NVIDIA RTX Spark: The Arm-Powered Superchip Aimed at Gaming and AI PCs
At Computex 2026, NVIDIA unveiled RTX Spark, a Blackwell-plus-Grace superchip with 128GB unified memory headed to 40-plus laptops and desktops this fall.

At Computex 2026, NVIDIA CEO Jensen Huang took the wraps off RTX Spark, a single-package superchip that pairs a Blackwell RTX GPU with an Arm-based Grace CPU. Announced jointly with Microsoft on May 31, it is NVIDIA's biggest swing yet at reshaping what a Windows gaming and AI PC looks like.
Quick answer
RTX Spark is a single-package superchip pairing a Blackwell RTX GPU (6,144 CUDA cores) with a 20-core Arm Grace CPU and up to 128GB of unified LPDDR5X memory, rated at roughly 1 petaflop of FP4 AI performance. More than 30 laptops and 10-plus desktops ship this fall from Asus, Dell, HP, Lenovo, Microsoft, and MSI. The big catch for gamers is that it runs Windows on Arm, so x86 games depend on emulation or native Arm builds, making compatibility, not raw power, the deciding question.
Key takeaways
- RTX Spark fuses a 20-core Arm Grace CPU and a Blackwell RTX GPU (6,144 CUDA cores) on one package with up to 128GB of unified LPDDR5X memory.
- NVIDIA rates it at up to 1 petaflop of FP4 AI performance, with up to 300 GB/s of memory bandwidth.
- It is an Arm-based Windows-on-Arm platform, so x86 game compatibility depends on emulation or native Arm builds, the make-or-break question for gamers.
- More than 30 laptops and 10-plus desktops ship this fall from Asus, Dell, HP, Lenovo, Microsoft, and MSI, with Acer and Gigabyte to follow.
- Huang outlined a three-generation roadmap: Grace Blackwell, then Vera Rubin (LPDDR6), then Rosa Feynman.
What RTX Spark Actually Is
RTX Spark is a superchip, meaning the GPU and CPU live on one tightly coupled package rather than as separate parts on a motherboard. NVIDIA connects a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores to a 20-core Grace CPU over the NVLink-C2C interconnect. The CPU is co-developed with MediaTek and combines 10 high-performance Cortex-X925 cores with 10 efficiency-focused Cortex-A725 cores. The two halves share a single pool of up to 128GB of unified LPDDR5X memory with up to 300 GB/s of bandwidth, and NVIDIA rates the package at roughly one petaflop of FP4 AI performance.
The headline detail for the PC world is the CPU. Grace is an Arm design, not x86, so RTX Spark machines are Arm-based Windows PCs. That is the same direction Microsoft, Arm, and NVIDIA have been teasing as a "new era of PC," and Spark is the product that makes it concrete.
Note
Unified memory means the CPU and GPU draw from the same 128GB pool instead of copying data between separate system RAM and VRAM. That is a big deal for large AI models and high-resolution assets that previously had to fit inside a smaller VRAM budget.
Here are the headline specifications at a glance:
| Component | Specification |
|---|---|
| GPU | Blackwell RTX, 6,144 CUDA cores, 5th-gen Tensor Cores |
| CPU | 20-core Arm Grace (10x Cortex-X925 + 10x Cortex-A725), co-developed with MediaTek |
| Interconnect | NVLink-C2C between CPU and GPU |
| Memory | Up to 128GB unified LPDDR5X |
| Memory bandwidth | Up to 300 GB/s |
| AI performance | Up to ~1 petaflop FP4 |
| OS platform | Windows on Arm |
| Availability | 40+ devices, fall 2026 |

Why Gamers Should Care
NVIDIA is positioning Spark as a single platform for AI, content creation, and gaming rather than a pure AI accelerator. It bundles the technologies gamers already rely on, including DLSS, Reflex, and G-SYNC, alongside the newer DLSS 4.5 frame generation shown at the same event. Spark also folds in NVIDIA's broader stack, from CUDA and RTX to FP4, TensorRT, and OptiX.
The pitch is a thin laptop or a small, efficient desktop that can run demanding games and heavy local AI workloads on the same chip. The 128GB unified memory pool is the standout for play that overlaps with AI, since large language models and generation tools that normally choke on limited VRAM get far more room to work.
The open question is software compatibility. Because Spark is Arm-based, games and tools built for x86 Windows will depend on emulation or native Arm builds. That has historically been the friction point for Arm PCs, and how smoothly today's game libraries run will decide whether Spark is a genuine gaming platform or mainly an AI workstation that happens to play games. Even on capable hardware, PC game stutter and shader compilation hitches can be worse under emulation, so early reviews of frame pacing will matter as much as raw benchmark numbers.
Availability and Lineup
NVIDIA says more than 30 laptops and over 10 desktops built on RTX Spark will launch this fall from major partners including Dell, HP, Asus, Lenovo, Microsoft, and MSI, with Acer and Gigabyte models to follow. That is an unusually broad day-one lineup, which signals NVIDIA wants Spark treated as a mainstream PC platform rather than a niche launch.
Fixed
A 40-plus-device launch across multiple major OEMs this fall means real buying choices at multiple price points, not a single halo product.
The Roadmap Beyond Spark
NVIDIA did not stop at the first chip. Huang outlined a multi-generation roadmap for the Spark line. The first generation is Grace Blackwell. A future Vera Rubin generation pairs upgraded chips with faster LPDDR6 memory, and a later Rosa Feynman generation is expected to push memory bandwidth further still. Laying out three generations at launch is a clear statement that NVIDIA sees the superchip approach as a long-term direction for personal computing, not a one-off experiment.
What to Watch Next
A few things will determine whether RTX Spark lands with gamers:
- Game compatibility on Arm. Native Arm builds and the quality of x86 emulation will make or break the gaming experience.
- Real pricing. A broad fall lineup is promising, but mainstream appeal depends on where the laptops and desktops actually land on price.
- Thermals and battery. The promise is high performance in thin and small chassis, so independent testing of sustained performance and battery life matters.
- Driver and game support cadence. Spark inherits NVIDIA's gaming stack, but how quickly studios certify titles on the new platform will shape the early experience.
Frequently asked questions
Is RTX Spark a gaming PC or an AI machine?
NVIDIA pitches it as both. The Blackwell GPU and full RTX feature set make it capable of demanding games, while the 128GB unified memory and 1-petaflop FP4 rating target local AI workloads. Whether it is a great gaming platform hinges on Arm game compatibility.
Will my Steam library run on RTX Spark?
It depends. RTX Spark runs Windows on Arm, so x86 games rely on emulation unless a native Arm build exists. Many titles work through emulation, but performance and compatibility vary, this is the biggest unknown until independent testing arrives.
When can I buy an RTX Spark laptop?
Fall 2026, from Asus, Dell, HP, Lenovo, Microsoft, and MSI, with Acer and Gigabyte following. Expect more than 40 models across laptops and compact desktops.
How is RTX Spark different from a normal gaming laptop?
A normal gaming laptop has a separate x86 CPU and a GPU with its own VRAM. RTX Spark fuses an Arm CPU and Blackwell GPU on one package sharing a single 128GB memory pool, which helps large AI models but introduces the Arm software-compatibility question that traditional x86 laptops avoid.
For now, RTX Spark is the clearest sign yet that the next wave of Windows PCs will blend gaming and on-device AI on a single Arm-based superchip. The hardware story is strong. The fall launch and real-world testing will tell us whether the gaming story keeps up.
Sources & further reading
- nvidia.com/en-us/geforce/news/computex-2026-nvidia-geforce-rtx-announcements/
- tomshardware.com/laptops/nvidia-unveils-rtx-spark-superchip-at-computex-2026-new-platform-promises-to-turn-windows-into-an-agentic-ai-os-with-arm-cpu-blackwell-gpu-and-128gb-unified-memory
- tomshardware.com/pc-components/cpus/nvidia-unveils-dgx-sparrk-roadmap-for-laptops-and-desktop-pcs-at-computex-2026-three-generations-outlined-rubin-followed-by-rosa-feynman
- nvidianews.nvidia.com/news/nvidia-microsoft-windows-pcs-agents-rtx-spark
- en.wikipedia.org/wiki/Nvidia_RTX_Spark


