| 1 |
On Mon, Sep 17, 2018 at 04:00:26PM +0200, Guilherme Amadio wrote: |
| 2 |
> Hi everyone, |
| 3 |
> |
| 4 |
> We have several packages (~35) with local USE=cuda. Should we make that |
| 5 |
> a global USE flag? It's a quite generic flag for GPU support, so I was |
| 6 |
> surprised to learn it was still local when I added support for it to a |
| 7 |
> package recently. |
| 8 |
|
| 9 |
Sounds good to me, I was also surprised when I added the cuda |
| 10 |
useflag to TensorFlow. |
| 11 |
|
| 12 |
> Another thing we might want to discuss is a global setting for the CUDA |
| 13 |
> architecture to be used¹. We compile from source, so it would be a shame |
| 14 |
> not to compile with the right features for newer GPUs². Maybe just |
| 15 |
> advertising the NVCCFLAGS one should put in make.conf is enough? |
| 16 |
|
| 17 |
Is this the CUDA Capability (ie https://developer.nvidia.com/cuda-gpus)? |
| 18 |
TensorFlow's configure currently autodetects it based on what the card |
| 19 |
supports and builds for that version (or versions). Or you can set |
| 20 |
TF_CUDA_COMPUTE_CAPABILITIES="6.1" in make.conf. If you would like to |
| 21 |
standardize this I'll gladly change to use that. |
| 22 |
|
| 23 |
I'm also updating cuda.eclass for EAPI7, I will send patches to the list |
| 24 |
soon. I'm also adding a couple other functions for things I use in TF, |
| 25 |
If there are more things you want to add, they should all go in at once. |
| 26 |
|
| 27 |
-- Jason |
| 28 |
|
| 29 |
> Cheers, |
| 30 |
> —Guilherme |
| 31 |
> |
| 32 |
> 1. https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#options-for-steering-gpu-code-generation |
| 33 |
> 2. https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#gpu-feature-list |
| 34 |
> |
Archives