# FP8 Training: H100 Native, RTX 4090'da Prematur — Transformer Engine Internals

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-fp8-training-h100-rtx4090-prematur
> Updated: 2026-08-13T18:51:47.428Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part X — Quantization Engineering
**TLDR:** FP8 = AI compute'un geleceği. H100 native (FP8 Tensor Cores + WGMMA + Transformer Engine). RTX 4090 (Ada) FP8 GEMM destekler ama ekosistem hazır değil — fallback yaygın, training pipeline buggy. Cookbook'un kuralı: RTX 4090'da bf16 training, FP8 inference (vLLM). H100'de FP8 training cookbook Part XIII Triton'da derin.