|
gptkbp:instanceOf
|
gptkb:personal_computer
|
|
gptkbp:compatibleWith
|
gptkb:Nvidia_RTX_40_series
standard PCIe SLI bridges
|
|
gptkbp:competitor
|
gptkb:AMD_Infinity_Fabric
|
|
gptkbp:connectorType
|
proprietary edge connector
|
|
gptkbp:enables
|
faster data transfer between GPUs
unified memory access
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
Nvidia NVLink
|
|
gptkbp:introducedIn
|
gptkb:Nvidia_Pascal_architecture
2016
|
|
gptkbp:manufacturer
|
gptkb:Nvidia
|
|
gptkbp:maxGPUsConnected
|
up to 16 (in some systems)
|
|
gptkbp:memoryBusWidth
|
up to 25 GB/s per lane (NVLink 2.0)
up to 50 GB/s per lane (NVLink 3.0)
|
|
gptkbp:purpose
|
high-speed GPU interconnect
|
|
gptkbp:replacedBy
|
PCI Express (for GPU-to-GPU communication)
|
|
gptkbp:supports
|
multi-GPU communication
|
|
gptkbp:supportsProtocol
|
point-to-point interconnect
|
|
gptkbp:usedIn
|
gptkb:Nvidia_A100_GPU
gptkb:Nvidia_Ampere_architecture
gptkb:Nvidia_DGX_systems
gptkb:Nvidia_H100_GPU
gptkb:Nvidia_Hopper_architecture
gptkb:Nvidia_P100_GPU
gptkb:Nvidia_Quadro_GPUs
gptkb:Nvidia_RTX_3080_Ti
gptkb:Nvidia_RTX_3090
gptkb:Nvidia_Tesla_GPUs
gptkb:Nvidia_Turing_architecture
gptkb:Nvidia_V100_GPU
gptkb:Nvidia_Volta_architecture
data centers
high-performance computing
supercomputers
AI workloads
Nvidia RTX GPUs (select models)
|
|
gptkbp:website
|
https://www.nvidia.com/en-us/data-center/nvlink/
|
|
gptkbp:bfsParent
|
gptkb:Nvidia
|
|
gptkbp:bfsLayer
|
5
|