Unverified70% confidenceFactExact time
The Transformer Engine of the Hopper architecture can dynamically switch precision between FP8 and FP16 on a per-layer basis.
2
Sources
70%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedThe Hopper architecture of the H100 introduced a dedicated Transformer Engine supporting FP8 precision computation.76% similarUnverifiedCompared with the previous generation A100, the Hopper architecture has significant improvements in FP8 precision, the Transformer engine and NVLink interconnect bandwidth, making it the de facto standard hardware for large-scale language model training since 2023.71% similarUnverifiedFuTransformer is a CUDA implementation targeting NVIDIA Hopper GPUs, with a separate FP8 prefill path and INT4 decode path designed for Llama-style Transformer blocks.71% similarUnverifiedH100的Transformer Engine支持FP8精度，张量核心在FP8模式下峰值算力可达3958 TFLOPS，FP16为1979 TFLOPS70% similarVerifiedBlackwell's second-generation Transformer Engine adds FP4 precision support, which in theory can improve Transformer model training throughput by 2–4x compared to the H100.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/511291API
curl https://kongchang.com/api/v1/knowledge/claims/511291MCP
get_claim(id=511291)