使用ソフトとはCUDA6.0のサンプルファイル「bandwidthTest」をVS2012でビルド。
■環境
CPU:3930k
GPU:Geforce TITAN
OS:Windows7 64bit
■前回データ(CUDA5.0)
Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5815.1
Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 6282.0
Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 220126.1
■今回実施結果(CUDA6.0)n=3
Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5924.5
33554432 6016.3
33554432 6015.6
Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 6361.2
33554432 6399.1
33554432 6408.2
Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 224998.1
33554432 224667.3
33554432 208915.6
若干高速化されてる?
割合にして3%ほど速くなったみたいだ。
ただ、PCIe Gen.2での話なのでGen.3だとどう変わるのか気になるところです。