GIGABYTE recently officially launched its first artificial intelligence supercomputer, the AI TOP ATOM, at Computex 2025 in Taipei. This device is a customized version based on the NVIDIA DGX Spark platform, equipped with the GB10 Grace Blackwell Superchip, providing up to 1000 AI TOPS of computing power. It’s designed for professional users such as individual developers, data scientists, and AI researchers. This compact device not only showcases GIGABYTE’s new strategic direction in the AI field but also integrates cutting-edge hardware technology through deep collaboration with NVIDIA and Micron.
The core of the AI TOP ATOM is the NVIDIA GB10 Grace Blackwell Superchip. This chip integrates a Blackwell GPU and a Grace CPU, achieving efficient collaboration through NVLink-C2C interconnect technology. The Blackwell GPU features 6144 CUDA cores and utilizes fifth-generation Tensor Cores, supporting FP4 precision computation to deliver 1 PFLOPS (1 quadrillion operations per second) of AI computing performance. This is ideal for fine-tuning and inference of large-scale AI models. The Grace CPU is based on the ARM architecture and includes 20 high-efficiency cores (10 Cortex-X925 and 10 Cortex-A725), balancing performance and energy efficiency. The NVLink-C2C technology provides 900 GB/s of bandwidth, which is five times that of fifth-generation PCIe, ensuring efficient and consistent data transfer between the CPU and GPU, optimizing memory-intensive AI tasks.
In terms of memory, the AI TOP ATOM is equipped with 128GB of LPDDR5X unified memory, utilizing the latest generation of memory chips provided by Micron. Its 256-bit bus provides a bandwidth of 273 GB/s. This unified memory architecture allows the CPU and GPU to share data, reducing latency and improving the processing efficiency of large models. For storage, it supports M.2 PCIe NVMe slots with optional 1TB or 4TB SSDs. Additionally, the device has a power consumption of only 170W, runs a customized Linux system called “DGX OS,” and is powered by a standard electrical outlet, balancing performance and energy efficiency.
The AI compute power of the AI TOP ATOM supports local operation of AI models with up to 200 billion parameters, covering the prototyping, fine-tuning, and inference needs of mainstream models from DeepSeek, Meta, Google, and more. Through the NVIDIA ConnectX-7 network interface, two AI TOP ATOMs can be interconnected for collaborative work, at which point they can support models with up to 405 billion parameters, suitable for more complex AI training tasks. This expandability makes it suitable not only for individual developers but also for the computing needs of small teams or research institutions. The device comes pre-installed with the NVIDIA AI software stack, simplifying the development process, and models can be seamlessly deployed to data centers or the cloud.
Design and Connectivity #
Visually, the AI TOP ATOM measures just 150 × 150 × 50.5 millimeters and weighs approximately 1.2 kilograms, resembling a mini PC with a simple and practical design. Its interfaces include four USB4 (40Gbps) ports, one RJ-45 port, one HDMI 2.1a port, and a 10GbE ConnectX-7 smart NIC that supports Wi-Fi 7 and Bluetooth 5.3, meeting the demands for high-speed data transfer and display output. Despite its understated appearance, its internal integration of cutting-edge components truly makes it a desktop AI supercomputer.
Industry Collaboration and Future Outlook #
GIGABYTE’s foray into the AI field marks its transformation from a traditional PC hardware manufacturer to an AI computing solutions provider. The AI TOP ATOM not only demonstrates GIGABYTE’s technical strength in hardware integration and optimization but also reflects NVIDIA’s strategic direction of promoting AI democratization. By opening up the design licensing for DGX Spark (originally codenamed Project DIGITS) at GTC 2025, NVIDIA announced a starting price of $3,000 USD (approximately 21,700 RMB), with customized versions launched by partners including GIGABYTE, ASUS, Dell, HP, and Lenovo. GIGABYTE has optimized the thermal and structural design of the AI TOP ATOM, ensuring efficient operation within its compact volume. Micron’s LPDDR5X memory and NVIDIA’s GB10 chip further enhance system performance, giving it a competitive edge in the AI development field.
In terms of video encoding and decoding, the GB10 chip integrates fifth-generation NVDEC and ninth-generation NVENC engines, supporting H.264 and H.265 formats with 4:2:2 specifications. Additionally, the Blackwell GPU’s fourth-generation RT Cores and architecture supporting FP32 computation enable it to excel in graphics rendering and general-purpose computing tasks, further expanding the device’s functional boundaries.
The launch of the AI TOP ATOM comes at a time of rapid growth in the AI hardware market. As the scale of generative AI models continues to expand, there’s a growing demand for efficient and flexible computing platforms. With its compact design, high-performance computing power, and accessible price, the AI TOP ATOM provides an easy entry point to high-performance AI computing for individuals and small teams.