Nvidia Hopper architecture

URI: https://gptkb.org/entity/Nvidia_Hopper_architecture
GPTKB entity
Statements (33)
Predicate Object
gptkbp:instanceOf gptkb:graphics_card
gptkbp:announced March 2022
gptkbp:architecture_type monolithic
gptkbp:developer gptkb:Nvidia
gptkbp:first_product gptkb:Nvidia_H100
https://www.w3.org/2000/01/rdf-schema#label Nvidia Hopper architecture
gptkbp:memory_bandwidth up to 3 TB/s
gptkbp:namedAfter gptkb:Grace_Hopper
gptkbp:notableFeature gptkb:Transformer_Engine
gptkb:multi-instance_GPU_(MIG)
DPX instructions
confidential computing
dynamic programming acceleration
improved sparsity support
gptkbp:predecessor gptkb:Nvidia_Ampere_architecture
gptkbp:process_node gptkb:TSMC_4N
gptkbp:releaseYear 2022
gptkbp:successor gptkb:Nvidia_Blackwell_architecture
gptkbp:supports gptkb:HBM3_memory
gptkb:NVLink_4.0
gptkb:PCIe_Gen5
gptkb:Tensor_Cores
FP16 precision
FP8 precision
gptkbp:targetMarket data centers
high-performance computing
AI acceleration
gptkbp:transistor_count 80 billion (H100)
gptkbp:usedIn gptkb:Nvidia_GH200_Grace_Hopper_Superchip
gptkb:Nvidia_H100_PCIe
gptkb:Nvidia_H100_SXM
gptkbp:bfsParent gptkb:Nvidia_NVLink
gptkbp:bfsLayer 6