Why prebuild binaries have train_net.exe, but originally caffe run as caffe train -solver examples/mnist/lenet_solver.prototxt why this build don't use original way?
Also why build use only 1 core? which of BLAS libs linked to it?
Just a benchmark report for MNIST:
CPU: Core 2 Duo E8500 - 2.5 hours
GPU: GeForce GTX 460 - 10 min
Platform: Windows 10
CUDA toolkit: cuda_7.5.18_win10_network.exe
Why prebuild binaries have
train_net.exe
, but originally caffe run ascaffe train -solver examples/mnist/lenet_solver.prototxt
why this build don't use original way?Also why build use only 1 core? which of BLAS libs linked to it?
Just a benchmark report for MNIST:
CPU: Core 2 Duo E8500 - 2.5 hours GPU: GeForce GTX 460 - 10 min
Platform: Windows 10 CUDA toolkit: cuda_7.5.18_win10_network.exe