All GPUs
DatacenterHopper

H200

NVIDIA H200

Price per hour
$3.40
Below market rate

The most powerful GPU for AI training and large language models with 141GB HBM3e memory.

Specifications

VRAM
141GB
Memory Type
HBM3e
Bandwidth
4800GB/s
CUDA Cores
16,896
Tensor Cores
528
FP32 Performance
67TFLOPS
FP16 Performance
134TFLOPS
TDP
700W

Architecture

Architecture
Hopper
GPU Chip
GH100

Best For

Large LLM TrainingMulti-model InferenceResearch

Cost Estimate

$3.40
per hour
$81.60
per day
$571.20
per week
$2448.00
per month