// HOPPER
Available
H200 SXM
141 GB HBM3e for long-context inference without the spill.
Best for
long-context inferenceRAG
Available in 5 regions
us-west-1us-east-1eu-central-1apac-sg-1ap-ph-1
- Architecture
- Hopper
- Form factor
- SXM
- Memory
- 141 GB HBM3e
- Bandwidth
- 4.8 TB/s
- NVLink
- Gen 4
- Board power
- 700 W
// SOFTWARE STACK
CUDANVLinkTensorRT-LLMvLLMSGLangNVIDIA DynamoNVFP4KV-cachespeculative decoding