Lenovo ThinkSystem SR680a V4 — Product Documentation List & Technical Specifications

Conclusion: The Lenovo ThinkSystem SR680a V4 is an 8U air-cooled AI training platform built on dual Intel Xeon 6 processors with P-cores and eight integrated NVIDIA B300 SXM6 GPUs. It supports up to 4TB of DDR5 memory across 32 DIMM slots, eight 2.5″ NVMe drives, four PCIe Gen5 x16 FHHL slots, and an OCP 3.0 slot, purpose-built for generative AI, HPC, and large-scale inference workloads.

Product Overview

The SR680a V4 is Lenovo’s flagship scale-up GPU server, developed entirely in-house and designed for air-cooled data centers that require maximum GPU performance. It pairs two Intel Xeon 6 processors (6700-series with Performance-cores) with eight NVIDIA HGX B300 GPUs on a single 8U chassis. Each GPU carries 270GB of HBM3e memory, delivering 2.16TB of total accelerator memory per node, interconnected via NVIDIA NVLink and NVSwitch for all-to-all GPU communication.

The platform targets the most demanding workloads: training and fine-tuning large language models (LLMs), generative AI and deep learning, HPC simulations, and large-scale inference. Lenovo claims up to 11x more inference performance than the previous GPU generation. Eight integrated ConnectX-8 800Gb/s OSFP links enable direct GPU-to-network connectivity with GPU Direct support, while XClarity Controller 3 provides out-of-band management with a hardware root of trust.

Technical Specifications

Documentation List