NVIDIA H200 PCIe 141GB GPU

The NVIDIA H200 PCIe 141GB GPU is built on the NVIDIA Hopper architecture and represents a significant advance in GPU memory capacity and bandwidth for enterprise AI and high-performance computing workloads. At its core, the H200 PCIe utilizes the same GH100 Tensor Core GPU as the H100, while introducing HBM3e memory technology that delivers 141GB of on-chip memory capacity — nearly double that of the H100 SXM5 — along with substantially increased memory bandwidth. This expanded memory footprint enables the inference and training of very large language models and multimodal AI models that previously required multi-GPU or multi-node configurations.

The H200 PCIe connects to host servers via a PCIe Gen5 x16 interface, making it compatible with a broad range of standard enterprise server platforms without requiring NVLink Switch System infrastructure. The GPU retains the fourth-generation Tensor Cores and Transformer Engine found in the H100, delivering high throughput for FP8, FP16, BF16, TF32, and FP64 precision workloads. NVLink support enables peer-to-peer GPU communication within a server node, and the card supports NVIDIA's full software ecosystem including CUDA, cuDNN, TensorRT, and the NVIDIA AI Enterprise software suite.

Designed for deployment in enterprise data centers, cloud service provider infrastructure, and on-premises AI compute clusters, the H200 PCIe 141GB addresses the growing memory requirements of generative AI inference, scientific simulation, and large-scale data analytics. Its PCIe form factor broadens accessibility across standard rack-mount server designs, while the combination of Hopper compute performance and HBM3e memory bandwidth makes it a capable platform for both training smaller frontier models and serving large deployed models with reduced latency and higher throughput per GPU.

Ideal for

• Large language model inference serving, enabling deployment of models with parameter counts that exceed the memory capacity of previous-generation GPUs within a single card
• Generative AI and multimodal model training for enterprises building proprietary foundation models or fine-tuning open-weight models on domain-specific datasets
• High-performance computing and scientific simulation workloads in fields such as computational fluid dynamics, molecular dynamics, and climate modeling that require both FP64 precision and large memory capacity
• Data analytics and in-memory data processing at scale, where the 141GB HBM3e frame allows large datasets to reside entirely on-GPU for accelerated query and transformation pipelines
• AI-driven drug discovery and genomics research requiring concurrent processing of large biological datasets and complex neural network models
• Enterprise AI infrastructure consolidation, replacing multi-GPU configurations previously needed to accommodate large model memory footprints with a single high-capacity accelerator

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.