从 5 月首秀,到 10 月预售,等了小半年,英伟达给 Windows 准备的「最强外援」RTX Spark 终于来了。
10 月 7 日,微软在旧金山举行 Windows 与 Surface 发布会。黄仁勋与纳德拉同台对谈,搭载 RTX Spark N1X 的 Surface Laptop Ultra 和第一批 RTX Spark OEM 机型启动预售,将于 10 月 16 日正式上市。

爱范儿也第一时间拿到了这台 Surface Laptop Ultra,并上手了现场多款 RTX Spark 电脑。从让 WorkBuddy 在本地干活,到直接拉开复杂的虚幻 5 工程,再到用 DLSS 5 玩游戏,这批 Windows AI PC 的实际表现,确实比预想中更扎实。
英伟达把 20 核 CPU、Blackwell 显卡和 128GB 内存打包塞进了这颗芯片,但落到实际的机型感受上,这台由微软操刀的 Surface Laptop Ultra,用起来究竟怎么样?
Surface 的硬件,老黄的芯片
全金属机身、利落平直的线条,仍旧是熟悉的 Surface 味道。正面这块 15 英寸 3:2 触控屏用上了 Mini LED 发光材料,HDR 峰值亮度达到了 2000 尼特,不管是审片调色还是跑虚幻 5 的光影演示,色彩和细节都让人印象深刻。
整机厚度做到了大约 18mm,重量刚好压在 2.0kg。虽然放进双肩包的电脑仓不算吃力,但只要你把它单手拿在手里,沉甸甸的扎实分量会立刻提醒你:它就是一台冲着表现来的随身「AI 超算」。
为了保证 RTX Spark 的表现极致释放,微软还给 Surface Laptop Ultra 上足了散热冗余,机身的厚度和分量显然为此做了让步。
右侧的磁吸 USB-C 接口,是一个让我爱不释手的小外观。
微软把它叫做 Magnetic Connect。插上官方磁吸线,它能像 MagSafe 一样防绊防扯;拔掉磁吸头,它仍旧是个能传数据、能接表明器的满血 Type-C。既兼顾了磁吸的顺手,又没有浪费宝贵的高速接口,这种细节对经常带着电脑四处跑的人格外友好。
右侧除了磁吸口,还留出了 USB-A 3.1、HDMI 2.1、全尺寸 SD 卡槽和 3.5mm 耳机孔;机身左侧则排布有两个满血 USB-C。作为一台移动工作站,它的接口表现确实很合格,出门在外处理工作,也不需要局促地挂着拓展坞。
不过,真正让这台 Surface 产生质变的,还是藏在精致外壳内部的那颗 RTX Spark 芯片。
RTX Spark N1X 平台最高给出 20 核 Grace CPU、6144 个 CUDA 核心的 Blackwell RTX GPU,以及最大 128GB 的统一内存。
官方标称它在 FP4 精度与稀疏计算下,最高 AI 算力能达到惊人的 1 PFLOPS,相当于每秒整整一千万亿次浮点运算。
很多人可能对这串天文数字没有直观概念——这个数字是当下主流轻薄 AI PC(40–50 TOPS 级别)的整整二十多倍,以往这种规模的算力,通常只能在机房里的数据中心或者大型工作站上才能见到。
但更关键的,还是最高兼容 128GB 的统一内存,这才是整台机器的杀手锏。
在传统 PC 上折腾过本地 AI 的人都知道,卡脖子的从来不是显卡算力,而是显存。哪怕买到顶配如 5080 显卡,也就 16GB 显存。
想跑参数稍大一点的模型?对不起,连权重文件都加载不进去,直接就报错闪退了。
而在统一内存架构下,显卡能直接调用整块 128GB 内存池,哪怕扣除掉操作平台和日常软件的开销,剩下留给模型的空间,也足够在本地跑起以往只能放在服务器上的超大模型了。
英伟达的 CUDA 生态,这次 RTX Spark 也能原生兼容。开发者能把原本在台式机或服务器上跑得顺手的整套工具链,无缝地搬到这台随身笔记本上,省去了重新折腾适配环境的麻烦。
无论是模型推理还是 3D 渲染,有成熟的 CUDA 生态兜底,硬件的表现都能无损转化为随时可用的生产力。
在英伟达和微软的号召下,Windows 阵营的 OEM 制造商们也推出了自家的 RTX Spark 机型,让这次新平台的首发阵容格外强悍。
譬如华硕,就在一台仅重 1.48kg、14 英寸的 ProArt P14 上,硬生生塞进了最高 128GB 的统一内存;联想和微星,则推出了能翻折、能手写的二合一 PC;还有戴尔 XPS 16、惠普 OmniBook 等也都推出了 RTX Spark 版本。
显然,这就是有史以来最强的 Windows on Arm 阵容——微软是铁了心要把这个平台扶起来。
智能、超能、全能
在拿到 Surface Laptop Ultra 的这段时间,爱范儿也感受了多个使用情形,其中让人印象最深刻的应用,是 WorkBuddy。
不同于市面上云端的 WorkBuddy,Surface Laptop Ultra 上的最新版本,已经可以下载本地模型,直接调用这台电脑的端侧算力执行 Agent 任务。
实际感受下来,响应和吐字速度都很跟手,部分任务在断网离线状态下也能直接跑完。
WorkBuddy 让人眼前一亮的地方,在于它直接把本地大模型做成了能落地的日常生产力。你不需要懂什么复杂的推理环境规格,它就能直接接管平台权限,调动各种本地工具替你处理文件、跑任务流程。
这就让 AI Agent 类机型,同时具备了端侧和云端算力。
遇到复杂的逻辑推理或者要查最新信息,就可以调用云端的大模型去深度思考;如果是整理本地文档、批量格式转换这种琐碎活,用笔记本自己的算力,就能在本地哐哐开干。
费脑子的重活让云端算,涉及本地文件的跑腿活就在电脑里应对,既能出色地完成任务,又能帮你剩下不少 Token。
尽管当下,WorkBuddy 还不兼容「端云混合」的模式,自动切换云端或本地模型,但 RTX Spark 的浮现,无疑为这样的情形给出了多元的想象力。
统一内存带来的底气,最直观的体现就是本地大模型。像 Qwen 3.5、Qwen 3.8 这些开源模型,现在都能在端侧达成真正的「Token 自由」。
笔者在本地借助 Ollama + Open WebUI 跑了实测:Qwen3.5 35B-A3B(Q4 量化)跑出了接近 60 Token 每秒(59.65 Token 每秒)的持续生成速度,两千多字一口气输出,感受行云流水;换到参数更密集的 Qwen3.8 27B,预热后也能达到 20.98 Token 每秒。
这对于需要跑长链条代码循环的工程,或者需要反复调试提示词的任务来说,在本地随便折腾都不用担心云端按次扣费的账单。而且就算在高铁飞机上断了网,手头的工作也完全不会停摆,商业机密和未公开代码更不需要上传到第三方服务器。
还有一种玩法,是让算力延伸到物理世界产生互动。
在活动现场,爱范儿也感受到了用来演示的 Mirobot 机械臂和各种小机器人,可以直接连着这台电脑做本地调试。
由于没有云端往返的网络延迟,改完一段控制逻辑,桌上的机械臂就能立刻给出物理动作反馈。这种边调边试、所见即所得的感受,对搞具身智能和硬件开发的人来说,应该很有吸引力。
Hugging Face 旗下的 Microduck 小鸭机器人,当下已经开放了训练流程。而借助 Home Assistant,笔记本电脑上的本地模型,甚至能帮你接管家里的智能家居,成为算力不间断的家庭智能中枢。
生产力部分,熟悉的「曼哈顿」也来了。
这是虚幻 5 引擎中的一个城市演示情形,有着极高密度的建筑,并且每一个平面都兼容用 RTX 光追成效开展渲染。
即便如此,整个工程文件仍旧可以流畅运行,这对于需要带着项目见客户、到现场调试思路的工程师来说,能随身打开演示项目,本身就是一种生产力。
除了跑 AI 跑工程,这台 Surface Laptop Ultra 跑游戏也很有一手。
按照英伟达的说法,它的图形表现大致对标移动版 RTX 5070。笔者在现场上手了《NBA 2K27》《毁灭战士:黑暗时代》等游戏,帧率表现相当流畅。
而 DLSS 5 的加入,则让游戏的画面更上一个台阶。它利用游戏画面和运动向量开展三维引导的神经渲染,直接参与画面的塑造、改善光照与材质表现。在《NBA 2K27》里,皮肤的光感、毛发透光与接触阴影,都让人物更有真实感。
光看官方展台的演示还不够,爱范儿也跑了几款主流的 3A 游戏,看看它在日常使用里到底能跑到什么水平。
先看重负载 3A 代表《赛博朋克 2077》(1080p、DLSS 自动、关帧生成):在关闭光追时,平均帧率能达到 67.17 FPS,最低帧也有 51.50 FPS;
在「光线追踪:超级」预设下,平均帧率也有 39.49 FPS,在光影全开的顶格画面里沉浸过剧情,视觉享受仍旧赏心悦目。
向左滑动查看更多内容
偏电竞的《CS 2》中,Surface Laptop Ultra 在 1080P 格外高画质的 Dust2 基准测试里拿下了 109.7 FPS 的平均帧,稍微降低一些画质,达到「电竞级」的表现没有难题。
真正带来惊喜的是《漫威争锋》。笔者这次提前感受到了原生适配 Arm 的版本。
在 1080P 全极高画质(Lumen 全局光照与反射)、DLSS 原生档并启动自动帧生成的设定下,游戏内置测试跑出了平均 171 帧的成绩,激烈团战情形下的 1% Low 也有 138 帧,画面丝滑。
从 3A 单机到竞技网游,这几项实测清晰地划出了一条基准线:它确实具备了正经游戏本的图形底盘。不过,要想让玩家真正放心地把它当主力机打游戏,还要翻过另一座大山。
过去提起 Windows on Arm 玩游戏,大部分玩家的第一反应都是怀疑:转译掉帧不说,再加上反作弊驱动这个老大难难题,很多网游连大厅都进不去。
不过,这次微软和老黄显然下定决心要把这块最大的短板补齐——底层有 Prism 负责转译兜底,像 Steam、XBOX 等平台上的游戏,大部分都能装上也都能流畅跑。
更难啃的反作弊大山(从 Easy Anti-Cheat 到腾讯 ACE、Vanguard),微软、英伟达也在和开发商陆续打通。
微软甚至直接砸出了一张王牌:全球最热门的 FPS 游戏《使命召唤》,将于 2027 年正式登陆 RTX Spark 平台。
虽然生态的彻底补齐仍旧需要时间,但 Windows on Arm 玩游戏,这次似乎真有希望了。
跑得动本地重度大模型,带得出门做三维工程渲染,回到家还能痛痛快快打 3A 游戏。RTX Spark 想做的「AI 时代全能本」,开始有了具象模样。
AI 时代,Windows 终于等来救星
作为 RTX Spark 平台的首秀机型,Surface Laptop Ultra 的感受确实挑不出什么毛病——唯一让人望而却步的,实际上就是售价。
其他 OEM 机型同样不便宜,联想 Yoga Pro 15 RTX Spark 的 64GB 版本售价 29999 元,128GB 顶配版也来到了 42999 元。
但考虑到 5060 游戏本的售价也已经涨到一万块出头,这个售价并非难以理解。
作为一个崭新的、面向将来的 AI PC 平台,RTX Spark 无疑还有很长的路要走。相较于已经在专业生产力领域打出名堂的其他竞品而言,微软和英伟达还要开拓更多的情形、连接更多的应用并且拿出更大的诚意,来让使用者接受 RTX Spark。
某种程度上讲,当下购入一台 RTX Spark 笔记本电脑,无疑是个「战将来」的挑选。
工欲善其事,必先利其器——在 PC 时代,往往会有创业者为了给出更顺畅的服务,砸重金包下一条网络专线;而在 AI PC 时代,拿下的算力越多、成本越低,能创造的机会也就越多。
对于需要在客户驻场部署模型、随时调试硬件机器人,或者 AI 行业的创业者而言,RTX Spark 的吸引力无可替代——最主流的开发生态、足够强劲的算力和一个双肩包就能背走的轻便。
但如果你不是这样的使用者,只是想买一个不错的游戏本或办公本,在 RTX Spark 之外仍旧有多元的挑选。
当下的 RTX Spark,更像是一款专门为极致硬核使用者量身定制的重型装备,它的象征意义和标杆价值,远大过普罗大众的消费热情。
它确实为 Windows 生态开辟了一块至关关键的试验田。软件厂商终于可以放开手脚,围绕大容量统一内存和本地算力去开发真家伙,而不是整天围着聊天框打转,为捉襟见肘的算力发愁。
看着微软和英伟达在台上站在一起,很难不让人联想起当年的 Wintel 黄金时代。
那时候,英特尔负责把处理器表现推到极致,微软负责把 Windows 铺满千家万户,软硬件紧紧咬合在一起滚雪球:软件越多,买电脑的人越多;使用者越多,开发者就越愿意为新硬件写程序。
如今,老黄和纳德拉重聚在 Arm 架构下,想要复制的,显然就是这种能让整个生态重新跑起来的正向循环。
只是这一次,横亘在面前的,不再是某个具体的硬件公司,而是无处不在、吞噬一切的云端大模型们。
而使用者的设备,无论是什么规格,只要能连上网,就能调用全世界最聪明的大模型。
AI 越来越厉害,并不天然构成人们掏出真金白银换电脑的理由。
这也正是 WorkBuddy 这类应用的价值所在。它把本地算力接进了现有的工作流,让使用者能在一次次调用中感受到本地算力的便利。
只有当本地算力真正嵌入日常操作、在无需网络的极端情形下也能应对实际难题, RTX Spark 和 Surface Laptop Ultra 才不会沦为展台上的炫技,而是成为桌面上离不开的生产工具。
英伟达需要继续降低模型在本地部署和调优的门槛,联合更多愿意押注端侧模型的合作伙伴;而微软,则必须证明自家平台管理 Agent 的安全性和稳定性。归根结底,人们在本地部署省下来的时间,不能转头就被糟糕的平台感受浪费掉。
从这个角度来看,Windows 甚至比英伟达更急迫地需要打赢这一仗。如果将来的 AI 统统留在云端,英伟达在数据中心照样日进斗金;但 Windows 却很可能被架空,沦为一个仅仅用来点开浏览器的无聊外壳。
微软必须向所有人证明:当 Agent 真正开始替人类打理日常工作时,个人电脑仍旧是掌控一切的绝对中枢。
实际感受过 Surface Laptop Ultra 之后,爱范儿确信笔者已经看到了 AI 时代全能本的轮廓。它有随身工作站的能力,也保留了创作和游戏的乐趣。
Windows 的救星确实已经赶到了,接下来,就要看微软如何兑现承诺,让人们觉得这笔昂贵的算力投入,真正物有所值了。
本文来源:https://www.ifanr.com/1684442
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
After waiting for half a year from the debut in May to the pre-sale in October, the "strongest foreign aid" RTX Spark prepared by NVIDIA for Windows has finally arrived.
On October 7, Microsoft held a Windows and Surface conference in San Francisco. Jensen Huang spoke with Nadella, and the Surface Laptop Ultra equipped with RTX Spark N1X and the first batch of RTX Spark OEM models will be on sale on October 16.

Ivan also got the Surface Laptop Ultra for the first time, and got a number of RTX Spark computers on site. From having WorkBuddy work locally, to pulling straight out the complex Unreal Project 5, to playing games with DLSS 5, the actual performance of these Windows AI PCs is indeed more solid than expected.
NVIDIA packaged a 20-core CPU, a Blackwell graphics card, and 128GB of memory into this chip, but it fell on the actual model. How does this Microsoft-driven Surface Laptop Ultra work?
The all-metal body and flat lines are still the familiar Surface flavor. The 15-inch 3: 2 touch screen on the front is made of Mini led luminous material, the HDR peak brightness reaches 2000 nits, and the colors and details are impressive, whether it is the color of the review or the illusory 5 light and shadow demonstration.

The thickness of the whole machine is about 18mm, and the weight is just 2.0kg. Although it is not difficult to put it in the computer compartment of the backpack, as long as you hold it in one hand, the solid weight of the heavy weight will immediately remind you that it is a portable "AI supercomputer" for the performance.
In order to ensure the ultimate release of RTX Spark performance, Microsoft also gave the Surface Laptop Ultra enough thermal redundancy, and the thickness and weight of the fuselage obviously gave way to this.
The magnetic USB-C connector on the right is a small look that I love.

Microsoft calls it Magnetic Connect. Plug in the official magnetic suction cable, it can prevent tripping and tearing like MagSafe; pull out the magnetic suction head, it is still a full blood Type-C that can transmit data and connect the indicator. This kind of detail is especially friendly to people who often run around with computers, taking into account the smoothness of magnetic suction without wasting valuable high-speed interfaces.
In addition to the magnetic suction port on the right, USB-A 3.1, HDMI 2.1, full-size SD card slot and 3.5mm headphone hole are left; two full-blooded USB-C are arranged on the left side of the fuselage. As a mobile workstation, its interface performance is indeed very qualified, and it does not need to hang the docking station in a cramped manner to handle work outside.
However, what really changed this Surface was the RTX Spark chip hidden inside the exquisite shell.

The RTX Spark N1X platform gives up to 20 cores of Grace CPUs, 6144 CUDA cores of Blackwell RTX GPUs, and up to 128GB of unified memory.
Officially, it is rated at FP4 accuracy and sparse calculations, and the highest AI computing power can reach a staggering 1 PFLOPS, which is equivalent to a full thousand trillion floating-point operations per second.
Many people may not have an intuitive concept of this string of astronomical numbers - this number is more than twenty times that of the current mainstream lightweight AI PC (40–50 tops level). In the past, the computing power of this scale was usually only seen in the data center or large workstations in the computer room.

But more importantly, it is compatible with up to 128GB of unified memory, which is the killer of the entire machine.
Anyone who has tossed local AI on a traditional PC knows that the card neck is never a graphics card, but a graphics memory. Even if you buy a top-fitting graphics card such as 5080, it is 16GB of graphics memory.

Want to run a slightly larger model? I'm sorry, I can't even load the weight file, so I just reported an error and flashed back.
Under the unified memory architecture, the graphics card can directly call the entire 128GB memory pool. Even after deducting the overhead of the operating platform and daily software, the remaining space left for the model is enough to run locally. The oversized model can only be placed on the server in the past.
NVIDIA's CUDA ecosystem, this time RTX Spark is also natively compatible. Developers can seamlessly move the entire toolchain that originally ran smoothly on a desktop or server to this portable notebook, eliminating the hassle of re-tossing the adaptive environment.

Whether it is model reasoning or 3D rendering, there is a mature CUDA ecological backbone, and the performance of the hardware can be converted into ready-to-use productivity without loss.
Under the call of NVIDIA and Microsoft, OEMs in the Windows camp also launched their own RTX Spark models, making the launch lineup of this new platform particularly strong.
For example, Asus, on a ProArt P14 weighing only 1.48 kg and 14 inches, hardened into a unified memory of up to 128GB; Lenovo and MSI, launched a 2-in-1 PC that can fold and write by hand; and Dell XPS 16, HP OmniBook, etc. also launched RTX Spark version.

Obviously, this is the strongest Windows on Arm lineup ever - Microsoft is keen to lift the platform.
During the time I got the Surface Laptop Ultra, Ivan also felt a number of use cases, of which the most impressive application was WorkBuddy.
Unlike WorkBuddy in the cloud on the market, the latest version on the Surface Laptop Ultra can already download local models and directly call the end-side computing power of this computer to perform Agent tasks.

Actually, the response and spitting speed are very close to each other, and some tasks can be completed directly when the network is offline.
The highlight of WorkBuddy is that it directly makes the local large model into the daily productivity that can land. You don't need to know the specifications of the complex reasoning environment, it can directly take over the platform permissions and mobilize various local tools to process documents and run tasks for you.
This allows the AI Agent class model to have both end-side and cloud computing power.

If you encounter complex logical reasoning or want to check the latest information, you can call the large model in the cloud to think deeply. If you are organizing local documents, batch format conversion, and using your notebook's own computing power, you can do it locally.
The hard work makes the cloud count, and the running errands involving local files are handled in the computer, which can not only complete the task well, but also help you have a lot of tokens left.
Although WorkBuddy is not yet compatible with the "cloud hybrid" model, automatically switching between cloud or on-premises models, the emergence of RTX Spark undoubtedly gives such a situation a diverse imagination.

The most intuitive manifestation of the strength brought by unified memory is the local large model. Open source models such as Qwen 3.5 and Qwen 3.8 can now achieve true "Token Freedom" on the end.
The author ran the actual test with Ollama + Open WebUI locally: Qwen3.5 35B-A3B (Q4 quantization) ran out of the continuous generation speed of nearly 60 tokens per second (59.65 tokens per second), and more than 2,000 words were output one by one to feel the flow of the cloud; switching to the more parameter-dense Qwen3.8 27b can also reach 20.98 tokens per second after preheating.

For projects that need to run long chain code loops, or tasks that require repeated debugging of prompts, there is no need to worry about cloud-based pay-per-view billing. And even if the network is disconnected on a high-speed rail plane, the work at hand will not stop at all, and trade secrets and undisclosed code will not need to be uploaded to third-party servers.
There is also a gameplay that extends computing power to the physical world to create interactions.
At the event, Ivan also felt the Mirobot robot arm and various small robots used for demonstration, which can be directly connected to this computer for local debugging.

Since there is no network latency for round-trip in the cloud, after changing a piece of control logic, the robotic arm on the table can immediately give physical action feedback. This feeling of experimentation and what you see and what you get should be very attractive to people with intelligence and hardware development.
Hugging Face's Microduck duckling robot is now open for training. With Home Assistant, a local model on your laptop can even help you take over your smart home and become an uninterrupted home smart hub.
In the productivity part, the familiar "Manhattan" is also coming.

This is a city demonstration in the Unreal Engine 5, with extremely high-density buildings, and each plane is compatible with rendering with RTX light chase performance.
Even so, the entire project file can still run smoothly, which is a productivity in itself for engineers who need to bring the project to the customer and go to the site to debug the idea.
In addition to running AI running projects, this Surface Laptop Ultra running game is also very good.

According to NVIDIA, its graphics roughly match the mobile version of the RTX 5070. I started playing games such as NBA 2K27 and Doom: Dark Ages on the spot, and the frame rate was quite smooth.
The addition of DLSS 5 takes the game to the next level. It uses the game screen and motion vectors to carry out three-dimensional guided neural rendering, directly participating in the shaping of the picture and improving the lighting and material performance. In NBA 2K27, the light sensation of the skin, the light transmission of the hair, and the contact with the shadows all make the characters more realistic.
It is not enough to see the demonstration of the official booth. Ai Fan 'er also ran several mainstream 3A games to see what level it can run in daily use.

First of all, pay attention to load 3A, which means "Cyberpunk 2077" (1080p, DLSS automatic, off-frame generation): when optical tracking is turned off, the average frame rate can reach 67.17 FPS, and the lowest frame also has 51.50 FPS;
Under the "Ray Tracing: Super" preset, the average frame rate is also 39.49 FPS. Immersed in the plot in the top frame with full light and shadow, the visual enjoyment is still pleasing to the eye.
Swipe left for more

In CS 2, the Surface Laptop Ultra scored an average frame of 109.7 FPS in the 1080p extra-high-quality Dust2 benchmark, slightly reducing the image quality to achieve "esports-level" performance.
The real surprise is Marvel Rivalry. The author felt the version of the native adaptation Arm in advance this time.

Under the setting of 1080p full high image quality (Lumen global illumination and reflection), DLSS native file and automatic frame generation, the in-game test ran an average of 171 frames, and 1% Low also had 138 frames in the fierce group battle situation, and the picture was silky.
From 3A stand-alone to competitive online games, these measurements clearly draw a benchmark line: it does have a graphical chassis with a serious gamebook. However, if you want to make players really feel at ease playing it as a mainframe game, you have to climb over another mountain.
In the past, when it came to playing games on Windows on Arm, most players' first reaction was suspicion: translating frames without saying, coupled with the big problem of anti-cheat driving, many online games could not even enter the lobby.

However, this time Microsoft and Huang are clearly determined to make up for the biggest shortcomings - Prism is responsible for translating the bottom, and most games on platforms such as Steam and Xbox can be installed and run smoothly.
The more difficult anti-cheat mountain (from Easy Anti-Cheat to Tencent ACE and Vanguard), Microsoft and NVIDIA are also working with developers.
Microsoft even smashed a trump card: Call of Duty, the world's most popular FPS game, will officially land on the RTX Spark platform in 2027.

Although it will still take time to completely complete the ecology, Windows on Arm seems to have real hope this time.
Run the local heavy model, bring out the door to do three-dimensional engineering rendering, and go home and play 3A games. RTX Spark wants to do "Almighty Ben in the AI era", which has begun to take on a figurative appearance.
As the debut model of the RTX Spark platform, the feeling of the Surface Laptop Ultra really does not matter - the only thing that is daunting is the price.

Other OEM models are also not cheap, the 64GB version of the Lenovo Yoga Pro 15 RTX Spark costs 29,999 yuan, and the 128GB top version also comes to 42,999 yuan.
But considering that the price of the 5060 gamebook has also risen to more than 10,000 yuan, this price is not difficult to understand.
As a new, future-proof AI PC platform, RTX Spark undoubtedly has a long way to go. Compared to other competitors that have already achieved fame in the field of professional productivity, Microsoft and NVIDIA have to open up more situations, connect more applications, and show more sincerity to make users accept RTX Spark.

In a way, buying an RTX Spark laptop now is definitely a “war of the future” choice.
If you want to do a good job, you must first profit from it - in the PC era, entrepreneurs often smash a network line in order to give a smoother service; in the AI PC era, the more computing power and lower the cost, the more opportunities can be created.
For entrepreneurs who need to deploy models on-site, debug hardware robots at any time, or in the AI industry, the appeal of RTX Spark is irreplaceable - the most mainstream development ecology, powerful enough computing power, and a backpack that can be carried.

But if you are not such a user and just want to buy a good game or workbook, there are still a variety of choices outside of RTX Spark.
Today's RTX Spark is more like a heavy equipment tailored for extreme hardcore users. Its symbolic meaning and benchmarking value far exceed the enthusiasm of the general public.

It does open up a critical field of experimentation for the Windows ecosystem. Software vendors can finally let go and develop real people around large-capacity unified memory and local computing power, rather than spinning around chat boxes all day, worrying about the stretched computing power.
Watching Microsoft and NVIDIA stand together on stage, it's hard not to recall the golden age of Wintel.
At that time, Intel was responsible for pushing the processor performance to the extreme, Microsoft was responsible for spreading Windows to thousands of households, and the software and hardware were tightly intertwined: the more software, the more people bought computers; the more users, the more developers were willing to write programs for new hardware.

Nowadays, Lao Huang and Nadella are reunited under the Arm framework. Obviously, this is the positive cycle that can make the whole ecology run again.
However, this time around, it is no longer a specific hardware company, but the big cloud models that are everywhere and devour everything.
And the user's device, regardless of specification, can call the world's smartest big model as long as it can connect to the Internet.

AI is becoming more and more powerful, and it does not naturally constitute a reason for people to exchange real gold and silver for computers.
That's the value of apps like WorkBuddy. It connects local computing power to existing workflows, allowing users to feel the convenience of local computing power in a single call.
Only when the local computing power is truly embedded in the daily operation and can cope with practical problems without the extreme situation of the network, RTX Spark and Surface Laptop Ultra will not be reduced to stunning skills on the booth, but become indispensable production tools on the desktop.

NVIDIA needs to continue to lower the threshold for local deployment and tuning of the model, and unite with more partners who are willing to bet on the end-side model; Microsoft, on the other hand, must prove the security and stability of its own platform management agent. In the final analysis, the time saved by people deploying locally cannot be wasted by the bad platform feeling.
From this perspective, Windows is even more urgent than NVIDIA to win this battle. If all future AI remains in the cloud, NVIDIA will continue to fight for money in the data center; but Windows is likely to be overhead and reduced to a boring shell used only to open the browser.
Microsoft must prove to everyone that the PC remains the absolute hub of control when the Agent truly begins to take care of the day-to-day work for humans.

After actually experiencing the Surface Laptop Ultra, I am convinced that I have seen the outline of the Almighty Book in the AI era. It has the power of a portable workstation, but also retains the fun of creation and games.
The savior of Windows has indeed arrived. Next, we will see how Microsoft delivers on its promise to make people feel that this costly computing investment is really worth the money.
This article is from: https://www.ifanr.com/1684442 For copyright issues, please write to: 13951431913#139.com or submit a review on the right.