"invalid device function" from forward-prop

soumith / cuda-convnet2.torch

Torch7 bindings for cuda-convnet2 kernels!

Apache License 2.0

40 stars 16 forks source link

"invalid device function" from forward-prop #12

Closed coventry closed 10 years ago

coventry commented 10 years ago

Can I use cuda-convnet2 with a GTX 650? The following snippet (extracted from benchmark.lua) fails with the error message

/tmp/luarocks_ccn2-scm-1-7061/cuda-convnet2.torch/cudaconv3/src/filter_acts.cu(2085) : getLastCudaError() CUDA error : filterActs: kernel execution failed : (8) invalid device function .

require 'ccn2'
n = ccn2.SpatialConvolution(64, 128, 9, 1):cuda()
i = torch.randn(64, 64, 64, 128):cuda()
n:forward(i)

soumith commented 10 years ago

you can, but right now i hard-coded the NVCC flags only for compute capability 3.5. Let me allow 3.0 as well.

soumith commented 10 years ago

done, if you reinstall the package, you should be good now.

coventry commented 10 years ago

Thanks. After reinstalling, the test script hangs my machine with the given inputs, but runs without errors if I halve all of them, which may well be an issue with the card (it has 2GB of memory.)

soumith commented 10 years ago

i think it's probably the memory, or that it's just really really slow for case 1 and case 2. gtx 650 isn't really the ideal card for the cases I wrote, i targeted the gtx titan :)