[Here](https://github.com/Infrasys-AI/AIInfra/blob/main/01AICluster/02L0L1Base/01Server.md#:~:text=GPU%EF%BC%9A8%20%E5%BC%A0%20NVIDIA%20HGX%20H100%EF%BC%8C%E5%9F%BA%E4%BA%8E%20Hopper%20%E6%9E%B6%E6%9E%84%EF%BC%8C%E6%94%AF%E6%8C%81%20FP8%20%E7%B2%BE%E5%BA%A6%E8%AE%A1%E7%AE%97%EF%BC%8C%E5%8D%95%E5%8D%A1%20FP8%20%E7%AE%97%E5%8A%9B%E8%BE%BE%20200%20PFlops%EF%BC%8C8%20%E5%BC%A0%E6%80%BB%E7%AE%97%E5%8A%9B%201%2E6%20PFlops) The official [doc](https://www.nvidia.com/en-us/data-center/h100/) says the H100 has around 2000 TFLOPS (dense) on fp8, a dgx device is then 16 PLOPS.
Here
The official doc says the H100 has around 2000 TFLOPS (dense) on fp8, a dgx device is then 16 PLOPS.