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
Requested GPU:0, but only XLA_GPU:0 exsits, tf-gpu1.14.0 #30748
Comments
@AmitMY You could use
Please let me know how it progresses. Thanks! |
Thanks @jvishnuvardhan
Although I only create the computation graph under I also see:
(To make sure I didn't change anything except for this config, when I remove the config I am getting that error again, still seeing my GPUs as XLA_GPU) |
@sanjoy would you please help to take a look? |
Can you please attach the full log? XLA creates an XLA_GPU device for every present on the system whereas TF creates a GPU device only for GPUs suitable for compute (i.e. ignores slower GPUs) so that could explain what you're seeing. TF logs out "Ignoring visible gpu device" when it does this enhanced filtering so you should see it in your logs. |
I wouldn't say 1080Ti is slow... I can't attach the full log including Titan X log output:
|
I got the same problem too. |
Actually, works fine on the same machine hardware with CUDA Version 10.0.130 My bad, I checked CUDA version on the wrong server when reporting the issue. Is this version of Tensorflow even supposed to work for v8? if not, I'll close this issue. |
The failure is almost certainly because of the cuda version, going by the log:
The build probably bakes in |
mine combination is tf 1.13 and cuda 10.. even cuda 9 is not working.
…On Fri, Jul 19, 2019 at 10:44 AM Amit Moryossef ***@***.***> wrote:
Actually, works fine on the same machine hardware with CUDA Version
10.0.130
It doesn't work on CUDA Version 8.0.61
My bad, I checked CUDA version on the wrong server when reporting the
issue.
Is this version of Tensorflow even supposed to work for v8? if not, I'll
close this issue.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#30748?email_source=notifications&email_token=AJKLC6GLBGBDXX4DO27SP4LQAH4PNA5CNFSM4ID7TGW2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2MIYOA#issuecomment-513313848>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AJKLC6D4A7UN7WRQCI7ZAGTQAH4PNANCNFSM4ID7TGWQ>
.
|
Same problem here , not detecting any GPU device after upgrading the tensorflow version |
I downgraded the tensorflow to an older version and the problem is
resolved..
command is like
conda install tensorflow-gpu==1.9.0
…On Wed, Jul 24, 2019 at 2:20 PM alaayadi ***@***.***> wrote:
Same problem here , not detecting any GPU device after upgrading the
tensorflow version
tensorflow only detect GPU XLA
But now , I cannot go back to tensorflow1.10.0 as it was been deleted from
pip
Any help ?
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#30748?email_source=notifications&email_token=AJKLC6C56BDJDTRSDWU762LQBDBQFA5CNFSM4ID7TGW2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD2XUDJA#issuecomment-514802084>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AJKLC6H37AS6XXGYE6MMM63QBDBQFANCNFSM4ID7TGWQ>
.
|
That's just working ! thanks |
If you have cuda 10.1 you need to update cuDNN to 7.6.2 That's what got it working for me without downgrading Tensorflow |
For me, it was the placement of ArgMax ops on GPUs that broke it; relocating them to CPU did the trick. |
I'm having the same issue after accidentally upgrading CUDA and then downgrading again, how did you manage to relocate the ArgMax ops to CPU? |
@albertNod something along these lines: https://github.com/aymericdamien/TensorFlow-Examples/blob/042c25ce2c3d91bf5a2e0a308fea578b1e290f82/examples/6_MultiGPU/multigpu_cnn.py#L110 |
I met the same problem on ubuntu 18.04, cuda 10.1 and Tensorflow 1.14.0. However, I uninstalled the pip version tensorflow using |
works great! |
Thank you! Saved my day. I'm on ubuntu 16.04, cuda 10.1. And this works: |
By installing CUDA 10.0, this problem may be solved.
|
I have similar issue with Tensorflow and Keras. And, don't have any issue with pytorch:
Notice that I got feedback Since I got libcudart.so.10.2 installed, I solved this issue by: make link Update: got similar issue again in jupyter, need to close, then start jupyter again, but doesn't solve CUDNN_STATUS_INTERNAL_ERROR See more in this repository |
Doesn't matter CUDA 10.0 , 10.1 10.2, for all dll not found, manually load them into ram. I have done all the setting PATH environment variable and nothing worked for me except this. I usually use wheels built by others because official tensorflow software builds lag the hardware (tf2.2 is only tested up to cudnn 7.4 from https://www.tensorflow.org/install/source_windows). Last year June when i was trying to install tf1.4, the official tf only went up to 10.0 but my GPU required 10.1 EDIT |
System information
pip
Describe the problem
tensorflow-gpu
latest version by performing:pip install --upgrade --force-reinstall tensorflow-gpu
tf.device('/gpu:0')
The error seems to say I have 4 gpus, but they don't match GPU, only XLA_GPU. I have no idea why, earlier versions of tensorflow do say I have GPU, but claim other bugs.
Error:
Even if I try:
tf.device('/job:localhost/replica:0/task:0/device:XLA_GPU:0')
I get:
The text was updated successfully, but these errors were encountered: