NVIDIA HGX H200 8-GPU Liquid-Cooled SXM Baseboard (1,128GB HBM3e)

Because of its power and thermal density. Eight H200 SXM GPUs draw on the order of 5.6kW under sustained load, and at that density direct liquid cooling is the efficient answer: it removes heat air struggles with, allows tighter rack packing, runs more efficiently at scale, and sustains the continuous all-GPU loads of large AI training. The liquid-cooled design is what enables the densest, most efficient H200 node deployments, and it is what distinguishes this variant from air-cooled options.

Both, and they are not contradictory. The GPU coldplates on the baseboard are passive, there are no fans or pumps on the board itself. Heat is carried away by coolant supplied from the system's liquid loop. So the board is passive at the component level (no active cooling parts on it) and liquid-cooled at the system level (it relies on the facility's liquid-cooling infrastructure). For an integrator, this means the board needs a compatible liquid loop to function.

No. This is the HGX H200 baseboard, the 8-GPU SXM carrier board only, with no host CPUs, system memory, storage, chassis, or power supplies, and no self-contained cooling. To run, it must be integrated into a system with a host platform, roughly 5.6kW of power delivery, and a compatible liquid-cooling loop. It is a component for builders with liquid-cooling capability, not a turnkey machine. For a finished server, MillionMiner offers complete H200 systems.

A compatible direct-liquid-cooling loop: coolant supply, manifolds, and the connections the GPU coldplates require, sized for roughly 5.6kW of heat from the assembly. The board does not include pumps or a self-contained loop, it draws on the facility or system's liquid-cooling infrastructure. MillionMiner confirms the specific coolant, flow, and connection requirements for the baseboard revision in the quote, and can advise on the surrounding system.

System builders and integrators with liquid-cooling capability who are assembling dense H200 nodes, and operators expanding the GPU assembly in an existing liquid-cooled HGX H200 platform. If you have or are building the chassis, host platform, power, and liquid loop, this baseboard is the high-density core. If you want a server that runs out of the box, or you do not have a liquid-cooling loop, the complete systems or air-cooled options are the better path.

The GPUs, memory (1,128GB HBM3e), and NVLink fabric are the same flagship Hopper assembly. The difference is cooling and density: this variant is built for direct liquid cooling, which suits the highest-density, most efficient deployments but requires a liquid-cooling loop, while an air-cooled baseboard fits air-cooled chassis but at lower density. Choose this one if you have liquid-cooling infrastructure and want maximum density; choose air-cooled if your facility is air-cooled. MillionMiner matches the variant to your data center.

Memory. The H200 is the high-memory Hopper part, with 141GB of HBM3e per GPU and more bandwidth than the H100, for 1,128GB across the baseboard. For the largest models and memory-bound training and inference, that extra capacity and bandwidth per GPU is the H200's defining advantage. The liquid-cooling here lets you deploy that memory advantage at the highest density.

Yes, unreservedly. This is flagship data center AI silicon: eight top-bin Hopper H200 GPUs with full HBM3e and full NVLink fabric, the configuration used to train and serve the largest AI models. Unlike the workstation and display cards elsewhere in this catalog, there is no overstatement to qualify, only the scope to note that it is supplied as a liquid-cooled baseboard for integration.

Eight H200 SXM GPUs on one HGX baseboard, 1,128GB of HBM3e total (141GB per GPU), full NVLink and NVSwitch interconnect, Hopper architecture, roughly 5.6kW power under load, built for direct liquid cooling with passive GPU coldplates fed by the system loop. MillionMiner confirms the exact baseboard revision, the per-GPU and aggregate performance specs, and the cooling and power requirements in the quote.

Submit your build, the liquid-cooled node you are integrating or the platform you are expanding, and quantity through the quote form. A MillionMiner specialist confirms the baseboard configuration and condition, the liquid-cooling and power requirements, export eligibility for Hopper-class hardware, and whether this liquid-cooled baseboard, an air-cooled one, 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 liquid-cooling-capable data centers available.