a4x-maxgpu-4g-metal by Google Cloud Platform

An accelerator-optimized bare-metal platform featuring four integrated GPUs and a high-core-count ARM processor for intensive parallel computing.

Google Cloud Platform a4x-maxgpu-4g-metal is an accelerator-optimized bare-metal server designed for intensive parallel computing. It features 144 dedicated arm64 vCPUs and 960.0 GB of system memory, yielding 6.67 GB of RAM per core. The primary computational power is delivered by four integrated NVIDIA GB300 GPUs. Tradeoffs of this configuration include the complete absence of local storage and zero complimentary public IPv4 addresses, necessitating network-attached storage and external routing solutions. Lacking x86-specific vector extensions, the arm64 architecture requires compatible software builds. This server is engineered for workloads such as large-scale machine learning model training, deep learning inference, high-performance computing, and complex data science pipelines requiring dense GPU acceleration and high-capacity dedicated host memory.