Benchmark Scores for NVIDIA and MediaTek’s Co-Developed GB10 Chip Revealed: Clock Speed Hits 3.9GHz

On May 12, according to tech media Notebookcheck, performance benchmark data for the NVIDIA GB10 Grace Blackwell Superchip, powering NVIDIA’s first personal supercomputer Project DIGITS, has surfaced in the Geekbench database, marking the official entry of this groundbreaking product into testing. This development further solidifies NVIDIA’s leadership in the AI computing arena.

From Concept to Reality: Technical Progress Post-CES Debut

NVIDIA initially unveiled Project DIGITS at CES earlier this year, with its standout feature being the GB10 Grace Blackwell Superchip, co-developed with MediaTek. The chip adopts a hybrid architecture combining the Blackwell GPU and Grace CPU, leveraging NVLink-C2C high-speed interconnect technology for low-latency, high-bandwidth communication between components. Equipped with 128GB of LPDDR5X memory and 1TB/4TB NVMe SSD storage, the system can locally run large language models (LLMs) with over 200 billion parameters, delivering data center-level AI performance for individual users.

Performance Breakthrough: 1 Petaflop AI Power

According to official specifications, the GB10 integrates NVIDIA’s latest CUDA cores and fifth-generation Tensor cores, achieving AI compute performance of up to 1 petaflop (one quadrillion floating-point operations per second) with significantly improved energy efficiency. This capability enables efficient handling of demanding tasks like generative AI and complex scientific simulations. Additionally, the integrated Grace CPU, based on Arm architecture and co-designed with MediaTek, features 20 high-performance cores, offering industry-leading connectivity and power efficiency to complement the GPU’s capabilities.

MediaTek Collaboration: A Powerhouse Alliance in Arm Ecosystem

The launch of the GB10 highlights the synergy between NVIDIA and MediaTek in chip design. As a global leader in Arm-based system-on-chip (SoC) design, MediaTek infused its expertise in low-power mobile technologies into the Grace CPU, enabling high performance while minimizing energy consumption. This partnership not only expands Arm’s footprint in supercomputing but also sets a technical benchmark for future edge AI devices.

Industry Impact: A Milestone in Personal AI Computing

The testing progress of Project DIGITS signals a shift of high-performance AI computing from the cloud to the edge. Users can now train and deploy massive AI models locally without relying on remote servers, a critical advancement for data-sensitive industries like healthcare and finance, as well as research institutions. With Geekbench data now public, industry observers anticipate the product will enter mass production in 2024, accelerating the democratization of AI.

NVIDIA has yet to announce pricing or a release timeline for Project DIGITS, but its hardware specs and performance have already captivated the tech world. If successfully commercialized, this device could become a cornerstone infrastructure for developers, researchers, and enterprises building private AI systems.