrow.svg","GENERAL-PURPOSE COMPUTE","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Feyeq_arch_mobile_right_arrow.svg","DL PERFORMANCE \u002F COMPUTE DENSITY",[151,154,158,162,166],{"id":22,"title":152,"sub_title":153,"sub_text":29},"CPU","Central Processing Unit",{"id":32,"title":155,"sub_title":156,"sub_text":157},"MPC","Multi-threaded Processor Cluster","More versatile than a GPU, and with higher efficiency than any CPU",{"id":36,"title":159,"sub_title":160,"sub_text":161},"VMP","Vector Microcode Processor","A wide vector (VLIW and SIMD) machine with exceptional performance for short integral types common in computer vision and deep learning algorithms.",{"id":60,"title":163,"sub_title":164,"sub_text":165},"PMA","Programmable Macro Array","A CGRA dataflow machine. Its unique architecture delivers outstanding performance for dense computer vision and deep learning algorithms that are unachievable in classic DSP architecture.",{"id":73,"title":167,"sub_title":168,"sub_text":169},"XNN","Deep Learning Accelerator","Dedicated high-performance AI engine. The main source of horse power for convolutional neural networks.",{"title":171,"list":172},"探索由 EyeQ™ 驱动的解决方案",[173,178,183],{"id":22,"img":174,"text":175,"path":176,"btn_text":177},"https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fbenchmark\u002Fsuper_vision_2.png","驾驶员运动脱离 | 注视","\u002Fsolutions\u002Fsuper-vision\u002F","Mobileye SuperVision™",{"id":32,"img":179,"text":180,"path":181,"btn_text":182},"https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fbenchmark\u002Fchauffer_2.png","驾驶员运动脱离 | 驾驶员视觉脱离","\u002Fsolutions\u002Fchauffeur\u002F","Mobileye Chauffeur™",{"id":36,"img":184,"text":185,"path":186,"btn_text":187},"https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fbenchmark\u002Fdrive_2.png","无需驾驶员","\u002Fsolutions\u002Fdrive\u002F","Mobileye Drive™",[189,198,205,213],{"id":190,"url":191,"title":192,"type":193,"image":194,"publish_date":195,"author_name":196,"img_alt":197},279,"the-mobileye-safety-methodology-for-fully-autonomous-driving","The Mobileye Safety Methodology for Fully Autonomous Driving","blog","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002F114f82911350ead2e8d23f09ea3935a9_1732626715488.jpg","2024-11-26T07:00:00.000Z","Prof. Shai Shalev-Shwartz and Prof. Amnon Shashua","Photo credit: VW",{"id":199,"url":200,"title":201,"type":193,"image":202,"publish_date":203,"author_name":29,"img_alt":204},280,"driving-ai","5 Takeaways from Mobileye's AI day ","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002F9d5275f3635f6d26bd00f9edf597d298_1732784251661.jpg","2024-11-27T07:00:00.000Z","Prof. Shai Shalev-Shwartz and Prof. Amnon Shashua President and CEO at Driving AI",{"id":206,"url":207,"title":208,"type":193,"image":209,"publish_date":210,"author_name":211,"img_alt":212},261,"autonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology","Autonomous decisions: The bias-variance tradeoff in self-driving technology","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002Ff8167f28d4d800bd54b3e04c9fda5a5d_1715717822589.jpg","2024-05-15T07:00:00.000Z","Prof. Amnon Shashua and Prof. Shai Shalev-Shwartz","Monolithic versus compound AI systems in LLMs and autonomous driving.",{"id":214,"url":215,"title":216,"type":217,"image":218,"publish_date":219,"author_name":29,"img_alt":220},258,"mobileye-eyeq6-lite-launches-to-speed-adas-upgrades-worldwide","Mobileye EyeQ™6 Lite launches to speed ADAS upgrades globally","news","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002Ffe5c0db6e4aa39bad155323a8a6a3e4c_1713310858467.jpg","2024-04-17T07:00:00.000Z","The EyeQ6 Lite and an 8-megapixel camera that enables broader data for ADAS applications","对比结果 | Mobileye EyeQ™6H 与 Jetson AGX Orin","低延迟、高效能、兼具成本效应优势。","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fbenchmark_social.jpg",53,1784714284011]