Unverified50% confidenceFactExact time
The FP8 format was jointly proposed by NVIDIA, ARM, and Intel in 2022.
1
Sources
50%
Confidence
Long-term
Relevance
8/19/2026
First Seen
Sources
魔改2080Ti本地部署Qwen3 27B实测：FP8量化低成本方案
Related Claims
VerifiedFP8 was first introduced with hardware support by NVIDIA in the H100 architecture, with two subtypes: E4M3 and E5M2.71% similarUnverifiedFP8由NVIDIA在Hopper架构（H100）中引入硬件原生支持，随后在Ada Lovelace（RTX 40系）和Blackwell（RTX 50系）架构中持续完善69% similarVerifiedFP8 is an 8-bit floating-point format natively supported by NVIDIA's Hopper architecture (H100), which preserves better dynamic range compared to INT8 quantization.68% similarUnverifiedFSR 4.1 will arrive on RDNA 3 architecture GPUs (Radeon RX 7000 series) in July 2025.66% similarUnverifiedNVIDIA在Ampere架构（RTX 30系，2020年）引入专用INT8 Tensor Core使INT8推理速度相比FP16提升约2倍，NVFP4是为Blackwell（RTX 50系，2025年）定制的4位浮点格式64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/775812API
curl https://kongchang.com/api/v1/knowledge/claims/775812MCP
get_claim(id=775812)