NVIDIA HGX A100 8-GPU SXM4 Baseboard, 40GB (320GB Total)

No. This is the HGX A100 baseboard (935-23587-0000-000), the 8-GPU SXM4 carrier board only. It has no host CPUs, system memory, storage, chassis, power supplies, or fans. To run, it must be integrated into a system with a host platform, the power an 8-A100 assembly requires, and appropriate cooling. It is a component for builders and integrators, not a turnkey machine. For a finished server, MillionMiner offers complete A100 systems.

This is the 40GB-per-GPU variant, 320GB of HBM2 across the eight-GPU board, the original A100 memory tier. The alternative is the 80GB version (640GB total). Both share the same Ampere architecture, SXM4 form, and NVLink/NVSwitch fabric; the difference is memory per GPU. Choose the 40GB for workloads that fit in 40GB per GPU at a more economical price, or the 80GB for larger or memory-bound models.

System builders and integrators assembling A100 nodes, and operators repairing or expanding the GPU assembly in an existing HGX A100 platform. If you have the chassis, host platform, power, and cooling, or are building them, the baseboard is the core 8-GPU assembly. If you want a server that runs out of the box, the complete A100 systems are the right path instead.

By NVLink and NVSwitch, at up to 600 GB/s, the high-bandwidth fabric that interconnects all eight GPUs for fast all-to-all communication, so they work as a single complex for multi-GPU training and inference. This full NVLink/NVSwitch fabric is what distinguishes a true HGX baseboard from eight separate PCIe cards, and is essential to scaling workloads across the eight GPUs efficiently.

AI training and inference, HPC, and data analytics. The A100 has strong FP64 double precision for scientific and engineering computing, Tensor cores for AI, high HBM2 bandwidth, and MIG partitioning for multi-tenant use. It is a generation behind the current Hopper and Blackwell silicon, but remains a capable and cost-effective platform for many workloads, particularly FP64 HPC and multi-tenant inference.

MIG (Multi-Instance GPU) partitions each A100 into as many as seven isolated instances, each with dedicated memory and compute. Across eight GPUs that allows fine-grained, multi-tenant resource allocation, serving many workloads or users with guaranteed quality of service, which is part of why A100 platforms suit cloud and multi-tenant data center deployments.

Eight A100 SXM4 GPUs on one HGX baseboard (935-23587-0000-000), 40GB of HBM2 per GPU at 1.6 TB/s and 320GB total, NVLink and NVSwitch interconnect up to 600 GB/s, Ampere architecture, PCIe Gen4 host interface. MillionMiner confirms the exact baseboard revision, the full performance specs, and the power and cooling requirements in the quote.

An 8-GPU A100 assembly draws several kilowatts under load and requires data center power delivery and cooling appropriate to the SXM4 baseboard, supplied by the host system it is integrated into. This is data center infrastructure, not a workstation part. MillionMiner confirms the power and cooling requirements for your deployment in the quote.

As an Ampere data center part, units may be new or secondary-market, and MillionMiner verifies each baseboard and confirms its condition in the quote. The A100 is also an export-regulated accelerator, so MillionMiner confirms destination eligibility and the compliance position before sale. Condition and compliance are established before you order.

Submit your build, the node you are integrating or the HGX A100 platform you are expanding, and quantity through the quote form. A MillionMiner specialist confirms the baseboard configuration and condition, the power and cooling requirements, export eligibility, and whether this 40GB baseboard, the 80GB variant, or a complete system best fits your build, then delivery. Every unit is verified before shipment and shipped worldwide DDP with duties handled, with hosting in MillionMiner's own data centers available.