error == cudaSuccess (2 vs. 0) out of memory

[Ihsan Ullah]

Hi,
Can any body guide me how to tackle below error.
I am usign UBUNTU 12.04,
GPU=  GeForce GTX 660


I0525 18:43:13.658552 17949 solver.cpp:319]     Test net output #1: loss = 7.16204 (* 1 = 7.16204 loss)
F0525 18:43:13.986717 17949 syncedmem.cpp:51] Check failed: error == cudaSuccess (2 vs. 0)  out of memory
*** Check failure stack trace: ***
    @     0x7f0e74155378  google::LogMessage::Fail()
    @     0x7f0e741552c4  google::LogMessage::SendToLog()
    @     0x7f0e74154cb2  google::LogMessage::Flush()
    @     0x7f0e74157bb1  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f0e744c9e52  caffe::SyncedMemory::mutable_gpu_data()
    @     0x7f0e744efa41  caffe::Blob<>::mutable_gpu_data()
    @     0x7f0e745f4647  caffe::PoolingLayer<>::Forward_gpu()
    @     0x7f0e745bf312  caffe::Net<>::ForwardFromTo()
    @     0x7f0e745bf67f  caffe::Net<>::ForwardPrefilled()
    @     0x7f0e74504ce2  caffe::Solver<>::Step()
    @     0x7f0e7450575a  caffe::Solver<>::Solve()
    @           0x409c38  train()
    @           0x40653b  main
    @     0x7f0e7367276d  (unknown)
    @           0x4069c1  (unknown)
Aborted (core dumped)

[Axel Straminsky]

I had the same issue, and in my case I solved it reducing the batch size.