Recently, GIGABYTE unveiled its inaugural AI supercomputer, the AI TOP ATOM, at Computex 2025. This device, a tailored edition of NVIDIA's DGX Spark platform based on the GB10 Grace Blackwell superchip, delivers an impressive computing capability of up to 1,000 AI TOPS. Targeted towards professionals such as independent developers, data scientists, and AI researchers, this compact machine not only signifies GIGABYTE's strategic entry into the AI sector but also showcases the integration of state-of-the-art hardware technology through profound collaborations with NVIDIA and Micron.

At its core, the AI TOP ATOM harbors the NVIDIA GB10 Grace Blackwell superchip—a sophisticated amalgamation of the Blackwell GPU and Grace CPU, connected seamlessly via NVLink-C2C technology. Featuring 6,144 CUDA cores, the Blackwell GPU leverages the fifth-generation Tensor Core, supporting FP4-precision computation to deliver staggering AI performance at 1 PFLOPS, suitable for large-scale AI model tuning and inference. The Grace CPU, built on the ARM architecture with 20 efficient cores (comprising of 10 Cortex-X925 and 10 Cortex-A725), strikes a balance between performance and energy efficiency. NVLink-C2C provides 900 GB/s bandwidth—quintupling the bandwidth of the latest PCIe generation—ensuring seamless data transfer between the CPU and GPU.

In memory specifications, the AI TOP ATOM is outfitted with 128GB LPDDR5X unified memory, harnessing Micron's latest memory chips along with a 256-bit bus bandwidth of 273 GB/s, facilitating shared data use between the CPU and GPU to minimize latency. Storage solutions include M.2 PCIe NVMe slots for either 1TB or 4TB SSDs. Despite its prowess, the device boasts a power consumption of just 170W, operating on "DGX OS," a specially configured Linux system, and can be powered via a standard outlet—maximizing performance while ensuring energy efficiency.

The AI computing prowess of the AI TOP ATOM enables it to handle local AI models with up to 200 billion parameters, perfectly addressing needs from prototyping to inference in major models like DeepSeek, Meta, and Google. Networking through NVIDIA's ConnectX-7 interface allows coupling of two AI TOP ATOM units, facilitating model handling with up to 405 billion parameters, ideal for complex AI training tasks. This scalability extends its appeal from individual developers to small teams and research entities. Pre-loaded NVIDIA AI software stack simplifies development, offering seamless model deployment in data centers or cloud environments.

Aesthetically, the AI TOP ATOM is compact at 150 x 150 x 50.5 millimeters and weighs approximately 1.2 kilograms, likened to a mini mainframe, embodied in a simple yet functional design. Connectivity options include four USB4 ports (40Gbps), an RJ-45 port, HDMI 2.1a, and a 10GbE ConnectX-7 smart card with Wi-Fi 7 and Bluetooth 5.3 for robust data transfer and display output. Despite its modest exterior, within lies an integration of cutting-edge components transforming it into a desktop-class AI supercomputer.

GIGABYTE positions the AI TOP ATOM as a key component in NVIDIA's DGX Spark ecosystem. Announced at GTC 2025, NVIDIA offers the DGX Spark (previously dubbed Project DIGITS) beginning at $3,000. Partners including Gigabyte, ASUS, Dell, HP, and Lenovo provide customizable versions. GIGABYTE has meticulously optimized the thermal and structural design of the AI TOP ATOM to maintain performance efficiency despite its small footprint. With Micron's LPDDR5X memory and NVIDIA's GB10 chip, system performance reaches new heights, ensuring a competitive edge in AI computing.

The GB10 chip's integration of the latest NVDEC and NVENC engines (fifth and ninth generation respectively) supports H.264 and H.265 encoding at 4:2:2 specifications. The Blackwell GPU's advanced RT cores and FP32 compute-enabled architecture empower superior performance in graphic rendering and general-purpose computing, greatly expanding the device's functionality.

GIGABYTE's foray into AI marks a pivotal shift from being a traditional PC hardware maker to a robust AI computing solution provider. AI TOP ATOM not only highlights GIGABYTE's expertise in hardware innovation and optimization but also underscores NVIDIA's strategic push to democratize AI universality. By sharing DGX Spark's design license, NVIDIA collaborates with premier global system manufacturers, accelerating the adoption of personal AI supercomputing.

The unveiling of AI TOP ATOM coincides with the burgeoning AI hardware market. With the escalation in generative AI model complexity, demand surges for efficient, adaptable computing platforms—the AI TOP ATOM, with its compact form, potent processing capability, and cost-effectiveness, presents an ideal solution for developers and small teams eager to exploit high-performance AI computing.