![\"\"](\"https:\/\/wp.study3.biz\/wp-content\/uploads\/2021\/01\/nvidia-smi-nvlink-c.jpg\")
\n[chibi@manjaro p2pBandwidthLatencyTest]$ .\/p2pBandwidthLatencyTest
\n[P2P (Peer-to-Peer) GPU Bandwidth Latency Test]
\nDevice: 0, GeForce RTX 2080 Ti, pciBusID: 81, pciDeviceID: 0, pciDomainID:0
\nDevice: 1, GeForce RTX 2080 Ti, pciBusID: 82, pciDeviceID: 0, pciDomainID:0
\nDevice=0 CAN Access Peer Device=1
\nDevice=1 CAN Access Peer Device=0<\/p>\n
***NOTE: In case a device doesn’t have P2P access to other one, it falls back to normal memcopy procedure.
\nSo you can see lesser Bandwidth (GB\/s) and unstable Latency (us) in those cases.<\/p>\n
P2P Connectivity Matrix CPU 0 1 CPU 0 1 NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled. [chibi@manjaro p2pBandwidthLatencyTest]$ .\/p2pBandwidthLatencyTest [P2P (Peer-to-Peer) GPU Bandwidth Latency T … \u7d9a\u304d\u3092\u8aad\u3080
\n D\\D 0 1
\n 0 1 1
\n 1 1 1
\nUnidirectional P2P=Disabled Bandwidth Matrix (GB\/s)
\n D\\D 0 1
\n 0 475.79 5.89
\n 1 5.85 542.91
\nUnidirectional P2P=Enabled Bandwidth (P2P Writes) Matrix (GB\/s)
\n D\\D 0 1
\n 0 527.62 47.10<\/strong>
\n 1 47.10<\/strong> 542.91
\nBidirectional P2P=Disabled Bandwidth Matrix (GB\/s)
\n D\\D 0 1
\n 0 532.76 8.25
\n 1 8.24 531.49
\nBidirectional P2P=Enabled Bandwidth Matrix (GB\/s)
\n D\\D 0 1
\n 0 527.22 94.15<\/strong>
\n 1 94.04<\/strong> 530.54
\nP2P=Disabled Latency Matrix (us)
\n GPU 0 1
\n 0 1.37 14.29
\n 1 12.45 1.33<\/p>\n
\n 0 3.33 10.16
\n 1 10.11 3.28
\nP2P=Enabled Latency (P2P Writes) Matrix (us)
\n GPU 0 1
\n 0 1.32 1.68
\n 1 1.81 1.32<\/p>\n
\n 0 3.34 2.68
\n 1 2.70 3.31<\/p>\n
\n[chibi@manjaro p2pBandwidthLatencyTest]$
\nAMD EPYC 7302P 16-Core Processor Manjaro Linux RTX2080Ti x2 CUDA 11.2 simpleP2P p2pBandwidthLatencyTest bandwidthTest deviceQuery<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"