Lenovo ThinkSystem HGX H100 SR675 V3 Server

Tier-one, compact H100. The ThinkSystem SR675 V3 is a 3U dual-socket Lenovo server supporting up to eight double-wide GPUs including the NVIDIA H100, for buyers who want enterprise hardware and support, a small footprint, and GPU flexibility rather than a maximal fixed training mesh. It joins a ThinkSystem estate the way the rest of that fleet is already managed.

Different machine entirely. This SR675 V3 is a 3U, PCIe-GPU, AMD EPYC H100 server. The Lenovo NVIDIA HGX H200 in the catalog is an 8U SXM system with 1.1TB of HBM3e on an NVSwitch mesh. This one leads on footprint, GPU flexibility, and enterprise support; that one leads on memory capacity and coupled training. MillionMiner quotes both depending on your workload.

This server uses double-wide PCIe GPUs, where the H100 cards run as flexible, largely independent accelerators. SXM systems like the Dell XE9680 and Quanta D74H wire eight GPUs into an NVSwitch mesh at 900 GB/s for tightly-coupled training. For inference, HPC, and mixed workloads where GPUs run their own jobs, the PCIe approach here is right and more flexible; for large distributed training, the SXM systems are the tool.

Dual 4th Gen AMD EPYC processors, up to 96 Zen 4 cores and 3TB of DDR5, the core density and memory capacity eight GPUs need during preprocessing and data loading. The sizing rule this catalog applies everywhere holds, system memory at or above the GPU pool, and the configured host is confirmed with your quote.

Up to eight double-wide PCIe Gen5 GPUs, including the NVIDIA H100 and L40 Tensor Core cards. The PCIe architecture is what gives it that flexibility across card types. The exact GPU configuration for your build is confirmed on quote.

spec sheet lists a liquid-to-air heat exchanger, a closed loop that rejects heat to air within the chassis and deploys in an ordinary data center without facility water. Some of the copy refers to it more loosely as liquid cooling, so MillionMiner confirms the exact cooling configuration of your unit on quote, since it determines site requirements.

Up to eight hot-swap drive bays accepting SAS, SATA, or NVMe, so you can run an all-NVMe dataset tier or mix interfaces. The fit-out is specified with your quote, and MillionMiner ships units pre-populated when the storage plan is known.

Support and estate fit. Tier-one hardware passes enterprise procurement because an organization running ThinkSystem manages, monitors, and services this node the way it already runs its fleet, rather than certifying a parallel platform. For corporate IT and regulated environments, that consistency is often the deciding factor over a cheaper non-OEM box.

Choose the SXM H100 systems, the Dell XE9680 or Quanta D74H, when you need the NVSwitch training mesh and 640GB pooled at 900 GB/s. Choose Lenovo's SR680a V3 H200 when memory capacity at 1.1TB per node leads. This SR675 V3 is the pick when footprint, GPU flexibility, and Lenovo support matter more than a coupled training fabric, and the quote routes it honestly.

Submit your workload and deployment details through the quote form. A MillionMiner specialist confirms the GPU and host configuration, the cooling setup, Lenovo warranty and support terms, destination eligibility under the US export controls that apply to Hopper-class accelerators, and delivery. Every system is tested before shipment and shipped worldwide DDP with duties handled, with hosting in MillionMiner's own data centers available.