RTX Spark laptops have arrived, aimed at the x86 halo

Microsoft and Nvidia opened preorders for the first laptops built on RTX Spark, an Arm-based superchip that shares its silicon with DGX Spark. Led by the $2,599 Surface Laptop Ultra, the systems pair high performance, CUDA support, and a premium design, aimed head-to-head at x86 premium, enthusiast, and gaming laptops.

Software support and workstation readiness are the questions to watch. Microsoft Surface Laptop Ultra. (Source: Microsoft) Microsoft opened preorders on October 7 for the first laptops built on RTX Spark, Nvidia’s Arm-based superchip for Windows PCs. The announcement came at a Windows event in San Francisco.

The lineup is led by the Microsoft Surface Laptop Ultra, whose pricing starts at $2,599, with partner systems from Asus, Dell, HP, Lenovo, and MSI. The OEM launch is October 16. These laptops are positioned head-to-head with x86 premium, enthusiast, and even gaming laptops.

They carry features that, until now, were seen only in high-performance laptops: a healthy CPU core count, a high-end GPU, full support for the Nvidia software ecosystem including CUDA, and premium platform touches such as bright HDR displays and, in the Surface case, a magnetic power connector.

Agentic AI workflows are the main use that Nvidia and Microsoft pitched for a system like this, including the interception of AI with gaming and creation. Figure 1. RTX Spark initial lineup of OEM partners. (Source: Microsoft) What RTX Spark is RTX Spark pairs an Arm CPU developed with MediaTek, with an Nvidia GPU.

It is the same silicon architecture as DGX Spark, with the same core counts and GPU at the top, though the power operating point is adjusted to suit a laptop’s thermal and power needs. Nvidia CEO Jensen Huang described it on stage as two giant chips fused together, and he put its performance at 1 PFLOPS (FP4), the same low-precision basis as the DGX Spark claim.

Task Manager in a Microsoft demo identified the GPU as a 6,144-core Blackwell part, and that is the same core count as a desktop RTX 5070, adjusted for mobile power envelopes, providing an idea of the performance tier—and it’s quite high. More detail specifications will follow soon.

The top-tier configuration of the Nvidia RTX Spark delivers roughly 1,000 TOPS (trillions of operations per second) of local AI compute. Nvidia officially rates this hardware package at 1 PFLOPS of FP4 AI performance with sparsity. This immense throughput is driven by its 5th-generation Tensor Cores embedded inside the custom Blackwell RTX GPU.

The silicon is only half of the story. Even if it brings discrete-level performance to an integrated GPU, RTX Spark also gives that silicon multiples of the memory access. A discrete card built on the same RTX 5070 silicon comes with 12 GB of memory.

RTX Spark can have up to 128 GB of memory, thanks to its unified memory architecture, where the CPU and GPU share one pool. For AI work, that translates into bigger and more capable models, with larger context windows, running locally on a laptop. Huang framed the design as a unification of everything Nvidia sells.

“For the first time, we took all the capabilities, what used to be in a game PC and a laptop, in a workstation, in a supercomputer in the cloud, [and] unified the architecture and made it run on one superchip,” Huang said. Huang added that more than 4,000 engineering years went into the project, and that Nvidia has tested more than 1,200 applications for performance and compatibility, with more software support to come.

The Surface Laptop Ultra Microsoft called the Surface Laptop Ultra the most powerful Surface it has built, aimed at developers and creators. Table 1 lists what the company claimed. Every figure in the table is a Microsoft claim.

Few testing methodologies or settings were disclosed, so independent testing will be necessary to validate all of them. Table 1. Surface Laptop Ultra price and claims. (Source: Microsoft) Figure 2.

Microsoft Surface Laptop Ultra. (Source: Microsoft) The demo targeted the sustained load question. Microsoft showed an intensive GitHub Copilot developer workflow, a film being mastered in Blackmagic Design’s DaVinci Resolve, and Gears of War , released the day before, running at 60 fps, all on the same system.

Microsoft’s narration described the workloads as running together, but the demo did not show them running at the same time. It also compared Windows on RTX Spark devices with the MacBook Pro M5 Pro, claiming up to 2.1 times faster time to first token, up to 4.3 times faster image generation, and up to 6.2 times faster video generation.

Microsoft did not say which models, configurations, or benchmarks produced those numbers. Figure 3. Microsoft’s claims for Windows on RTX Spark devices versus the MacBook Pro M5 Pro. (Source: Microsoft) Local AI is the pitch Microsoft spent much of the event on what these machines are for.

Its answer is hybrid intelligence, in which software routes work between cloud models and models running on the laptop, based on task complexity, cost, and privacy. Microsoft CEO Satya Nadella made the case. “There will never be a moment where we will go and look and say, ‘Oh, this runs locally.

This runs in the cloud.’ You will expect this hybrid intelligence to be everywhere and pervasive,” Nadella said. The 128 GB of unified memory is what makes that pitch credible on a laptop, because it lets large, compressed models stay resident. How Windows makes local agents safe to run, and what Microsoft is building around them, deserves its own look, and we cover it separately.

The software question, and what we still don’t know The Arm transition is the real test. The promise of software applications running natively on Arm has only just started, and its pace depends on how aggressively Nvidia and Microsoft work with the ISVs. Huang’s count of more than 1,200 tested applications is a start, but testing is not the same as native optimization, and definitely not ISV certification, and Microsoft did not detail either.

The event answered a lot, but many more details will be unveiled in the next weeks as RTX Spark systems become available. We saw prices for the Surface Laptop Ultra and the Dev Box, but not for the partner laptops, and we do not know how the $2,599 starting-point price maps to memory and storage configurations.

Performance, TDP operation points for the silicon, and many other aspects of the platform will have to wait for independent testing, which JPR is already working on. Running Windows instead of Linux, as DGX Spark does, changes usage and workflows. It’s important to notice that this chip doesn’t have an NPU.

A GPU is far faster for AI workloads but less power efficient for sustained AI work, and again, only independent testing will show how that plays out on battery. Then there is the workstation question. Sriram Krishnan, former senior White House policy adviser on AI, who moderated the panel with Nadella and Huang, put it simply: “I want a workstation for Christmas.” RTX Spark is not a workstation today, and no vendor claimed it is.

We are pointing to the following needs, if Nvidia, Microsoft, and the OEMs ever want to go there. A workstation needs these things: DGX Station for Windows The event’s last hardware reveal went the other direction, up from the laptop. DGX Station for Windows, from the Dell Precision Pro and the HP ZGX later this year, marks the first time a DGX Station runs Windows.

Microsoft pitched it as data-center-class performance with enterprise-grade security and manageability, able to run models of more than a trillion parameters locally. Its slide promised last quarter’s frontier at your desk. In the HP demo, work scaled across 32 agents, which Microsoft called a token factory for development teams.

DGX Stations are the clearest sign yet that Microsoft and Nvidia see Windows as a serious platform for local AI, from the laptop to the desk, beyond the mini PCs that will be based on RTX Spark. Figure 4. DGX Station for Windows. (Source: Microsoft) What do we think?

Qualcomm has been working for years, attempting to break into the x86 compute market and finding how hard it is to build the technology, enabling the software ecosystem, and overcoming the brand recognition x86 has built over more than five decades. Nvidia arrives with more of the resources needed to face those challenges—and with brand recognition that is orders of magnitude ahead.

It also brings gaming capabilities, CUDA software support, and the funding to work with ISVs. That could make RTX Spark and future Nvidia client platforms not only a local powerhouse for agentic AI workflows, but also a truly general-purpose platform. That is why we see innovation that the premium and enthusiast mobile market has not seen in a very long time.

Is DGX Spark silicon in a laptop, running Windows, all this will be? Time will tell. We will watch the TDP operating point, which is we’ve already started testing at JPR Labs.

We are excited to see how sustained AI performance behaves, how quickly native Arm software arrives, and dream of the day when RTX Spark earns ISV certifications and a place in OEM workstation lines. Until the workstation criteria stated above are met, RTX Spark is a premium and enthusiast platform: great at AI-enhanced workflows and capable of gaming, as support and enabling expands in the future.

Stay tuned for more details about this innovative platform. — The Workstation Cavalry AND IF YOU LIKED WHAT YOU READ HERE, DON’T BE STINGY, SHARE IT WITH YOUR FRIENDS. Disclaimer: This article is not financial advice.

The content is for informational purposes only. The author may own shares in companies mentioned, and readers should not consider this article a recommendation to buy, sell, or hold any assets.