Lenovo ThinkSystem SR675 V3 supply

About this model

The ThinkSystem SR675 V3 is the GPU flagship of our catalogue: a 3U system carrying eight GPUs on dual 4th-generation AMD EPYC hosts, positioned in the listing for AI training and high performance computing. It is the most fully specified accelerator node we publish, and the anchor of the GPU range.

Configuration choices

Eight accelerators in one node is the training topology the frameworks are tuned for, and the dual EPYC hosts exist to keep them fed — host memory, local NVMe staging and network throughput are sized around the cards, not the other way round. A serious inquiry for this model names the accelerator model, the framework and the largest model to be trained; those three facts are the configuration.

Power is the fourth fact. A loaded SR675 V3 is among the heaviest continuous loads in any rack; the per-rack budget at the delivery site belongs in the first message, not the last.

Where it fits

Buyers put this model to work in three recurring patterns:

• AI training nodes and training cluster building blocks.
• HPC systems with accelerator-resident codes.
• Large-model inference served from one shard group.

In service

Operating an eight-GPU node is operating a small power station that computes. Its service plan starts with the electrical facts — dedicated feeds, monitored PDUs, thermal headroom watched at the rack level — and continues through accelerator care: driver baselines frozen per training programme, firmware moved deliberately between programmes rather than during them, and utilisation telemetry kept as the node's report card.

Spares strategy for this class is component-level and honest: a failed accelerator or a host board is a programme-stopping event, so estates running training deadlines hold the critical spares or hold a maintenance arrangement that guarantees them. The purchase conversation is the right place to decide which, in writing — the middle of a training run is the wrong place to discover the answer.

The specification, in one place

The published figures for the SR675 V3 are the ones carried on this page: form factor: 3U GPU server; accelerators: 8 GPUs; processors: 2 × AMD EPYC 4th Gen; positioning in the listing: AI training and high performance computing. Those figures are the anchor of the supply offer — they are what the configuration starts from, and every line beyond them (processor alternatives, memory populations, drive models and counts, power and rail options) is settled in writing on the quote rather than assumed from the badge.

Buyers comparing this model should hold the comparison to the same standard: configuration against configuration, line by line. In the Lenovo range this model is quoted alongside its ThinkSystem siblings in the same format, so a two-model comparison on one brief is always available. Ask for serial numbers before shipment, and keep the final written configuration with the asset record — it is the document the machine's whole service life will be checked against.