NVIDIA H200 specifications (web)
Citation
NVIDIA. “NVIDIA H200 GPU.” Product web page with specifications, date not stated. https://www.nvidia.com/en-us/data-center/h200/. Read 2026-11-09.
Bottom line
- The H200 is a Hopper-generation GPU with 141GB of HBM3e memory at 4.8TB/s. NVIDIA says that is “nearly double the capacity” of the H100.
- Its compute figures match the H100 SXM, and NVIDIA says it has “the same power profile as the H100”.
Background
NVIDIA makes GPUs for AI and high-performance computing (HPC). This is its product page for the H200, with performance claims and a table for the H200 SXM and H200 NVL. The page marks the table as preliminary. The page gives no price.
Key claims
- “the NVIDIA H200 is the first GPU to offer 141 gigabytes (GB) of HBM3e memory at 4.8 terabytes per second (TB/s) —that’s nearly double the capacity of the NVIDIA H100 GPU with 1.4X more memory bandwidth.” (Higher Performance With Larger, Faster Memory)
- “The H200 boosts inference speed by up to 2X compared to H100 GPUs when handling LLMs like Llama2.” (Unlock Insights With High-Performance LLM Inference)
- “This cutting-edge technology offers unparalleled performance, all within the same power profile as the H100.” (Reduce Energy and TCO)
- Specifications, SXM then NVL: ”| Max Thermal Design Power (TDP) | Up to 700W (configurable) | Up to 600W (configurable) |” (Specifications).
- ”| FP8 Tensor Core² | 3,958 TFLOPS | 3,341 TFLOPS |”, where “2 With sparsity” and “1 Preliminary specifications. May be subject to change.” (Specifications).
Evidence and method
A vendor’s product page. Performance claims are NVIDIA’s own tests, with settings in footnotes. The specifications are preliminary.
Limits
- No price or cost of ownership.
- The specifications may change, as the page says.