Acer introduced the SFF RTX Spark on September 2, 2026, at IFA 2026, presenting a compact desktop built for agentic AI workloads on the user’s own device. The system combines NVIDIA’s RTX Spark superchip with a space-saving design, putting substantial AI compute into a smaller footprint.
The main appeal is not just size. Acer’s desktop is designed to run personal AI agents locally, which keeps data on the device instead of sending every task to a cloud model. That pairing gives the SFF RTX Spark two clear goals: provide serious computing power and support on-device data privacy.
A Superchip Built for AI and Graphics
The RTX Spark Superchip supplies the desktop’s core computing platform and brings NVIDIA’s AI and graphics technologies together. It combines a Blackwell RTX GPU with 6,144 CUDA cores, fifth-generation Tensor Cores with FP4 precision, and a 20-core NVIDIA Grace CPU.
NVIDIA links those components through its NVLink-C2C chip-to-chip interconnect, creating a unified platform rather than treating the processor and graphics hardware as separate parts. MediaTek collaborated with NVIDIA on the custom Arm-based CPU design used in the system.
The platform reaches up to 1 petaflop of AI compute and supports up to 128 GB of high-speed unified memory. That capacity gives the compact desktop room to handle demanding AI workloads without relying on a separate cloud system for every request.
Jensen Huang, CEO of NVIDIA, described the broader shift in simple terms: “The PC is being reinvented.” He added, “For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask, and the PC does the work.”
What the Platform Can Handle
NVIDIA says the platform can render 90 GB 3D scenes with OptiX and DLSS, giving the system a role in demanding graphics work as well as AI. It can also edit 12K 4:2:2 video with the Blackwell decoder, a workload that requires substantial processing and memory capacity.
For large language models, the platform can run models with 120 billion parameters and support up to 1 million tokens of context. That combination is aimed at workloads that need both a large model and the ability to work with extensive amounts of information in one session.
The RTX Spark platform also supports gaming. NVIDIA says it can play AAA games at 1440p resolution above 100 frames per second with ray tracing, DLSS, and Reflex. Those figures show how the same hardware can serve AI, creative work, and graphics-intensive entertainment.
Adding Controls for Personal AI Agents
The local AI focus connects with new Windows tools developed by NVIDIA and Microsoft. The companies created Windows security primitives and the NVIDIA OpenShell runtime so personal AI agents can run on a user’s primary device under defined controls.
The Windows security primitives provide identity, containment, policy, and end-to-end security capabilities. Those controls matter because an agent can do more than answer a question: it can operate within permissions that determine what tasks it may perform and what information it may access.
OpenShell lets users define what agents can and cannot do. It can also route queries to local models based on privacy policies, giving users a way to keep selected requests on the SFF RTX Spark when privacy matters.
When a query does go to a cloud model, OpenShell can obscure personal information before sending it. That approach gives the user more control over the boundary between local and cloud AI, rather than forcing every request into the same path.
Agent developers including OpenClaw and Hermes Agent are adopting the security layer in new Windows apps. Their adoption places the SFF RTX Spark within a broader push to make personal AI agents useful while keeping their actions and data under user-defined rules.
Acer’s compact desktop brings together the hardware needed for those workloads: a Blackwell RTX GPU with up to 6,144 cores, a Grace CPU with up to 20 cores, up to 1 petaflop of AI compute, and up to 128 GB of unified memory. The result is a small system built around the idea that an AI assistant can work on the PC itself, with performance and privacy handled in the same box.
Based on
