NVIDIA RTX Spark Explained

Nvidia unveiled RTX Spark at Computex 2026, its a new Arm-based system-on-a-chip made up of its Grace CPU, a Blackwell RTX GPU, and up to 128GB of LPDDR5X unified memory. Designed for Windows laptops and desktop PCs, it’s intended for local artificial intelligence, gaming, and creative uses. Nvidia collaborated with MediaTek in the design of the chip itself, as well as with Microsoft to get everything working in Windows.

RTX Spark is a 1-petaflop Superchip that offers full CUDA and RTX ecosystem support as well as Windows-native agents. Finishing off the overview of this chip family is the statement that it “reinvents Windows PCs for the era of personal AI agents.” There are a lot more statements on who it’s aimed at and why it’s such an important chip over on the official press release, but below we’ll go a bit more into what it means for most users.

There are plenty of reasons why RTX Spark is interesting, but the fact it is Arm-based and not an x86 CPU could possibly be the most important. This means it has to use the Windows-on-Arm operating system, using Microsoft’s Prism Emulator to translate x86 code for Windows-on-Arm. Huge gaming back catalogues still very much favor x86, so the emulation is vital to the success of RTX Spark.

Windows has been steadily improving Windows-on-Arm for years, and it’s now in a healthy state, although not everything works perfectly. Having such a big player as Nvidia on the Arm side will help here, and given Nvidia’s links to the gaming industry, it could mean that Arm support catches up to x86 Windows. This won’t just affect RTX Spark, but also Qualcomm’s Snapdragon processors.

Nvidia already sounds pretty confident that it has worked out a lot of the kinks on this front, and is making positive noises about getting some of the more important, low-level anti-cheat applications up and running on Windows-on-Arm. That may not seem like much, but it’s one of the things that has held back gaming on Linux. Getting these anti-cheat apps up and running could change our PCs significantly.

That’s the CPU side covered, but what about the GPU? Nvidia promises 100fps at 1440p in the latest games and claims it can achieve RTX 5070 levels of performance, although that is dependent on the application, which probably comes down to that emulation layer again. The raw performance is clearly there, if there’s one thing you can be certain of, it’s that Nvidia knows how to piece together a GPU.

RTX Spark will be available with up to 6,144 RTX Blackwell CUDA cores, matching the RTX 5070. This will be the specification for the top end chip, with rumors suggesting that more modest versions could go down to as little as 2,048 CUDA cores. This wouldn’t exactly be a gaming powerhouse but given the right power envelope could make for some thin and light gaming on the go. It’s going to be interesting to see what price points manufacturers can hit using these chips.

Gaming is an important part of the RTX Spark proposition, but it isn’t just about gaming. There are a couple of things that make this obvious, from the fact that it supports up to 128GB of unified memory to that 1-petaflop of performance figure being AI-specific. If you’ve been following Nvidia’s fortunes at all, this shouldn’t come as much of a surprise, as it has shifted from being primarily a gaming company to one whose future is pretty much defined by AI. It clearly believes that AI agents are going to be pivotal at least in the short term, and RTX Spark is all in on that.

Whether RTX Spark succeeds depends on how well Windows-on-Arm continues to mature and whether developers fully adopt Arm-native software. The hardware itself looks compelling, combining Nvidia’s GPU expertise with an efficient Arm-based platform and enough memory to support demanding AI workloads.

If Nvidia can deliver on its performance promises while maintaining compatibility with existing Windows applications and games, RTX Spark could become one of the most significant changes to the Windows PC ecosystem in years. It is still early days, but RTX Spark represents Nvidia's clearest attempt yet to bring AI and gaming ecosystems together in one platform.

The first RTX Spark laptops are expected in the fall of 2026, from global OEMs with compact desktop systems not far behind.