Nvidia RTX Spark Can Light Windows Fire on Arm

Tie: Nvidia is “reinventing the personal computer,” according to CEO Jensen Huang. Microsoft and Nvidia have been teaming up in preparation for Nvidia’s highly anticipated RTX Spark launch. It’s a new Arm-based system-on-chip (or “SoC”) platform that brings Nvidia’s Blackwell architecture to thin and light Windows laptops. The goal is to provide high-power processing performance to use personal agents, creative work and play, but without the space, power requirements and cooling requirements that are often imposed by different graphics.

The RTX Spark joins Qualcomm’s Snapdragon X processors running Windows on Arm, with similar claims of “all-day battery life.” Snapdragons achieve that, but one thing to remember about Nvidia’s chip is that it is designed for much heavier loads than Snapdragon processors.

Those aren’t meant to “render massive 90GB-plus 3D scenes, edit 12K 4:2:2 video, produce 4K AI videos, use 120B parameter LLM with a core of up to 1 million tokens using local agents, and play AAA games at 1440p with over 100 hours of battery life.” It remains to be seen if the Spark can live up to that under normal use.

This is the first of what Nvidia says it plans to be a line of chips at various price points. These first models are scheduled to ship this fall:

• Microsoft Surface Laptop Ultra
• Dell XPS 16
• Asus ProArt P14 and P16
• HP Omnibook X 14, Omnibook Ultra 16
• Lenovo Yoga Pro 9n
• MSI Prestige N16 Flip AI

The 15-inch Surface Laptop Ultra is particularly notable because Microsoft hasn’t updated its screens in a very long time, and Surfaces (both desktop and laptop) have never included the discrete GPUs their prices seem to require. The Ultra has a high-resolution (262ppi) 15-inch mini LED touchscreen that supports HDR (with a maximum brightness of 2,000 nits), unlike the old, meh model. Microsoft hasn’t updated its Surface Laptop Studio in three years, and this is the chip and screen we need if Microsoft plans to bring it back from the dead.

There will be small desktops. There seems to be a resurgence of this — at least an increase in the number of manufacturers offering it — thanks to developers. RTX Spark models will compete with AMD Ryzen AI Halo based models for example. They are expected from companies such as Acer, Asus, Dell, HP and Lenovo, among others.

Nvidia plans to have desktop, laptop and workstation chips for each generation of chips.

Given the current price volatility, we won’t know how much they will cost until they are closer to shipping. AI’s strong demand for components — and the resources needed to make them — has created a severe shortage of memory, processors and SSD storage, driving up computer and phone prices and affecting available configuration options.

It exploded

The chip is an offshoot of the DGX Spark (GB10), which powers Linux-based computers aimed specifically at developers and now the Windows-based DGX Channel. The Spark was designed in collaboration with MediaTek, and has the same specs as the DGX: 6,144 CUDA cores, 20-core Grace CPU, the ability to access 128GB RAM and more. Nvidia says it supports up to 120B virtual agents with 1M cores. (For reference, AMD says its top Ryzen AI Max Pro 400 series chip can handle 300B parameter models).

Its GPU specs are roughly comparable to the RTX 5070, but the integrated memory architecture means it has access to more RAM than 12GB. Nvidia says the system configuration can be scaled down to 16GB, though, which means it could be a hindrance where a dedicated 5070, with 12GB VRAM, might not be possible. The company gave 100fps 1440p as its gaming performance benchmark (though it wasn’t clear if that was with or without DLSS 4.5 enabled).

Nvidia says the chip’s overall AI performance is one PFLOPS (a billion floating-point operations per second), but that’s based on FP4 calculations. On the other hand, FP4 is the current favorite of data formats because it is faster than other floating point options and more accurate than integer, but there are some trade-offs. (Procyon has a good visual example of what speed versus accuracy tradeoffs can mean in image reproduction.) But among consumer SoCs, this is the first to support it in hardware.

The real competition for this is the M5 Pro and M5 Max MacBook Pros, which are aimed at the same users, but the M5 line does not support FP4 and FP8 data types, which may turn out to be an obstacle.

The component itself can run anywhere from “single digits” to 80W, which means you’ll really need to pay attention to whether the laptop is running at full power or if the manufacturer is overclocking it. In other words, it sounds like the performance, especially on the battery, could be very different. Generally, mobile processor power envelopes are thin bands; for example, Intel Core X9 388H specifies 15W-85W.

It has an NPU, which Nvidia doesn’t seem to like to talk too much about, but programs with Spark are considered Copilot Plus-qualifying, so it should be able to hit at least 40 TOPS.

RTX Spark may seem powerful, but Nvidia maintains its strict separation between the pro and consumer markets. For example, it does not plan to implement a system to verify applications or support ECC memory.

In addition to being one of Nvidia’s launch partners for the Surface Laptop Ultra, Microsoft has been working on making the necessary updates to Windows to use the new chip.

Like Qualcomm’s Snapdragon X series processors, Windows doesn’t natively support the Arm command set the way it does for Intel and AMD’s x86-architecture chips, which were the basis for the PC. Instead, Arm-based systems use an emulation layer called Prism to interpret instructions. Emulation is partly the reason why early systems based on Qualcomm chips encountered performance and compatibility issues.

Windows Repair

Many of the updates to Windows that are needed to support the hardware are under the hood, but one will be right in your face: Microsoft is putting Spark-run agents on the Taskbar.

Many of the changes we’ve seen in Windows recently are laying the groundwork for this. Prism was written for Qualcomm’s SoCs, because it was the only Arm-based silicon the operating system needed to run. Supporting RTX Spark meant updating Prism and other core Windows components to effectively distribute workloads across CPU cores, balance cooling and performance, address and intelligently manage the large amount of integrated memory available on the GPU (for AI processing with TensorRT) and more.

Qualcomm doesn’t have nearly as much money in Windows gaming performance as Nvidia does, for obvious reasons. For example, Nvidia has been working with Microsoft to improve compatibility with anti-cheat software (such as Epic’s Easy Anti-Cheat), which prevented certain games from running on the devices, and support for the Xbox app, which is key to Microsoft’s game strategy in general.

Adobe is also improving parts of its capture engines to fit directly into Spark, especially with several new pipelines to accelerate many GPU- and AI-intensive features such as rendering complex timelines in Premiere Pro and improving natural brushes in Photoshop. While CUDA and TensorRT are already running on Nvidia’s mobile GPUs, using them properly in this unique architecture requires some updating. Apps will also be able to interact with Windows agents.

Also, Nvidia is bringing OpenShell — its security protocol for active agents — to Windows, using new controls that Microsoft will reveal at its Build conference in the first week of June. OpenShell, in theory, allows you to define guardrails for your agents, route queries to authorized local models based on your privacy policies and let it “hide” personal information when querying cloud-based models.

Nvidia is trying to expand daily operations beyond developers, with the idea that “wider adoption is limited by the inability to run agents safely and privately on users’ primary PCs.” I suspect the trust issues are more complicated than that. The company says that OpenShell will be included in the current popular agents, OpenClaw and Hermes.