All GPUs
DatacenterNone available
H200
NVIDIA H200
The most powerful GPU for AI training and large language models with 141GB HBM3e memory.
Specifications
VRAM141 GB
Memory TypeHBM3e
Bandwidth4800 GB/s
CUDA Cores16,896
Tensor Cores528
FP32 Performance67 TFLOPS
FP16 Performance134 TFLOPS
TDP700 W
ArchitectureHopper
GPU ChipGH100
Best For
Large LLM TrainingMulti-model InferenceResearch
Price per hour
$3.80
Cost Estimate
per hour$3.80
per day$91.20
per week$638.40
per month$2736.00