🏠IT之家•Stalecollected in 1m
Nvidia Announces RTX Spark PC Processor with MediaTek CPU
💡Nvidia's new 1-petaFLOP PC chip could redefine local AI performance for developers and creative professionals.
⚡ 30-Second TL;DR
What Changed
Features a 20-core MediaTek Grace CPU paired with a Blackwell RTX GPU (6144 CUDA cores).
Why It Matters
This integration signals a major shift toward local AI-native hardware, potentially reducing reliance on cloud-based inference for creative workflows. It sets a new benchmark for high-end consumer and workstation AI computing.
What To Do Next
Evaluate your local AI model deployment strategy to leverage the Blackwell architecture for edge-based inference.
Who should care:Developers & AI Engineers
Key Points
- •Features a 20-core MediaTek Grace CPU paired with a Blackwell RTX GPU (6144 CUDA cores).
- •Delivers 1 petaFLOP of AI performance with 128GB of unified memory.
- •Supports major creative software like Adobe Premiere and Photoshop with up to 2x performance gains.
- •First devices from partners like Acer, ASUS, Dell, and Microsoft arriving this fall.
🧠 Deep Insight
Web-grounded analysis with 23 cited sources.
🔑 Enhanced Key Takeaways
- •The RTX Spark processor is also identified as the GB10 Grace Blackwell Superchip, particularly in the context of the NVIDIA DGX Spark, a personal AI supercomputer designed for developers and researchers.
- •The Blackwell GPU architecture, a core component of the RTX Spark, is built on a custom TSMC 4NP process, features 208 billion transistors, and employs a dual-die design interconnected by a 10 terabytes per second (TB/s) chip-to-chip interface.
- •Nvidia's collaboration with MediaTek on the Grace CPU design leverages MediaTek's extensive experience in creating power-efficient and high-performance CPU architectures, memory subsystems, and high-speed interfaces.
- •The RTX Spark, also referred to by its rumored codenames N1X/N1, is specifically engineered for Windows on Arm platforms, incorporating a 32 and 64-bit x86 Prism emulator and optimizations for local agentic AI use-cases, including Nvidia's Nemotron models.
- •The platform integrates Microsoft Power and Thermal Framework (MPTF) for efficient cooling and workload profile scheduling (WPS) to maximize hardware utilization, supporting up to 128GB of unified RAM with a 600GB/s memory bandwidth.
📊 Competitor Analysis▸ Show
| Feature / Competitor | Nvidia RTX Spark | Apple M4 | AMD Ryzen AI (e.g., Ryzen AI 9 HX 370) | Intel Core Ultra | Qualcomm Snapdragon X Elite |
|---|---|---|---|---|---|
| CPU Architecture | 20-core MediaTek Grace (Arm-based) | M4 (Arm-based) | Ryzen AI (Zen-based, with NPU) | Core Ultra (x86, with NPU) | Oryon (Arm-based) |
| GPU Architecture | Blackwell RTX (6144 CUDA cores) | Integrated (part of M4 SoC) | RDNA 3.5 graphics | Integrated (part of Core Ultra SoC) | Integrated (part of Snapdragon X SoC) |
| AI Performance | 1 petaFLOP (FP4) | Powerful Neural Engine | Up to 50 TOPS (for Ryzen AI 9 HX 370) | NPU for AI tasks | NPU for AI tasks |
| Memory | 128GB unified memory | Unified memory architecture | System RAM (e.g., 128GB RAM for high-end AI PC) | System RAM | System RAM |
| Power Profile | 45-80W | Not specified | Not specified | Not specified | Not specified |
| Pricing (Typical PC) | N/A (DGX Spark, a personal supercomputer using GB10/RTX Spark, starts around $3,999) | High-end segment; typically $1,300 to $2,500+ | High-end segment; typically $1,300 to $2,500+ | High-end segment; typically $1,300 to $2,500+ | High-end segment; typically $1,300 to $2,500+ |
🛠️ Technical Deep Dive
- Blackwell GPU Architecture: Features 208 billion transistors, manufactured using a custom TSMC 4NP process.
- Dual-Die Design: Blackwell GPUs utilize two reticle-limited dies connected by a 10 TB/s chip-to-chip interconnect, forming a unified single GPU.
- Tensor Cores: Incorporates fifth-generation Tensor Cores with native support for FP4 precision, enhancing AI compute capabilities.
- AI Management Processor (AMP): A dedicated RISC-V based scheduler chip on the GPU designed to offload scheduling from the CPU and improve GPU resource control, utilized through Windows Hardware-Accelerated GPU Scheduling (HAGS).
- NVLink-C2C Interconnect: Connects the Grace CPU and Blackwell RTX GPU, providing 1.8 terabytes per second (TB/s) of bidirectional bandwidth.
- Grace CPU: A 20-core Arm-based CPU, co-designed with MediaTek, focusing on power efficiency and performance.
- Unified Memory: 128GB of LPDDR5x unified memory, coherently shared between the CPU and GPU, offering 600GB/s memory bandwidth.
- Manufacturing Process: The GB10 SoC, which the RTX Spark is based on, is built on TSMC's 3nm process.
- Power Profile: Designed with a 45-80W power profile.
- Emulation Support: Includes a 32 and 64-bit x86 Prism emulator for compatibility with non-Arm native applications.
🔮 Future ImplicationsAI analysis grounded in cited sources
Nvidia's entry with RTX Spark will significantly accelerate the adoption of Windows on Arm PCs.
Nvidia's robust GPU and AI software ecosystem, combined with Microsoft's optimization efforts (WPS, MPTF, x86 emulation), addresses key weaknesses of previous Windows on Arm attempts, particularly in gaming and creative applications, making the platform more viable for a broader market.
The collaboration between Nvidia and MediaTek could lead to broader AI integration across various device categories beyond PCs.
MediaTek's expertise in power-efficient Arm-based SoCs and its existing collaborations with Nvidia in automotive and IoT suggest a strategic move to infuse AI capabilities from edge devices to cloud infrastructure.
The RTX Spark will intensify competition in the high-performance AI PC market, pushing rivals to innovate further in NPU performance and software integration.
With its 1 petaFLOP of AI performance and strong software support, the RTX Spark sets a high benchmark, compelling competitors like Apple, AMD, and Intel to enhance their AI-specific hardware and software ecosystems to remain competitive.
⏳ Timeline
2010-09
Windows ran on Nvidia Tegra Arm for the first time, showcasing Nvidia's early attempts in the Arm PC space.
2023-10
Reuters reported that Nvidia was developing Arm-based CPUs capable of running the Windows OS.
2024-03-18
Nvidia officially announced the Blackwell GPU architecture at GTC 2024.
2025-01-08
Nvidia and MediaTek confirmed their partnership on Arm chips for Windows PCs at CES 2025, unveiling Project DIGITS (DGX Spark) featuring the jointly developed GB10 superchip.
2025-09-12
NVIDIA revealed the GB10 SoC architecture at Hot Chips 2025, detailing its 3nm manufacturing process and dual-die design.
2026-06-01
Nvidia officially unveiled the RTX Spark PC Processor, combining a 20-core MediaTek Grace CPU with a Blackwell-based RTX GPU.
📎 Sources (23)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗