홈 › 칩 › NVIDIA H100

NVIDIA H100

칩 GPU 출하

The GPU of the ChatGPT era — the unit of account for the 2023–25 AI build-out.

현재 보고 계신 것은 칩 레코드입니다: 이 프로세서가 무엇인지, 주요 사양, 설계·제조·부품 공급사, 그리고 어디에 활용되는지를 다룹니다.

Hopper's H100 became the defining commodity of the AI boom: clusters were sized in H100s, startups raised funding denominated in them, and export policy was written around them. Its Transformer Engine pairs fourth-generation tensor cores with FP8 precision, roughly tripling A100 training throughput per chip.

xAI's Colossus assembled roughly 100,000 of them in 122 days; Meta, Microsoft, OpenAI, Google and every major cloud deployed them in the hundreds of thousands. Blackwell succeeded it in 2025, but H100/H200 fleets still carry a large share of world AI compute.

사양 및 주요 정보

각 수치에는 근거 유형이 표시됩니다: 벤더 사양, 벤더 주장, 보고된 값, 추정치, 또는 공식 발표. "공개되지 않음"은 그대로의 의미이며, 값을 임의로 만들어내는 일은 없습니다.

Overview

Designer NVIDIA

Architecture Hopper (GH100)

Announced March 2022 (GTC); shipping volume from early 2023

Status Shipping — succeeded by Blackwell at the frontier

Manufacturing

Foundry / process TSMC 4N (custom 5 nm-class)Vendor spec

Transistors 80 billionVendor spec

Die size 814 mm²Vendor spec

Packaging CoWoS-S with 5 or 6 HBM stacksVendor spec

메모리

Memory 80 GB HBM3 (SXM); 94 GB HBM3 (NVL); 80 GB HBM2e (PCIe)Vendor spec

Memory bandwidth 3.35 (SXM) / 3.9 (NVL) TB/sVendor spec

컴퓨트

TF32 tensor (dense) 495 TFLOPSVendor spec NVIDIA datasheets often headline throughput with structured sparsity; figures here are dense unless noted.

BF16/FP16 tensor (dense) 990 TFLOPSVendor spec

FP8 tensor (dense) 1,979 TFLOPSVendor spec FP8 via the Transformer Engine was Hopper's signature feature

INT8 tensor (dense) 1,979 TOPSVendor spec

FP64 tensor 67 TFLOPSVendor spec

AI Performance

MLPerf Set the reference results for training and inference benchmarks across 2023–24 submissionsMLPerf

전력

TDP up to 700 (SXM) / 350 (PCIe) WVendor spec

Physical

Form factors SXM5, PCIe, NVL (paired PCIe)

Interconnect NVLink 4 — 900 GB/s; PCIe 5.0Vendor spec

가격

Launch price Not publicly disclosed

Market price ~$25,000–40,000 per GPU at the 2023–24 peakReported

Cloud rental Roughly $2–7 per GPU-hour across major and specialist clouds during 2025 — falling steadilyReported varies with commitment and provider

전체 스택에서의 연결 구조

이 레코드의 링크를 한눈에 확인하세요 — 그룹을 클릭하면 해당 위치로 이동합니다. +11 아래에 추가 그룹이 있습니다.

설계사

NVIDIA

제조

TSMC

공정 노드

TSMC N5 / 4N (5 nm class)

패키징

CoWoS

패키징 공급업체

TSMC

메모리 기술

HBM3

메모리 공급사

SK hynix

인터커넥트 및 네트워킹

NVLink & NVSwitch InfiniBand

포함된 위치

Dell PowerEdge XE9680/XE9712 Supermicro NVL72 & 4U Liquid-Cooled Systems NVIDIA HGX H100/H200

실행

Triton (OpenAI) PyTorch Megatron-Core vLLM

지원 주체

CUDA

학습에 사용됨

Llama 3.1 405B Llama 4 Grok 3

출처

Amazon Web Services Microsoft Google (Alphabet) Oracle CoreWeave Lambda

배포처

xAI Colossus

기초 자료

NVIDIA H800

후속 제품

NVIDIA H200

이전 제품

NVIDIA A100

경쟁 제품

AMD Instinct MI300X Google TPU v5p Groq LPU AWS Trainium2 Intel Gaudi 3 Huawei Ascend 910C

출처

NVIDIA H100 page ↗ · Hopper architecture whitepaper ↗

검토 기준일 2026-01-15 — 보기 방법론.