ASUS 7U NVIDIA HGX™ H100/H200 eight-GPU server with dual 5th Gen Intel Xeon Scalable processors , designed for large-scale AI and HPC with up to 12 PCIE slots

The ASUS 7U HGX H100/H200 eight-GPU server is a dual 5th Gen Intel Xeon Scalable platform for organizations building large-scale generative AI and HPC systems, with NVLink GPU interconnects and support for up to 12 PCIe slots.

Recommended For

• AI research and infrastructure teams training or fine-tuning large models across eight tightly interconnected NVIDIA GPUs, where HGX, NVLink, and NVSwitch match the workload.
• Organizations running large-model AI inference or generative-AI services that can keep multiple GPUs busy and benefit from high-bandwidth GPU-to-GPU communication.
• HPC and scientific-computing teams with workloads designed to scale across GPUs, such as GPU-accelerated simulations and data-intensive research.
• Enterprise AI operators building a rack-based GPU cluster who need room for high-throughput networking alongside the GPU workload.

Real-World Use Scenarios

• An AI team fine-tunes a large language model on company data; the eight-GPU system lets the team run training workloads on a single server, while final performance depends on the model, software, and data pipeline.
• A company hosts an internal generative-AI service for employee or customer requests; the server can run large-model inference workloads, with actual response speed and user capacity depending on model size, serving software, and request volume.
• A research lab trains models across multiple servers when a job exceeds the practical scale of one node; the HGX system is designed to participate in a cluster, where distributed training can shorten runs but depends on the network, software configuration, and workload.
• A scientific-computing group runs GPU-accelerated simulations or other HPC jobs; workloads that are compatible with the GPU software stack can use the server’s parallel compute resources, while results vary by application and dataset.
• An AI research team evaluates language and vision-language models under sustained GPU workloads; published tests on eight-H100 HGX systems show these platforms are used for LLM and VLM benchmarking, though results vary with cooling, model, and test setup.

Spec Summary