Gigabyte showed off its new AI TOP ATOM at Computex this year. The company built this tiny supercomputer based on NVIDIA DGX Spark tech. It packs serious power in a small case with its GB10 Grace Blackwell Superchip. The device delivers about 1,000 AI TOPS, making it 20 to 50 times faster than current NPU engines.
Inside the machine, fifth-generation Tensor Cores with FP4 support sit. These cores can handle a trillion operations every second for AI tasks. The system uses Micron LPDDR5X memory chips with high bandwidth. The team labeled the GB10 chip as an engineering sample, meaning the final product might change before release.
Gigabyte claims the ATOM can work with 70 billion parameters for AI models. Two units connected through NVIDIA Connect-X can boost that number to 405 billion parameters. The plain-looking box comes with HDMI, Type-C, and ConnectX-T ports. This compact AI powerhouse represents where computer tech heads next.
Inside the machine, fifth-generation Tensor Cores with FP4 support sit. These cores can handle a trillion operations every second for AI tasks. The system uses Micron LPDDR5X memory chips with high bandwidth. The team labeled the GB10 chip as an engineering sample, meaning the final product might change before release.
Gigabyte claims the ATOM can work with 70 billion parameters for AI models. Two units connected through NVIDIA Connect-X can boost that number to 405 billion parameters. The plain-looking box comes with HDMI, Type-C, and ConnectX-T ports. This compact AI powerhouse represents where computer tech heads next.
