NVIDIA's Hopper H200 AI GPU Breaks MLPerf Record, Outperforming Competitors

With significant upgrades including 141GB of HBM3E memory, the H200 achieves up to 3x faster GenAI performance, now shipping to customers.

NVIDIA's Hopper H200 AI GPU sets a new MLPerf benchmark record, achieving up to 3x faster GenAI performance with TensorRT-LLM optimizations.
The H200 GPU's performance boost is attributed to its upgraded memory, featuring 141GB of HBM3E and up to 4.8TB/sec of memory bandwidth.
NVIDIA's H200 outperforms Intel's Gaudi2 AI accelerator by a significant margin in the latest MLPerf results.
The H200 GPU is now shipping to customers, with custom designs allowing for up to 1000W of power.
NVIDIA continues to lead in the AI hardware market, while AMD and other competitors lag behind in benchmark submissions.