文｜硅碳变量，作者｜温宁
在AI PC赛道热闹了两年之后，Windows阵营终于拿出了真正对标Mac生态的旗舰筹码。
微软联合英伟达推出搭载RTX Spark超级芯片的Surface Laptop Ultra，这不再是微软单独下场做硬件的一次常规迭代，而是Windows生态首次拿到英伟达这款一体化顶级SoC，依靠端侧大算力补齐长期存在的短板。
微软Windows与设备执行副总裁帕万·达武卢里周三在旧金山举行的一场公司活动上表示，Surface Laptop Ultra现已开放预购，并将于10月16日开始发货。该电脑配有磁吸USB-C端口，类似苹果的MagSafe充电接口。
Windows借英伟达顶级SoC补齐端侧AI核心短板？
过去两年大量推向市场的AI PC，大多建立在传统PC硬件底座之上，更偏向存量硬件的附加改造。
无论是英特尔、AMD的集成NPU方案，还是高通移植至Windows的Arm版本，核心思路都是在成熟硬件框架之上增加轻量化AI加速单元，很少针对高阶生成式AI任务，完成端侧算力体系的底层重构。
这也使得绝大多数Windows AI PC存在明显局限：端侧专用AI加速单元算力天花板不高，仅适合文本摘要或简单修图这类轻度AI任务。
一旦尝试百亿级大模型本地推理、超长上下文解析、高清图文视频生成、多步骤AI智能体自主执行等高负载场景，算力与内存的约束会快速显现。
这也成为长期以来在高端创作者与开发者市场，Windows生态竞争力不及Mac的关键因素。
苹果依托Apple Silicon自研SoC，实现CPU、GPU与统一内存的深度架构耦合，从硬件底层完成端侧AI全链路优化，为创作者和开发者搭建起稳定高效的本地AI算力底座。
反观Windows阵营，长期受硬件碎片化困扰，跨设备的软硬件协同优化难度更高。在这之前，行业尚未形成一套专门适配高阶AI任务的完整端侧硬件方案，复杂AI任务大多仍要依托云端Copilot完成，本地算力更多只能支撑轻量化AI需求。
英伟达RTX Spark超级芯片的落地，有望改写高端AI PC的竞争逻辑，也意味着微软在AI PC赛道不再独自承担系统层的攻坚工作，联合英伟达与一众整机厂商搭建起Windows高端 AI 生态的核心同盟。
作为英伟达面向个人AI智能体打造的旗舰级超级芯片，RTX Spark整合Grace架构CPU与Blackwell架构RTX GPU，最高可配备128GB统一内存，FP4算力最高可达1Petaflop，能够在本地承载120B参数大模型、支撑百万级Token超长上下文窗口，并原生兼容全套 CUDA生态。
在FP4精度AI推理场景下，这套硬件规格让消费级轻薄本拥有接近专业AI工作站和小型服务器的本地端侧算力，足以承载过去只能在大型设备上运行的大模型任务。
过去多数PC厂商是在芯片方案定型之后开展整机开发，工作集中在主板、外观与散热调校层面，芯片资源调度和底层系统适配的调整空间有限。
而在RTX Spark项目中，微软全程参与平台定义与打磨，从整机散热方案、供电逻辑、功耗调控，延伸到Windows底层算力调度、内存分配和硬件协同机制，进行全维度联合优化。
这套方案的目标是在该平台上让操作系统能够精准高效调用芯片CPU、GPU与统一内存资源，缓解Windows生态长期存在的软硬件适配割裂问题。
对英伟达而言，RTX Spark的价值不只是一款高端PC芯片，它也是英伟达将CUDA生态向高端消费PC场景延伸的关键布局。
通过引入与云端、专业工作站同源的CUDA软件栈到个人终端，开发者可在本机完成AI大模型原型调试、3D渲染、AI视频生成与智能体验证。依托这套环境，开发者可以在本地完成模型原型搭建、小规模微调与部署验证，大幅降低对云端实例的依赖，缩小消费端与云端AI开发环境之间的迁移成本。
依托英伟达的顶级算力与完整CUDA生态，Windows在这条高端硬件线上，补上了此前难以在轻薄形态下运行高阶端侧AI任务的短板，大幅降低对云端AI的调用需求。
微软瞄准创作者基本盘，用差异化体验完成阵地争夺
跳出硬件参数来看，Surface Laptop Ultra在不少体验设计上吸收了Mac已经被市场验证的高端创作设备思路，但微软并没有简单复刻，而是保留Windows生态独有的差异化优势。
最有代表性的是Magnetic Connect磁吸USB-C接口，这是笔记本原厂首次把磁吸结构集成在标准USB-C端口上。设计思路瞄准Mac磁吸充电防拉扯的使用痛点，同时保留USB-C通用拓展能力，替代了微软沿用多年和生态封闭的Surface Connect私有接口，兼顾接口安全与通用性。
除此之外，PixelSense Ultra触控屏峰值HDR亮度可达2000尼特，高于M5 Pro版本MacBook Pro 的1600尼特峰值，同时大尺寸触控板也加入了触觉反馈。
不过微软没有全盘照搬Mac的取舍逻辑，保留了Windows受众看重的实用属性：机身配齐HDMI、USB-A、SD读卡器、3.5毫米耳机口，拓展能力完整，在满足配套线材与规格前提下，最多可同时驱动三台4K显示器；SSD支持用户自行拆装更换，为创作者和AI开发者预留硬件扩容空间。
这套设计取舍，瞄准的核心目标人群十分明确：创作者与AI开发者。
这部分群体恰好是苹果在高端生产力市场里最稳固的基本盘。长期以来，大量视频剪辑、3D建模、AI模型调试从业者选择Mac，不只是芯片算力的优势，更是统一内存架构、整机稳定性、软硬件深度协同共同构成的综合使用体验。
微软的竞争思路也由此清晰：依托RTX Spark带来的完整CUDA生态，以及特定场景下更突出的AI视频生成性能，再搭配对标Mac水准的精致硬件体验，尝试撬动苹果长期把持的高端创作用户群体。
这一系列操作释放的信号十分明确：微软不再满足于守住大众PC市场的基本盘，试图向高端创作和本地AI开发这一苹果优势领地发起正面角逐。
2599美元的起售价低于同档位16英寸MacBook Pro，这套定价策略本质是生态卡位，而非单纯降价让利。
微软想要借助这款标杆产品证明：Windows高端AI PC，能够在本地AI任务上提供持平甚至更强的算力表现，同时保留Windows生态的软件兼容性、CUDA开发能力，以及外设拓展和硬件可升级的自由度。
但放到更大的产业视角看，仅凭这一台旗舰硬件很难直接推动Windows在高端创作者市场重新占据主流。
高端生产力市场的用户迁移，从来不是单一硬件参数或者一款标杆机型就能促成。创作者群体的迁移成本很高：工程文件、项目模板、插件生态、团队协作规范，都深度绑定现有工作流。
很多团队选择Mac不只是看重本地AI算力，还有跨设备协同和长期版本稳定性，以及大量专业创作软件针对Apple Silicon的原生持续优化。这是苹果多年积累形成的软件生态壁垒，短期很难靠一台Windows旗舰抹平。
另一个现实约束来自产品定位，由于RTX Spark属于高端SoC，成本高且产能有限，短期只会落地在少数旗舰机型，很难快速下沉到主流价位段PC。绝大多数普通Windows AI PC依旧沿用英特尔、AMD、高通 NPU 方案，硬件碎片化和软硬件协同参差不齐的老问题不会消失。
也就是说，这套优秀的端侧 AI 底座，短期内无法普惠整个 Windows 阵营。Windows 依然会呈现两极分化：少数旗舰机型拥有媲美 Mac 的一体化 AI 能力，海量主流机型仍然停留在轻量化 NPU 的阶段。
更大的考验落在Windows Agent的落地质量上，这套旗舰硬件所承载的AI价值，最终要依靠系统层智能体来兑现。倘若Agent在多步骤任务调度、本地混合模型推理、应用深度联动上的体验不及预期，再亮眼的芯片规格也容易沦为纸面参数。
而智能体体验的打磨，需要海量软件开发者参与适配、大量第三方应用深度改造，这类生态培育的周期，会远长于硬件研发周期。
从全局看，这台设备的战略定位更像Windows高端AI生态的能力样板。它的目标并非立刻转化大量Mac存量用户，而是向开发者和创作者证明：Windows平台有潜力搭建起承载高阶本地 AI、大模型调试、AI 视频生成的一体化硬件底座。
硬件能不能撑起Windows Agents？
整台Surface Laptop Ultra的硬件设计，核心取舍都围绕一个目标展开：支撑Windows Agent，也就是能够在本地运行的个人AI智能体。
其和当下大家熟悉的Copilot存在明显层级差异。
Agent不再局限于简单问答交互，目标是跨应用自动串联多步骤任务：读取本地文档、调用绘图工具、剪辑视频、管理本地文件，依托多模型协同完成完整复杂工作流，并且尽可能在终端本地完成运算，减少云端依赖，以此保障数据隐私。
这类自动化智能任务对硬件提出了严苛门槛：大容量统一内存、持续稳定的AI推理算力、CPU与GPU之间的高速互联通道，这正是RTX Spark芯片的设计原点。
英伟达还配套推出OpenShell，为智能体搭建本地安全运行沙箱，通过权限策略管控，让Agent在约束之下安全调用本机各类硬件与软件资源。
但硬件就位并不代表Windows Agent就能顺利落地，中间横亘着两层现实鸿沟。
第一层鸿沟，来自软件生态的适配缺口。
算力底座只是基础，Agent想要跑通全链路离不开 Windows底层API、第三方应用、模型调度框架的多方协同。目前Adobe已宣布针对RTX Spark重构Photoshop、Premiere底层架构，优化AI加速能力，但海量中小开发者与垂直行业软件，仍需要漫长的迁移与改造周期。
即便硬件足以承载120B参数大模型，如果应用层缺少配套的Agent联动能力，用户拿到手，它大概率依旧只是一台高性能创作本，而非具备自主执行能力的AI智能体设备。
第二层鸿沟，则落在功耗控制、系统稳定性与真实使用体验层面。
持续运行由大模型驱动的智能体会让整机长期处于高负载状态，对散热设计、系统调度算法以及电池管理形成持续考验。发布会的演示场景属于理想环境，在长时间叠加多任务的真实场景下，是否会出现性能降频和续航大幅缩水，仍有待真实用户长期使用来验证。
硬件只是赋予可能性。Windows Agent能否从发布会Demo，蜕变为日常稳定可用的生产力工具，很大程度上取决于后续Windows系统迭代以及整个软件生态的建设进度。
随之而来的争议是：Surface Laptop Ultra会不会只是一台曲高和寡的标杆演示机型，很难实现大规模放量？
定价已经说明了它的定位：2599美元起，128GB顶配版本达到5899美元，对标专业工作站，核心面向AI开发者和高端创作者，并非瞄准大众消费市场的走量产品。微软也并未指望依靠这款设备吸纳海量用户，它更核心的使命是完成整套平台定义。
微软与英伟达已经官宣，华硕、戴尔、惠普、联想、微星等PC厂商将在今年秋季陆续推出搭载RTX Spark的轻薄本与小型主机，宏碁、技嘉也会后续跟进。
不过大众AI PC的普及，依旧要面对多重约束。
首先是成本问题，RTX Spark作为旗舰 SoC，芯片制造良率和整机BOM成本都偏高，短期内很难下探到万元以内的主流消费区间。
其次是功耗的取舍问题，面向普通用户的轻薄产品必须在AI 算力、续航、噪音之间寻找平衡点，很难复刻旗舰机型的满血算力规格。
最后是应用生态的普及逻辑：如果普通用户无法直观感受到端侧AI带来的实际价值，便很难为AI能力对应的产品溢价买单。
微软联合英伟达推出RTX Spark旗舰PC本质并非一场单纯的硬件参数竞赛，而是 Windows在个人AI时代重构高端生产力生态的一次重要押注。
短期来看，这款旗舰产品更像是开发者与创作者的试验田。中长期成败系于两件事：一是微软能否把Windows Agent从演示特性打磨成稳定可靠的系统原生能力；二是PC厂商和软件开发者能否围绕 RTX Spark 持续丰富配套应用场景。
AI PC的竞争，早已不再停留在有没有NPU的浅层比拼。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Cyclium Silicon Carbon Variable, by Winning
After two years of intense activity on the AI PC track, the Windows camp finally came up with a real flag chip on Mac ecology.
The introduction of Surface Laptop Ultra by Microsoft together with Young Weaverda, which carries the RTX Spark superchips, is no longer a routine integrator of Microsoft's next hardware, but is the first time that the Windows ecology has been given the top integrated SoC of Britain, which has been built up with long-standing shortboards with long-standing side-calculations.
Microsoft Windows and Pavan Davuluri, Executive Vice-President of Equipment, stated at a corporate event in San Francisco on Wednesday that Surface Laptop Ultra was now open for pre-purchase and would commence delivery on 16 October. The computer is equipped with a United StatesB-C port of magnetic ingestion, a MagSafe charging interface similar to apples.
Windows borrows the top SoC of Weeda to fill the side of the AI core?
Over the past two years, AI PCs, which have been heavily marketed, have mostly been built on the base of traditional PC hardware and have been more oriented towards retrofitting inventory hardware.
Whether in Intel, AMD's integrated NPU program, or in Arm's version, which is highly transitory to Windows, the core idea is to add a light quantization AI acceleration unit to the mature hardware framework, with very few low-level generation AI tasks to complete the bottom-up re-engineering of the side-calculations system.
This also makes clear the limitations of the vast majority of Windows AI PCs: the end-side dedicated AI AC unit algorithm is low and suitable for light AI tasks such as text summary or simple drawings.
Once the 10 billion-degree LLM local reasoning, super-long context resolution, high-level graphic video generation, multi-step AIAI agent self-execution, etc. are tried, the constraints on calculation and memory will emerge quickly.
This has also been a key factor in the long-standing market for high-end creators and developers, where Windows is less ecologically competitive than Mac.
Apple has developed the SoC of Apple Silicon to align CPU, GPU and the unified memory deep architecture, optimize the end-side AI full-link from the bottom of the hardware, and create a stable and efficient local AI algorithm base for creators and developers.
Retrospect the Windows camp, which has long been plagued by the fragmentation of hardware, makes it more difficult to coordinate and optimize hardware across equipment. Prior to this, the industry had not developed a full-end hardware programme specifically suited to high-level AI missions, and most of the complex AI tasks were still to be completed by cloud-based Copilot, with more local calculations supporting light quantization AI requirements.
The landing of the British RTX Spark Superchip is expected to rewrite the competitive logic of the high-end AI PC, which also means that Microsoft will no longer be alone in the system-level effort on the AI PC track, and that it will unite with a whole group of machine manufacturers to build a core alliance of Windows high-end AI ecology.
As a flagship superchip for individual AIAI agent, RTX Spark integrates the Grace architecture CPU with Blackwell architecture RTX GPU, with a maximum of 128GB, with a FP4 calculator of up to 1 Petaflop, capable of carrying 120B parameter LLM, supporting the million-grade Token super long context window, and fully compatible with the original CUDA ecology.
In the FP4 precision AI reasoning scenario, this hardware specification gives consumer-grade lightweights local side computing power close to professional AI workstations and small servers sufficient to carry LLM tasks that could only be operated on large equipment in the past.
In the past, most PC manufacturers have been developing the whole machine after the chip programme has been shaped, focusing on the main panel, the appearance and thermology levels, and there is limited space for adjusting the chip resource transfer and bottom systems.
In the RTX Spark project, Microsoft was involved in platform definition and grinding, extending from the whole machine heat programme, power flow logic, power consumption regulation, to Windows bottom computing, memory distribution and hardware synergetic mechanisms, with a combination of full-dimensional optimization.
The objective of the package is to enable the operational system to access the CPU, GPU and the unified memory resources on the platform with precision and efficiency and to mitigate the long-standing problem of fit-out of the Windows ecology.
RTX Spark is not just a high-end PC chip, but it is also a key layout for the extension of the CUDA ecology to the high-end consumer PC scene.
By introducing the CUDA stack to the personal terminal, which is the same source as the cloud, professional workstations, the developers are able to complete the AZLLM prototype debugging, 3D rendering, AID video generation and AI agent validation on this machine. Based on this set of environments, developers can locally complete model prototype construction, small-scale fine-tuning and deployment validation, significantly reducing reliance on cloud examples and reducing the costs of migration between consumer and cloud AI development environments.
With the top computing power and complete CUDA ecology, Windows, on this high-end hardware line, filled with a short panel that had previously made it difficult to run high-end side AI missions in light form, significantly reduced the call demand for cloud AI.
Microsoft aimed at the creator's base disk and used the differential experience to finish the battle.
Out of the hardware parameters, Surface Laptop Ultra has, in a number of empirical designs, absorbed the idea of a high-end creative device that Mac has been certified by the market, but Microsoft does not simply reset it, but preserves the unique differential advantage of Windows ecology.
The most representative is the Magnetic Conect Magnetic USB-C interface, which was the first time that the laptop plant had integrated the magnetic suction structure into a standard USB-C port. Designed to target the mac Magnetic Insorption Charge to protect the use of pull-proof pain points, while retaining the USB-C universal outreach capability, replacing Microsoft ' s multi-year and ecologically closed private interface, which combines security and interoperability.
In addition, the PixelSense Ultra touch screen peak of HDR can be as bright as 2,000 Nits, above the M5 Pro version of MacBook Pro ' s peak of 1600 Nits, and large-scale touchboards also include tactile feedback.
However, Microsoft did not fully follow the Mac's trade-off logic and retained the practical attributes of the Windows audience: the body was fully equipped with HDMI, USB-A, SD card readers, 3.5 mm headphones, full outreach capability, with a maximum of three 4K monitors being driven simultaneously, subject to the fulfilment of the corresponding threads and specifications; and the SSD supported users in dismantling and replacing their own and providing hardware expansion space for creators and AI developers.
The design package is based on a clear target population: creators and AI developers.
This segment of the population happens to be the most solid base of apples in high-productivity markets. For a long time, a large number of video clippings, 3D modelling, and AI model debugging practitioners have chosen Mac not only for the advantage of chip arithmetic, but also for the integrated memory architecture, the integrity of the whole machine, and the combined use experience of software and hardware depths.
Microsoft's competitive thinking is also clear: it tries to pry apples into a long-term high-end creative user group, based on the complete CUDA ecology brought by RTX Spark, and the more prominent AI video generation performance in a given scene, coupled with a fine hardware experience at the Mac level.
The signal from this series of operations is clear: Microsoft is no longer content to hold onto the fundamentals of the popular PC market and is trying to make a positive race for the high-end creation and local AI development of this apple advantage.
The start-up price of $2599 was lower than the 16-inch MacBook Pro of the same slot, and the pricing strategy was essentially eco-calculation rather than a simple discount.
Microsoft would like to use this label product to prove that Windows High-end AI PC can provide even more balanced or better calculator performance on a local AI mission, while preserving Windows eco-compatibility, CUDA development capabilities, and the freedom of outreach and hardware upgrades.
However, from a broader industry perspective, it is difficult for this flagship hardware alone to directly drive Windows back into the mainstream of the high-end creator market.
The migration of users in high-end productivity markets has never been facilitated by single hardware parameters or a sticker. The costs of moving the creator group are high: engineering documents, project templates, plugin ecology, teamwork norms, all of which deeply bind the existing workflow.
Many of the teams chose Mac not only to focus on local AI algorithms, but also on cross-equipment synergetic and long-term version stability, as well as a large number of professional creative software continuously optimised for Apple Silicon. It is a software ecological barrier that has been built up over the years, and in the short term it is difficult to flatten it with a Windows flagship.
Another practical constraint arises from product positioning, as RTX Spark is a high-end SOC with high costs and limited production capacity, leaving only a few flagship aircraft in the short term, making it difficult to sink quickly into the mainstream price segment of PC. The vast majority of ordinary Windows AI PCs continue to follow the Intel, AMD, HD NPU programs, and the old problems of fragmentation of hardware and uneven synergy of hardware and software will not disappear.
In other words, this fine side of the AI base is not going to cover the entire Windows camp in the short term. Windows will continue to be polarized: a small number of flagship aircraft have an integrated AI capability comparable to Mac, and the big mainstream is still in the light of quantization NPU.
The greater test lies in the landing quality of Windows Agent, which ultimately relies on system layer AI agent for the AI values carried by this flagship hardware. If Agent's experience with multi-step tasking, local hybrid model reasoning, application depth connection is less than expected, resonant chip specifications can also become paper-based parameters.
The grinding of the AI agent experience requires the participation of a mass software developer in fit-for-fitting and in-depth retrofitting of a large number of third-party applications, which can take longer periods than hardware development.
Overall, the strategic positioning of the equipment is more like a high-end AI ecological model of Windows. Its objective is not to immediately transform a large number of Mac stock users, but to demonstrate to developers and creators that: The Windows platform has the potential to build local AI, LLM debugging, and AI video-generated integrated hardware floors.
Is the hardware holding up Windows Agents?
The entire hardware design of Surface Laptop Ultra, the core trade-offs are built around a goal: to support Windows Argentina, the individual AIAI agent, capable of running locally.
There is a clear hierarchy between this and the now familiar Copilot.
Agent is no longer limited to simple question-and-answer interaction, but aims to safeguard data privacy by applying an automatic series of multi-step tasks: reading local documents, calling drawing tools, editing videos, managing local files, working with the Todo model to complete a complete complex workflow and, to the extent possible, completing local calculations at the end and reducing cloud dependency.
Such automated smart missions impose stringent thresholds on hardware: large, unified memory, constant and stable AI reasoning, high-speed connectivity between CPU and GPU, which is the point where the RTX Spark chip was designed.
It has also been accompanied by the launch of OpenShell, the construction of a locally safe running sandbox for AI agent, and the control of the permission policy, which has allowed Agent to secure access to the hardware and software resources of the machine under constraints.
But the fact that hardware is in place does not mean that Windows Agent can land well, with two layers of realism in the middle.
The first layer of the divide comes from the adaptation gap in the software ecology.
The substrate is the foundation, and Agent wants to run through the entire chain without the multiple synergies of the bottom API of Windows, third-party applications, model scheduling frameworks. To date, Adobe has announced the optimization of AI Accelerator capabilities for the re-engineering of the Photoshop and Première sub-structures for RTX Spark, but long migration and retrofit cycles are still required for small and medium-sized developers and vertical industry software.
Even if the hardware is sufficient to carry 120B parameter LLM, if the application layer lacks the associated Agent connection capacity and the user gets it, the probability is still that it is a high-performance creation, not an AIAI agent device with autonomous execution capability.
The second divide falls at the level of power consumption control, system stability and real use experience.
The continuous operation of AI agent, driven by LLM, will keep the whole machine on a long-term high load and will continue to test dispersive design, system scheduling algorithms and battery management. The presentation of the publish conference is an ideal environment, and the possibility of a significant reduction in performance frequency and continuity of flight in the real context of a long and multitasking task is still to be verified by the true user.
Hardware just grants possibilities. Windows Agent's ability to move from publish to a daily stabilization of productivity tools available largely depends on the subsequent evolution of the Windows system and the progress of the entire software ecology.
The ensuing controversy is: Is Surface Laptop Ultra just a warp and oligarch model that makes it difficult to achieve large-scale discharges?
Pricing has shown its positioning: US$ 2,599 starts, 128 GB top editions reach US$ 5899, and the target professional workstations are targeted at AI developers and high-end creators, not at the mass consumption market. Microsoft also does not expect to rely on this equipment to absorb large numbers of users, and its more central mission is to complete the full platform definition.
Microsoft and Young Weidar have announced that PC manufacturers such as Wahe-seok, Dale, Hewlett-Packard, Union, Microstar and others will be launching light copies and small hosts carrying RTX Spark this fall, and will follow up with Hong Jian and Tweedy.
However, the spread of AI PC still faces multiple constraints.
The first is the cost, with RTX Spark as the flagship SoC, with high rates of chip production and the cost of completing BOM, making it difficult to reach the mainstream consumption zone within a million dollars in the short term.
The second is the trade-off in power consumption, where light products for ordinary users have to find a balance between AI calculus, continuity, and noise, and it is difficult to reset the calculator specifications of the flag type.
Finally, the application of ecological diffusion logic: if ordinary users are not able to visualize the real value of end-of-side AI, it is difficult to obtain a product premium for AI.
The launch of the RTX Spark flagship PC by Microsoft together with Weda is not simply a hardware parameter competition in essence, but rather an important pledge by Windows to rebuild high productivity ecology in the personal AI era.
In the short term, this flagship product is more like a test field for developers and creators. The short- and long-term success is due to two things: whether Microsoft can grind Windows Agen from demonstration to a stable and reliable system raw capacity; and whether PC manufacturers and software developers can continuously enrich the application landscape around RTX Spark.
AI PC competition has long since ceased to exist on the basis of the existence of NPUs.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.