900-21010-0120-030

• NVIDIA Tesla H100 Accelerator card, Hopper GPU Architecture
• 94GB HBM2e Memory on a 6016-Bit Bus
• 3938GB/S Memory Bandwidth
• 1080M Hz Core clock, 1785M Hz Boost Clock
• PCIe 5.0 x16 System interface, Dual-Slot PCIe Form Factor
• ECC Memory Enabled by default; 350W TDP; Professional Compute Class

The 900-21010-0120-030 is an NVIDIA Tesla H100 accelerator card with 94GB of HBM2e memory. Built on the Hopper architecture, it runs at a 1080M Hz core clock with a 1785M Hz boost clock and pairs a 6016-bit memory bus with 3938GB/s of memory bandwidth. The card uses a dual-slot PCIe form factor with a PCIe 5.0 x16 system interface, has ECC memory enabled by default, and carries a 350W TDP. The 900-21010-0120-030 is a professional-class accelerator rather than a display graphics card, intended for compute rather than desktop rendering.

H100-class accelerators are deployed for large-scale AI training and inference, high performance computing simulation, and data analytics, where memory bandwidth and capacity determine how large a model or dataset can be held on a single device. The near-4TB/s HBM2e bandwidth is the defining characteristic for transformer training and inference serving, and ECC-enabled memory is required for long-running jobs where a silent bit error would corrupt results. The PCIe form factor allows deployment in standard servers rather than proprietary module chassis. Buyers are typically cloud and AI infrastructure operators, HPC and research computing centers, enterprise AI teams, and server OEMs and integrators building GPU nodes.