Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Caffe Window on 3 GPUs #3883

Open
smithdir101 opened this issue Mar 24, 2016 · 3 comments
Open

Caffe Window on 3 GPUs #3883

smithdir101 opened this issue Mar 24, 2016 · 3 comments
Labels

Comments

@smithdir101
Copy link

Hi All,

I am dowload the caffe window version from https://github.com/BVLC/caffe/tree/windows.

Curretenly, I have 3 GPUS on my dell machine all of them Titan X.

I run this command and I get these error:

Please see the attached image.

If you can help me it would be great.

Thanks,

Jhon

3gpu_errors

@pavlejosipovic
Copy link

Did you try to run UT (test_all) project to verify that everything is setup correctly?
Next you can try running sample training on a single GPU.

If both pass, you will have to start debugging this assertion you are getting..

@SvenTwo
Copy link

SvenTwo commented Apr 1, 2016

I also had problems with 3 GPUs; this pull request fixed it for me: #3586 (not sure if it's applicable to current master though)

@smithdir101
Copy link
Author

I solved this issue by activating p2p mode for each of 3 cards.
Use this command c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe -fdm 1 in order to activate TCC driver mode, not WDDM one. Restart is required! In case your display is connected to the GPU that is going to be in TCC mode, take into account it will operate in 800x600 mode only! It is a good idea to have additional "simple" GPU in order to connect system's display.

Next thing to consider. You cannot use more than 2 GPUs connected to one CPU unless you use tricks like 2GPU on one board (K80) or PCI bus switch on motherboard (there are vendors available). Therefore people often build 2 CPUs system in order to be able to use 4 GPUs. The catch here is that inter-CPU bus QPI still has very high latency relative to inter-PCI bus connections making work of 4 GPUs way less effective comparing to 4GPU's that are connected to the same one CPU.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

3 participants