Infrastructure atlas
NVIDIA/Hopper

H200 Tensor Core

Hopper’s memory-heavy sibling, built around 141 GB of HBM3e for larger models and KV caches.

GPUtraininginferencelong contextCUDA
H200 Tensor Core hardware

01 / Memory

141 GB HBM3e

SXM or NVL

02 / Bandwidth

4.8 TB/s

GPU memory bandwidth

03 / Power

Up to 700 W

Configurable SXM TDP

Precision & peak compute

PrecisionPublished peakBasis
FP8 Tensor3,958 TFLOPSSXM, with sparsity
FP16 / BF16 Tensor1,979 TFLOPSSXM, with sparsity
INT8 Tensor3,958 TOPSSXM, with sparsity

Peak FLOPS and TOPS are theoretical hardware figures. Sparse and dense modes, chip and system totals, and form factors are not interchangeable.

NVIDIA H200 specifications

Technical specification

Silicon & memory
Architecture
Hopper
Memory
141 GB HBM3e
FP8 Tensor
3,958 TFLOPS SXM
BF16 / FP16 Tensor
1,979 TFLOPS SXM
Confidential computing
Supported
System & interconnect
Form factors
SXM or PCIe / NVL
Scale-up
NVLink 4 · NVSwitch systems
Partitioning
Up to 7 MIGs
Software
CUDA, TensorRT, NCCL, NIM