NVIDIA Grace CPU Superchip (GB10 Developer System)

The NVIDIA Grace Blackwell GB10 Superchip Developer System is a compact, self-contained AI computing platform built around the GB10 Superchip, which tightly integrates 72 NVIDIA Blackwell GPU cores with a 72-core NVIDIA Grace CPU via a high-bandwidth NVLink-C2C interconnect delivering 900 GB/s of chip-to-chip bandwidth. This architecture eliminates traditional PCIe bottlenecks between CPU and GPU, enabling unified memory access across 128 GB of LPDDR5X system memory shared coherently between compute domains. The system delivers up to 1 PFLOP of AI performance at FP4 precision, making it a formidable platform for on-device large language model inference, fine-tuning of smaller models, and AI pipeline prototyping without requiring data center infrastructure.

Designed for enterprise AI-at-the-edge deployments and developer productivity, the GB10 Developer System runs the full NVIDIA AI software stack including CUDA, cuDNN, TensorRT, and the NVIDIA AI Enterprise platform. Organizations can deploy production-grade inference workloads—running models up to 70 billion parameters locally—directly at the edge without reliance on cloud round-trips, which is critical for latency-sensitive and data-sovereignty-constrained applications. The system's compact desktop chassis consumes a fraction of the power of a full server rack while providing server-class AI throughput.

Launched in Q2 2025, the Grace Blackwell GB10 Developer System is positioned as the definitive tool for AI developers, research institutions, and enterprises building and validating AI applications before scaling to data center deployments. Its role as a sovereign, low-power, high-throughput inference node makes it uniquely suited for regulated industries, manufacturing floors, retail environments, and healthcare facilities where cloud connectivity is restricted or undesirable. The system ships with NVIDIA's comprehensive developer tools and supports popular AI frameworks including PyTorch, JAX, and ONNX Runtime natively.

Ideal for

• On-premises large language model inference for healthcare organizations requiring strict patient data locality and regulatory compliance
• Real-time AI-powered quality inspection and anomaly detection on manufacturing and industrial production lines
• Edge AI application development and validation for teams prototyping models intended for eventual data center or cloud deployment
• Retail and point-of-sale intelligent analytics running computer vision and NLP models entirely on-device without cloud dependency
• Government and defense edge computing nodes requiring sovereign AI processing with no external data transmission
• University and research laboratory AI experimentation platform providing 1 PFLOP of compute in a desktop-deployable footprint

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.