Newly-Launched NVIDIA DGX Spark Features GB10 Superchip Co-Designed by MediaTek

Rhea-AI Summary

MediaTek (OTC:MDTKF) co‑designed the GB10 Grace Blackwell Superchip used in NVIDIA DGX Spark, a personal AI supercomputer now publicly available as of October 15, 2025. GB10 pairs a 20‑core Arm Grace CPU with the latest Blackwell GPU and 128GB unified memory to deliver up to 1 PFLOP of AI performance for model tuning and real‑time inference.

The DGX Spark supports single‑system work on models up to 200 billion parameters and can link two systems using ConnectX‑7 networking to enable inference on models up to 405 billion parameters. The system targets desktop form factors with standard outlet power usage.

Positive

• AI performance up to 1 PFLOP
• 128GB unified memory for large models
• Supports models up to 200B parameters
• Scales to 405B parameters via two linked systems
• Desktop form factor with standard outlet power

Negative

• None.

Continues MediaTek's long history of collaboration with NVIDIA to advance AI innovation from supercomputers to vehicles, IoT devices, data center solutions, and beyond

HSINCHU, Oct. 15, 2025 /PRNewswire/ -- MediaTek has teamed with NVIDIA on the design of the GB10 Grace Blackwell Superchip in NVIDIA DGX Spark, a personal AI supercomputer that allows developers to prototype, fine-tune, and inference large AI models on the desktop. Announced earlier this year, DGX Spark will be available to the public starting today to drive the next wave of AI development across industries.

The GB10 Grace Blackwell Superchip leverages MediaTek's experience in designing power-efficient and high-performance CPU, memory subsystem, and high-speed interfaces to power the Grace 20-core Arm CPU. Combined with the latest generation Blackwell GPU and 128GB of unified memory, GB10 delivers up to 1 PFLOP of AI performance to accelerate model tuning and real-time inferencing. This configuration allows developers to work with large AI models up to 200 billion parameters or use the built-in ConnectX-7 networking technology to connect two DGX Spark systems together to enable inference on models up to 405B parameters. DGX Spark packs all this performance into a compact design that can easily fit onto a desktop and is power-efficient enough to use a standard electrical outlet.

"DGX Spark will usher in the next era of AI prototyping and advance our mission of making great technology more accessible from the edge to the cloud, while solving performance and power consumption challenges," said Vince Hu, Corporate Vice President of MediaTek's Data Center and Compute Business Group. "The GB10 Superchip utilizes our high-performance computing expertise for the data center in combination with our power savings technologies for consumer devices, custom-built to run AI workloads."

The GB10 collaboration builds on MediaTek's work with NVIDIA across multiple verticals, bringing advanced AI capabilities to hyperscale data centers, IoT applications, and software-defined vehicles.

MediaTek partners with leading brands to turn breakthrough ideas into products that scale, from everyday devices to enterprise and cloud systems. Our platforms unify compute, AI, connectivity, and system software to deliver industry–leading performance per watt in innovative, reliable designs. With deep expertise in architecture, integration, manufacturing, and lifecycle services, we help customers accelerate time-to-market and sustain products at global scale.

To learn more about MediaTek's ASIC solutions, please visit: https://www.mediatek.com/asic.

About MediaTek Inc.

MediaTek (TWSE: 2454) is a global leader in fabless semiconductor design, providing solutions from the edge to the cloud. Powering over 2 billion connected devices every year, MediaTek's leading-edge technology keeps the world connected and enhances everyday life.  At the forefront of innovation, MediaTek drives advancements in transformative technologies such as AI, 5G/6G, and Wi-Fi 7/Wi-Fi 8. Our high-performance, power-efficient products form the foundation for a smarter, more connected world, enabling devices from smartphones, smart homes and AI PCs to high-performance computing, automotive, and AI data centers. As a trusted partner to the world's leading brands, MediaTek leads the industry in creating solutions that meet the evolving needs of the global community, ensuring access to world-class technology for everyone. Our commitment to accelerating AI underscores our dedication to enriching the future of humanity. Visit http://www.mediatek.com for more information.

PR Contact: Kevin Keating, pr@mediatek.com

View original content to download multimedia:https://www.prnewswire.com/news-releases/newly-launched-nvidia-dgx-spark-features-gb10-superchip-co-designed-by-mediatek-302584156.html

SOURCE MediaTek Inc.

FAQ

What did MediaTek and NVIDIA announce about DGX Spark on October 15, 2025 (MDTKF)?

They announced MediaTek co‑designed the GB10 Superchip powering NVIDIA DGX Spark, available to the public starting October 15, 2025.

How much AI performance does the GB10 Superchip deliver in DGX Spark (MDTKF)?

GB10 delivers up to 1 PFLOP of AI performance for model tuning and real‑time inference.

What model sizes can DGX Spark handle with MediaTek GB10 (MDTKF)?

Single DGX Spark supports models up to 200 billion parameters; linking two systems enables up to 405 billion parameters.

What CPU and memory specs are in the GB10 Superchip used in DGX Spark (MDTKF)?

GB10 integrates a 20‑core Arm Grace CPU and 128GB unified memory combined with a Blackwell GPU.

Is DGX Spark designed for desktop use and standard power outlets (MDTKF)?

Yes. DGX Spark is described as a compact desktop design that is power‑efficient enough to run from a standard electrical outlet.

How does DGX Spark scale networking for larger inference workloads (MDTKF)?

DGX Spark uses built‑in ConnectX‑7 networking to link two systems, enabling inference on models up to 405B parameters.