CUDA Peer-to-Peer Memory Access
GPTKB entity
Statements (22)
Predicate | Object |
---|---|
gptkbp:instanceOf |
GPU technology
|
gptkbp:accessibleBy |
cudaDeviceEnablePeerAccess API
cudaMemcpyPeer API |
gptkbp:benefit |
deep learning
high-performance computing scientific simulations |
gptkbp:compatibleWith |
all GPU models
|
gptkbp:developedBy |
gptkb:NVIDIA
|
gptkbp:documentation |
NVIDIA CUDA Toolkit Documentation
|
gptkbp:enables |
direct memory access between GPUs
|
https://www.w3.org/2000/01/rdf-schema#label |
CUDA Peer-to-Peer Memory Access
|
gptkbp:improves |
data transfer speed between GPUs
|
gptkbp:introducedIn |
CUDA 4.0
|
gptkbp:reduces |
CPU involvement in GPU-to-GPU transfers
|
gptkbp:requires |
compatible hardware
CUDA-enabled GPUs same PCIe root complex for GPUs |
gptkbp:supportedBy |
gptkb:NVIDIA_NVLink
gptkb:PCI_Express |
gptkbp:usedIn |
multi-GPU systems
|
gptkbp:bfsParent |
gptkb:CUDA_C++
|
gptkbp:bfsLayer |
7
|