NVIDIA H100 NVL 94GB PCIe Dual-GPU for ThinkSystem SR670 V3

The NVIDIA H100 NVL 94GB PCIe Dual-GPU for ThinkSystem SR670 V3 (Lenovo part number 4X67A84518) is a purpose-engineered dual-GPU accelerator module built on NVIDIA's Hopper architecture, designed for validated deployment within the Lenovo ThinkSystem SR670 V3 rack server platform. The NVL form factor pairs two H100 GPUs on a single PCIe board interconnected via NVLink, delivering a combined 94GB of HBM3 memory and substantially increased aggregate memory bandwidth compared to single-GPU configurations — enabling models and datasets that exceed the capacity of a single accelerator to be treated as a unified, high-bandwidth memory pool.

The Hopper architecture introduces the Transformer Engine, which dynamically applies FP8 mixed-precision computation to dramatically accelerate large language model training and inference. The H100 NVL also provides fourth-generation NVLink bridging between the two dies on the module, delivering significantly higher GPU-to-GPU bandwidth than PCIe alone, making it well suited for tightly coupled parallel workloads. Hardware-level support for confidential computing via the NVIDIA Confidential Computing feature set enables trusted execution environments for sensitive enterprise AI workloads.

Validated as a Lenovo ThinkSystem option, this module integrates with Lenovo's systems management toolchain, XClarity provisioning environment, and the SR670 V3's cooling and power subsystems, ensuring enterprise-grade reliability and serviceability. The combination of massive HBM3 memory capacity, high-bandwidth NVLink interconnect, and Hopper-generation compute makes this accelerator a primary choice for organizations running generative AI inference, large-scale model training, high-performance computing simulation, and GPU-accelerated data analytics in datacenter environments across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) training and fine-tuning where combined 94GB NVLink-unified memory allows multi-billion parameter models to reside in a single accelerator module without CPU offloading
• Generative AI inference serving for enterprise deployments requiring high throughput and low latency, leveraging FP8 Transformer Engine acceleration for transformer-based model families
• High-performance computing (HPC) simulations in scientific research, computational fluid dynamics, and molecular dynamics that benefit from the unified high-bandwidth memory pool and PCIe host connectivity
• GPU-accelerated data analytics and in-memory database acceleration for large enterprise datasets requiring fast parallel processing across both GPUs in the module
• Multi-model AI inference consolidation, where the large memory capacity allows multiple concurrent model instances to be loaded and served simultaneously within a single server slot
• Confidential AI workloads in regulated industries such as financial services and healthcare, utilizing NVIDIA's hardware-based Confidential Computing capabilities to protect model weights and inference data

Technical specifications

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