We usually open AI in a browser, type a question and wait for an answer. RTX Spark opens up another possibility: a capable local model on the computer we use for our projects, entertainment and work. NVIDIA also wants that device to appeal to people editing video, creating 3D content or playing demanding games.

That makes this news relevant to a much wider audience than AI developers. If you are curious about the next wave of Windows computers, it is worth looking at the announcements, early impressions and questions that still need answers.

What is RTX Spark, and which configurations exist?

RTX Spark is a platform for slim Windows 11 laptops and compact desktops. The current NVIDIA specification table lists these N1X variants:

  • Higher specification laptop: 20 Grace CPU cores, 6,144 Blackwell RTX GPU cores and up to 128 GB LPDDR5X memory.
  • Lower specification laptop: 18 CPU cores, 5,120 GPU cores and up to 64 GB memory.
  • Compact desktop: 20 CPU cores, 6,144 GPU cores and up to 128 GB memory.

These are platform maximums. Memory is unified: the CPU and GPU share the same pool, so the stated capacity is not additional dedicated video memory. Check the specific model’s memory.

DGX Spark is a separate product with a GB10 chip and DGX OS, focused on AI development. Its specifications and clustering capabilities should not be attributed to RTX Spark PCs.

An AI assistant in your laptop: what could you do?

Imagine giving an assistant a few notes and documents from your project, then asking it to find information, compare versions or prepare a draft. Or a tool that helps you explore code and experiment with your own model. These are examples of tasks that make local AI interesting.

The benefit is the possibility of keeping some processing on your device. When a suitable model and application work entirely locally, the task can run without sending content to a remote AI service. Check what each application actually does locally and which features need the internet.

In its original announcement, NVIDIA describes models with up to 120 billion parameters and context of up to one million tokens. That reflects the platform’s ambition; model choices depend on memory, format and software. A large model is not automatically right for every task.

Concept illustration of an AI assistant processing selected local documents
Original AI illustration of local work with selected documents. A concept image, rather than a photograph of a specific RTX Spark device.

Hardware is only part of the story. The assistant needs to understand the task, use appropriate tools and return a result you can check. A small project with a few familiar files is a good starting point: you will quickly see where AI helps.

What about games? RTX Spark also wants a gaming role

In its October announcement, NVIDIA describes AAA games at 1440p and more than 100 FPS with DLSS. This is a manufacturer claim. Actual results depend on the game, settings and configuration. Assessing it requires knowing the game, graphics settings and how the displayed frame rate was achieved.

For a player, the important question is how their favourite titles run. A slim chassis is attractive when carrying a laptop, while long sessions also depend on cooling, noise, the display and keyboard. A new chip creates interest; the complete laptop is what you buy.

Tom’s Guide reports positive impressions from early gaming demonstrations, explaining that the lack of an FPS counter prevented confirmation of the promised 100+ FPS. A first encounter tells us something about the experience; detailed tests should establish performance and limits.

Video, 3D and generative AI: the appeal for creators

People working with video, photography or 3D projects have another reason to follow RTX Spark: how much demanding work can fit into a computer that is easy to carry?

In its initial announcement, NVIDIA describes 3D scenes exceeding 90 GB, 12K 4:2:2 video processing and up to twice the performance in selected Adobe AI and graphics workloads. These are announced capabilities under suitable conditions; results depend on the application, project and configuration.

The practical question is easier than the specifications: does your project open smoothly, how long does export take, and what happens when several tasks run together? Some people will care most about editing, others about a 3D scene or experimenting with local image and video models. One overall result can hardly describe all those needs.

A petaflop sounds huge. What does it measure?

NVIDIA’s original announcement describes up to one petaflop of AI performance at FP4 precision. FP4 is a low precision format used in certain AI calculations. That makes the figure a specific measure, rather than a universal score for everything you do on a computer.

For a local model, look at response speed and memory use. For video, look at editing and export time. For a game, look at responsiveness and stable output. Software, the specific device and workload matter in all these situations, so a presentation figure belongs in a wider picture.

What do early reports and demonstrations show?

AP highlights the wider PC market shift and new competition for NVIDIA. Windows Central follows configurations and launch timing. Its September overview predates the current official table used for specifications here.

TechRadar shares promising IFA impressions, with limited scope for independent testing. In its later article on battery-powered operation, Tom’s Guide separates a performance demonstration from actual battery life, which still needs measurement.

Those reports offer a useful first look, while buying decisions need tests of the model you want. Pay particular attention to battery life under your workload: watching video, gaming and running an AI model are different tasks.

Your files, your assistant, your permissions

An AI agent can use files and tools across several steps. That can help with preparing material or working on a project, but you need to know what it can access and which actions it can perform.

Microsoft’s October announcement describes Windows controls for agent identity, isolation and oversight. A practical start is a selected folder, a saved copy of important files and a review before sending, publishing or deleting.

Also check whether the application stores history or calls a cloud service. A local model makes local processing possible; the behaviour of the whole tool depends on its settings. If you use such tools at work too, clear rules for AI and business data are useful.

Windows on Arm: a detail worth checking

RTX Spark uses Arm architecture. NVIDIA provides a guide to preparing applications and CUDA environments, while Microsoft’s FAQ explains emulation of many x86/x64 applications in Windows 11. Kernel-mode and print drivers require appropriate Arm64 versions.

Before choosing a device, check the programs, games, plugins and peripherals you already use. This matters especially for a particular plugin, older printer, audio device or specialised application. If you also use the computer for work, check your VPN, card reader or electronic signature. Compatibility is much easier to check before a purchase.

When does it arrive, and what about price?

According to NVIDIA’s October 7 announcement, laptop preorders are open and availability is scheduled from October 16, 2026. Compact desktops are expected in November 2026. Timing depends on the manufacturer and market.

As of this update on October 14, laptop shipments are still scheduled to begin in the coming days. We have not confirmed availability or prices in Serbia. The specific model’s price, warranty and local purchasing terms matter when buying.

Who is likely to find RTX Spark most interesting?

Creators with demanding projects, people experimenting with local AI models and users drawn to work and games on the same device have the strongest reasons to follow it. Even if you do not plan to buy one, this platform shows the direction NVIDIA and Microsoft want for the Windows PC.

If you mainly use a computer for a browser, films, email and documents, the decision will depend more on price, comfort and battery life. Local AI becomes attractive when you find a task that gives you a reason to use it regularly.

RTX Spark is worth following through specific devices and their tests. The most interesting question is still ahead: how well will the promised mix of AI, creativity and gaming fit into everyday use?

If this story gets you thinking about computers or IT support at work, explore NBG TEAM computers and workstations and our business IT support.

Sources and further reading

Official specifications and announcements are linked separately from editorial reports and demonstrations. Checked on October 14, 2026.