NVIDIA RTX Spark PCs Open for Pre-Orders With Arm CPUs, Blackwell RTX Graphics, and Serious AI Power
NVIDIA is stepping into a new era of personal computing with RTX Spark PCs, a new class of sleek laptops and compact desktops built to combine high-performance AI, modern PC gaming, and creator-focused tools in one platform. Pre-orders are now open, with the first systems expected to arrive on October 16.
Unlike NVIDIA’s traditional role in the PC market, where it has mainly supplied discrete graphics cards, RTX Spark marks the company’s move into a more complete client PC platform. At the center of these machines is the new N1X system-on-chip, combining an Arm-based Grace CPU, Blackwell RTX graphics, unified memory, and NVIDIA’s AI software stack.
The result is a Windows on Arm PC platform designed to challenge both traditional x86 laptops and existing Arm-based Windows devices, while offering the kind of graphics and AI acceleration NVIDIA is known for.
RTX Spark PCs are aimed at three major groups: developers, creators, and gamers. Developers can run NVIDIA AI tools, models, and workflows locally without having to rewrite their software. Creators get access to 5th-generation Tensor Cores, NVFP4 support, AV1 acceleration, 4:2:2 video encode and decode, ray tracing, and DLSS features across production workflows. Gamers, meanwhile, are promised smooth 1440p gameplay at over 100 frames per second in supported AAA titles using DLSS 5, Reflex, and G-SYNC.
Major PC brands are already preparing RTX Spark systems, including Acer, ASUS, Dell, Gigabyte, HP, Lenovo, MSI, and Microsoft. Pricing is expected to start at $2,599 for models using the 18-core CPU configuration, a 5,120-core Blackwell RTX GPU, and 24GB of LPDDR5x unified memory. One example of this entry configuration is expected to appear in Microsoft’s Surface Laptop Ultra EP2.
The N1X chip will launch in two main versions.
The higher-end model features a 20-core Grace CPU paired with a 6,144-core Blackwell RTX GPU. It supports unified memory configurations starting at 24GB and scaling up to 128GB, making it suitable for demanding AI workloads, advanced creative projects, and high-end gaming. This version will be available in both laptops and compact desktop PCs at launch.
The second version includes an 18-core Grace CPU and a 5,120-core Blackwell RTX GPU. It supports 24GB to 32GB of unified memory and will initially appear in laptops, with small desktop systems expected later.
On paper, the graphics core counts are impressive. The top N1X configuration matches the CUDA core count associated with the GeForce RTX 5070, while the 18-core version includes more CUDA cores than an RTX 5060 Ti. Real-world performance will depend heavily on power limits, thermals, and device design, but RTX Spark systems are positioned to deliver some of the strongest integrated graphics performance available in a consumer PC.
NVIDIA says RTX Spark PCs will be capable of 1440p gaming at over 100 FPS with the help of DLSS technologies. The platform is expected to support DLSS 5, DLSS 4.5 Ray Reconstruction, and multi-frame generation up to 6x in supported titles. Reflex and G-SYNC support are also included to reduce latency and improve smoothness during gameplay.
Game support is a key part of the platform’s success. NVIDIA has highlighted growing developer and publisher support for Windows on Arm gaming, with companies such as Electronic Arts, Embark, Tencent, Ubisoft, KRAFTON, NetEase, Riot Games, and Epic Games involved in expanding the game catalog. Current and upcoming titles mentioned for Arm-based Windows PCs include Fortnite, Alan Wake 2, World of Warcraft: Forever, League of Legends, Valorant, PUBG: Battlegrounds, and Genshin Impact. NVIDIA also says Call of Duty is planned for RTX Spark in 2027.
For creators, RTX Spark PCs are designed to handle demanding media and 3D workloads. NVIDIA claims the platform can render large 90GB 3D scenes using OptiX and hardware ray tracing. Hardware-accelerated 4:2:2 video encode and decode should also make these systems attractive for video editors, colorists, and content creators working with high-quality footage.
AI is one of the biggest selling points of RTX Spark. The platform brings the full NVIDIA AI stack to compact consumer systems, enabling local AI agents with up to 120 billion parameters and context lengths up to 1 million tokens, depending on memory configuration and workload. NVIDIA also points to Windows Agent Framework support and unified memory optimizations as key advantages for running AI tasks locally.
The company claims RTX Spark PCs can deliver major local AI performance gains compared with an Apple MacBook Pro 16-inch with M5 Pro. NVIDIA cites up to 2.1x faster time to first token, up to 4.3x faster AI image generation, and up to 6.2x faster AI video generation.
Despite the powerful hardware, RTX Spark laptops are expected to arrive in slim designs, with some measuring around 14mm thick. NVIDIA is also promoting all-day battery life, although actual battery performance will vary depending on workload, device size, display configuration, and manufacturer tuning.
Under the hood, the RTX Spark N1X platform shares ideas with NVIDIA’s GB10 Grace Blackwell Superchip design. The chip uses a multi-die approach, combining a CPU die and GPU die in one advanced package. It is built using TSMC’s 3nm process technology and relies on a high-speed, low-power chip-to-chip interface to connect the different components.
The CPU side is based on the Arm v9.2 architecture and includes up to 20 custom Grace cores designed in collaboration with MediaTek. The 20-core configuration is arranged in two clusters of 10 cores, with each core receiving private L2 cache and each cluster receiving 16MB of L3 cache, for a total of 32MB.
The GPU side is based on NVIDIA’s Blackwell architecture and includes 5th-generation Tensor Cores, RTX ray tracing cores, CUDA support, DLSS, Reflex, G-SYNC, ACE, and NVFP4 acceleration. NVIDIA lists performance at up to 31 TFLOPs of FP32 compute and up to 1,000 TOPS of NVFP4 AI performance. The GPU also includes 24MB of L2 cache.
Memory is another major part of the RTX Spark design. The platform supports 256-bit LPDDR5x unified memory at speeds up to 9,400 MT/s, delivering up to 301GB/s of raw bandwidth. With aggregate bandwidth over the system interface reaching up to 600GB/s, the CPU and GPU can access a shared memory pool more efficiently. Memory capacity can scale up to 128GB on the highest-end systems, which is especially important for large AI models and professional creative workloads.
NVIDIA’s broader GB10 platform also supports features such as ConnectX-7 networking, CUDA, TensorRT, vLLM, SLANG, NVFP4, NVLINK C2C, and unified system memory. These technologies are designed to make it easier for developers and professionals to move workloads between local RTX Spark systems, larger workstations, cloud environments, and accelerated data centers.
With RTX Spark, NVIDIA is not just launching another laptop chip. It is introducing a full PC platform built around AI acceleration, advanced graphics, and Arm-based efficiency. If the company’s performance claims translate well into real-world use, RTX Spark could become one of the most important new Windows PC platforms for gaming, content creation, and local AI development.NVIDIA RTX Spark and GB10 Superchip Could Redefine Windows AI PCs
NVIDIA’s RTX Spark platform is shaping up to be one of the most interesting developments in the next wave of AI-powered Windows PCs. Built around the GB10 Superchip, the platform combines Grace CPU cores, Blackwell GPU architecture, unified memory, advanced connectivity, and NVIDIA’s mature RTX software ecosystem into a compact system designed for creators, developers, AI workloads, and high-performance mobile computing.
One of the most important features of the GB10 Superchip is how it handles unified memory. Instead of forcing users to enter the UEFI BIOS to change memory allocation, NVIDIA will allow users to manually assign memory to the GPU directly through Windows. On a system with 128 GB of unified memory, as much as 111 GB can be dedicated to the GPU. That is a major advantage for AI models, generative AI applications, 3D rendering, large creative projects, and other GPU-heavy workloads that benefit from massive memory capacity.
The GB10 Superchip also includes 16 MB of System Level Cache, functioning as L4 cache for the CPU. This cache helps the different engines inside the system-on-chip share data more efficiently while keeping power consumption under control. NVIDIA is also using a high-bandwidth, low-power chip-to-chip interface based on its NVLink architecture, helping the CPU and GPU communicate more effectively.
Connectivity is another strong point of the GB10 design. The chip supports PCIe, USB, Ethernet over PCIe, and up to four simultaneous displays. That includes three DisplayPort outputs plus one HDMI output, with support for up to 4K at 120Hz through DisplayPort Alt Mode and up to 8K at 120Hz through HDMI 2.1a. For security, the chip includes Dual Secure Root support, an SROOT processor, an OSROOT processor, and support for both firmware TPM and discrete TPM. Despite its capabilities, the entire chip is rated at a 140W TDP.
Scalability is where the GB10 Superchip becomes even more compelling. Multiple GB10 systems can be connected using NVIDIA ConnectX technology, allowing users to scale throughput, bandwidth, and total memory capacity for larger AI models. The ConnectX NIC links to the GB10 SoC through a PCIe Gen5 x8 interface, while systems communicate with each other over Ethernet. This opens the door for compact AI clusters built from RTX Spark machines, giving developers and researchers a flexible alternative to much larger workstation setups.
Software could be NVIDIA’s biggest advantage
Hardware is only part of the story. NVIDIA’s biggest strength may be its software ecosystem, which is already deeply established across desktop PCs, laptops, workstations, and AI systems.
Unlike newer attempts to push alternative Windows PC platforms, NVIDIA already has a proven foundation with RTX. Its GPUs are widely used in gaming laptops, creator notebooks, desktop PCs, and professional workstations. Driver support is mature, developer tools are widely adopted, and the company’s software stack is already familiar to millions of users.
CUDA, TensorRT, RTX acceleration, and AI-focused development tools give NVIDIA a major advantage on Windows. Linux support has also improved significantly, especially around DGX Spark systems. That matters because many AI developers move between Windows and Linux depending on the project, and NVIDIA’s ecosystem gives them a more consistent environment across both platforms.
Still, the largest unknown is not the GPU. It is the CPU.
Grace has already appeared in data center and AI-focused environments, but as a Windows PC processor, it remains largely unproven. Most benchmarks seen so far have come from DGX-based systems, Linux testing, or emulated environments rather than real-world consumer Windows laptops. NVIDIA will need to prove that Grace can deliver smooth everyday PC performance, strong application compatibility, and efficient battery life in mobile systems.
Early demonstrations, however, appear promising. Applications and workloads such as Blender, Unreal Engine, LLM Studio, OpenClaw, and other creative and AI tools have reportedly been shown running at solid speeds. NVIDIA’s internal testing points to up to 10x faster AI performance and up to 2x faster generative AI performance in certain workloads. Gaming performance has also been demonstrated in demanding titles such as Alan Wake II, Fortnite, and Indiana Jones, suggesting that RTX Spark systems may be far more versatile than typical AI-focused machines.
NVIDIA is planning more than one generation
A major sign of confidence is NVIDIA’s long-term roadmap. RTX Spark is not being treated as a one-time experiment. Instead, NVIDIA is laying out a multi-generation plan that extends through the end of the decade.
The first generation of RTX Spark systems will use the Grace CPU architecture paired with Blackwell GPU architecture and LPDDR5X memory. These systems are expected to arrive in 2026.
The second generation is planned for 2028. It will move to Vera CPU architecture and Rubin GPU architecture, along with LPDDR6 memory support. That should bring higher bandwidth, better efficiency, and stronger AI performance.
The third generation is planned for 2030, using Rosa CPU architecture and Feynman GPU architecture. While memory details for that generation have not been fully revealed, the roadmap shows that NVIDIA intends to keep RTX Spark evolving as a serious PC platform.
Each RTX Spark generation is expected to include at least two performance tiers. The higher-end first-generation Grace Blackwell design is rated at around 1 PFLOP of AI performance, while a lower-end version is expected to deliver around 400 TFLOPs. That split should allow NVIDIA and its partners to target different price points, from ultra-premium AI laptops to more compact systems for developers and creators.
The first RTX Spark laptops are coming
NVIDIA has already revealed several RTX Spark laptop designs from major PC brands. The first wave includes:
MSI Prestige N16 FLIP AI
Dell XPS 16
Lenovo Yoga Pro 9n
ASUS ProArt P14
HP OmniBook Ultra 16
Microsoft Surface Laptop Ultra
These systems are expected to target users who need far more AI and GPU performance than a typical premium laptop can provide. That includes AI developers, 3D artists, video editors, game developers, engineers, and professionals working with large local models.
One important question is memory configuration. NVIDIA has mentioned support for up to 128 GB, and current DGX Spark and OEM versions have been shipping with 128 GB configurations. If RTX Spark laptops are limited to that amount at launch, prices could land firmly in the premium category, especially with ongoing DRAM supply pressure. More flexible RAM and SSD options would make the platform more accessible to a wider range of buyers.
RTX Spark will also come to Mini PCs
NVIDIA’s RTX Spark strategy is not limited to laptops. The platform is also expected to appear in Mini PCs, giving desktop users a compact but powerful option for AI development, creative production, and high-performance computing.
RTX Spark Mini PCs are expected in fall 2026, with systems planned from Acer, ASUS, Dell, Gigabyte, HP, MSI, and Lenovo. These machines could fill an important gap between small-form-factor PCs and full DGX workstations. For users who want local AI performance without a massive desktop tower or enterprise-class workstation, RTX Spark Mini PCs could become a very attractive option.
As NVIDIA continues building out the DGX and RTX Spark ecosystem, it is possible that larger desktop systems will follow as well. That would give the company a complete lineup ranging from laptops and Mini PCs to professional workstations and scalable AI systems.
RTX Spark could become a major shift for AI PCs
The NVIDIA GB10 Superchip and RTX Spark platform represent more than another premium laptop launch. They bring together unified memory, Blackwell GPU technology, Grace CPU architecture, scalable AI performance, and a mature software ecosystem in a way that could challenge traditional PC designs.
The platform still has questions to answer, especially around Grace CPU performance in Windows, pricing, battery life, and software compatibility. But NVIDIA has one key advantage: it already owns much of the GPU, AI, creator, and developer ecosystem.
If RTX Spark delivers on its promise, it could become one of the most important AI PC platforms of the next several years.
