NVIDIA HGX A100 8-GPU 640GB SXM Baseboard (935-23587)

No. This is the HGX A100 baseboard, the integrated eight-GPU SXM module, NVIDIA part 935-23587, sold as a component. It is the part system builders mount into a chassis with CPUs, memory, storage, power, and cooling to make a finished server. If you need a complete machine, MillionMiner supplies the A100 systems built around this board, and the newer H100 and H200 systems.

Usually not. A100 is the Ampere generation, before H100 and H200, so a new build generally starts with Hopper. The A100 baseboard's buyer is protecting an existing investment, servicing a failed board, or adding nodes to a fleet already standardized on A100. For a greenfield project, MillionMiner will point you to the Hopper systems unless A100 economics specifically suit the workload.

Eight A100 SXM GPUs and the NVSwitch fabric connecting them, 640GB of HBM2e in total at the 80GB-per-GPU configuration this part number denotes. It does not contain CPUs, system memory, storage, or networking, those live on the host tray you mount it into. It is the GPU-and-interconnect heart of an A100 SXM system.

Three buyers. Operators servicing an A100 fleet who need to replace a failed baseboard with the matching part. Data centers standardized on A100, with validated images and host platforms in place, adding capacity in the same generation rather than introducing a second. And integrators building or refurbishing A100 systems for buyers who want Ampere economics. All three value generation continuity.

For many workloads, yes. A100 trained much of the model generation that defined the field and remains capable, and at secondary-market economics it is often sufficient. Choose it with eyes open, though: H100 brings roughly triple the training throughput and FP8, and H200 adds large memory. MillionMiner helps weigh extending A100 against moving to Hopper for your specific spend.

640GB of HBM2e across the eight GPUs, which is the 80GB-per-GPU A100 SXM configuration that the 935-23587 part number's variant denotes. The earlier A100 generation also came in 40GB per GPU; this is the 80GB board. MillionMiner confirms the exact variant of the board you receive in the quote.

HGX A100 baseboards mount into host platforms designed for them, the A100-generation chassis and host trays. Compatibility is specific, and for fleet service the matching host matters, so MillionMiner confirms host and chassis compatibility for your build or repair in the quote rather than leaving it to chance.

A warranty term is confirmed for the specific board in your quote. As an Ampere-generation and often secondary-market part, the coverage picture differs from a new complete system, and MillionMiner clarifies the board's condition and coverage before you order, the same way it handles the catalog's used A100 systems.

Yes. MillionMiner supplies the complete A100 systems built around this board, the newer H100 and H200 systems if moving generation makes sense, and hosting in its own data centers. The quote establishes whether servicing the board, buying a complete A100 system, or stepping to Hopper is the right call.

Submit your build or service details, host, chassis, and purpose, through the quote form. A MillionMiner specialist confirms the baseboard variant, compatibility, cooling and power requirements, warranty and condition, destination eligibility under the US export controls that apply to A100-class accelerators, and delivery. The board is verified before shipment and shipped worldwide DDP with duties handled.