The NVIDIA H100 94GB Tensor Core GPU (commonly deployed as H100 NVL) is a data-center accelerator engineered for large-scale Generative AI and high-throughput LLM inference, while also excelling in HPC and advanced analytics. Built on the NVIDIA Hopper™ architecture, it pairs 94GB of high-bandwidth memory with exceptional bandwidth to keep large models fed efficiently—helping reduce memory bottlenecks and increase tokens/sec in production AI serving.
Designed for modern PCIe server deployments, H100 94GB is ideal for scalable inference clusters and GPU-accelerated data centers that need high compute density, fast memory bandwidth, and efficient scaling for demanding transformer workloads.
Key Highlights
- 94GB GPU memory for larger models, higher batch sizes, and longer context workloads.
- Up to 3.9 TB/s memory bandwidth (H100 NVL) to accelerate memory-bound AI inference.
- Hopper architecture + Tensor Cores with Transformer Engine (FP8) optimized for transformer/LLM performance.
- Optimized for LLM inference with high compute density and energy efficiency for data-center serving.
- NVLink-based scaling (NVL configuration) enables higher effective memory pool and faster multi-GPU communication for large inference workloads (platform/config dependent).
