Shenzhen, China – July 2025 – In a breakthrough for AI and large language model computing, NVIDIA has launched the H100 141GB NVL Original, a next-generation solution designed to push the boundaries of AI performance, memory capacity, and energy efficiency. Built with enterprise-scale AI workloads in mind, the H100 NVL 141GB module is tailored to meet the demands of today's cutting-edge data centers, language model training, and inferencing at unprecedented speed and scale.
3X Faster LLM Inference – Powered by NVL Architecture
Unlike standard SXM or PCIe GPUs, the H100 NVL is a unique form factor featuring two H100 GPUs connected via NVLink and packaged as a dual-GPU solution with 141GB of shared HBM3 memory and a memory bandwidth of over 7TB/s. This makes it the ideal GPU for:
GPT-3, GPT-4, and LLaMA 3 training
Large-scale transformer inference workloads
Multi-modal AI tasks (vision-language, speech, etc.)
According to NVIDIA’s internal testing, H100 NVL provides up to 12x faster inference for large language models compared to the A100 80GB.
