<div dir="auto">Hi,<div dir="auto"><br></div><div dir="auto">So I actually have been using the CUDA from Manzil.</div><div dir="auto"><br></div><div dir="auto">While it did help in resolving other issues, recent issue after the file system shutdown seems to happen with pytorch only. (Not on tensorflow)</div><div dir="auto">Using Manzil's CUDA (old one and again got a copy for another one after the probldm, just in case) did not resolve the problem.</div><div dir="auto"><br></div><div dir="auto">The problem was only resolved after I went ahead and installed python, pip locally. </div><div dir="auto"><br></div><div dir="auto">With this experience, I am suspecting that the currently provided conda has some problem. (Altough the error messages indicate only CUDA errors) Or maybe it was just a hack around, but this did fix the issue.</div><div dir="auto"><br></div><div dir="auto">Cheers!</div><div dir="auto">Jay-Yoon</div></div><br><div class="gmail_quote"><div dir="ltr">On Thu, Sep 6, 2018, 11:29 AM Barnabas Poczos <<a href="mailto:bapoczos@cs.cmu.edu">bapoczos@cs.cmu.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi All,<br>
<br>
I'm somewhat confused:<br>
<br>
* Do I understand correctly that Manzil actually is using the CUDA<br>
libraries installed by himself<br>
(/zfsauton/home/manzilz/local/cuda-9.0/) and not the system libraries<br>
(/usr/local/cuda/lib64) ?<br>
* Since he is using different CUDA libraries is that the reason that<br>
pytorch is working for him and not for the other users? If so, should<br>
we double check the system libraries?<br>
* Do we know anyone who can use pytorch now with the CUDA system<br>
libraries ? If so, those users please let us know your system env<br>
variables.<br>
* As a quick solution, should we ask Manzil to copy his cuda libraries<br>
to a public place where others could access them?<br>
<br>
Best,<br>
Barnabas<br>
<br>
======================<br>
Barnabas Poczos, PhD<br>
Associate Professor<br>
Co-Director of PhD Program<br>
Machine Learning Department<br>
Carnegie Mellon University<br>
On Wed, Sep 5, 2018 at 5:33 PM Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>> wrote:<br>
><br>
> Here is my related env variables:<br>
><br>
><br>
><br>
> CUDA_HOME=/zfsauton/home/manzilz/local/cuda-9.0/<br>
><br>
> LD_LIBRARY_PATH=/zfsauton/home/manzilz/local/lib64:/zfsauton/home/manzilz/local/lib:/zfsauton/home/manzilz/local/cuda-9.0/lib64:/usr/local/cuda/lib64:<br>
><br>
> PATH=/zfsauton/home/manzilz/local/bin:/zfsauton/home/manzilz/.local/bin:/zfsauton/home/manzilz/local/cuda-9.0/bin:/usr/local/cuda/bin:/usr/lib64/qt-3.3/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin<br>
><br>
> C_INCLUDE_PATH=/zfsauton/home/manzilz/local/include:<br>
><br>
><br>
><br>
> From: Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>><br>
> Sent: Wednesday, September 05, 2018 5:29 PM<br>
> To: Yichong Xu <<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a>><br>
> Cc: Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>>; <a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>; Predrag Punosevac <<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a>>; Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>>; <a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a><br>
> Subject: Re: PyTorch problem<br>
><br>
><br>
><br>
> If the CUDA examples work for anyone, can they share their PATH and LD_LIBRARY_PATH variables?<br>
><br>
><br>
><br>
> Thanks<br>
><br>
><br>
><br>
> On Wed, Sep 5, 2018 at 5:27 PM Yichong Xu <<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a>> wrote:<br>
><br>
> I think with Biswajit’s and my problem with cuda, we should isolate the problem with just CUDA (and drivers) instead of wandering around python or pytorch.<br>
><br>
> Predrag can you test the CUDA examples? I sort of agree with Manzil that this might be a user account problem.<br>
><br>
><br>
><br>
> Thanks,<br>
><br>
> Yichong<br>
><br>
><br>
><br>
><br>
><br>
><br>
><br>
> On Sep 5, 2018, at 5:14 PM, Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>> wrote:<br>
><br>
><br>
><br>
> I just tried Yichong's way of testing cuBLAS, and get the same error as earlier:<br>
><br>
><br>
><br>
> [Matrix Multiply CUBLAS] - Starting...<br>
><br>
> GPU Device 0: "TITAN Xp" with compute capability 6.1<br>
><br>
><br>
><br>
> MatrixA(640,480), MatrixB(480,320), MatrixC(640,320)<br>
><br>
> CUDA error at matrixMulCUBLAS.cpp:275 code=1(CUBLAS_STATUS_NOT_INITIALIZED) "cublasCreate(&handle)"<br>
><br>
><br>
><br>
> So I believe it is not a conda error. I also tried removing .nv, doesn't help either. Maybe someone can share the PATH env variable?<br>
><br>
><br>
><br>
> On Wed, Sep 5, 2018 at 5:08 PM Emre Yolcu <<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>> wrote:<br>
><br>
> Manzil, could you share your `conda env export` (or equivalent) output for the environment you use for pytorch? It's still not working for me after reboot, maybe I can try replicating your exact setup and try with that.<br>
><br>
><br>
><br>
> Thanks,<br>
><br>
><br>
><br>
> Emre<br>
><br>
><br>
><br>
> On Wed, Sep 5, 2018 at 4:56 PM, Predrag Punosevac <<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a>> wrote:<br>
><br>
> Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>> wrote:<br>
><br>
> > It was working me before reboot as well. PyTorch does work on all<br>
> > nodes for me.<br>
><br>
> Aha! Gotcha.<br>
><br>
> ><br>
> > I am trying to say is that i think it is not issue at system level but<br>
> > at user account level. I might be wrong though.<br>
><br>
> That was my hunch as well. They were trying to convince me in a 150<br>
> e-mails chain over the weekend that pytorch was broken when I replaced a<br>
> failed HDD on the main file server. That didn't make any sense.<br>
><br>
> Could you please share your binaries and setup with outher pytorch<br>
> users?<br>
><br>
> Cheers,<br>
> Predrag<br>
><br>
> ><br>
> ><br>
> > -------- Original message --------<br>
> > From: Predrag Punosevac <<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a>><br>
> > Date: 9/5/18 4:44 PM (GMT-05:00)<br>
> > To: Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>><br>
> > Cc: Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>>, Yichong Xu <<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a>>, Emre Yolcu <<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>>, <a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a><br>
> > Subject: Re: PyTorch problem<br>
> ><br>
> > Should I go ahead and reboot all GPU computing nodes? Can somebody else confirm that a reboot fixes the issue?<br>
> ><br>
> > Predrag<br>
> ><br>
> > On Wed, Sep 5, 2018 at 4:42 PM, Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a><mailto:<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>>> wrote:<br>
> > It does work for me and my friends<br>
> ><br>
> ><br>
> ><br>
> ><br>
> > -------- Original message --------<br>
> > From: Predrag Punosevac <<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a><mailto:<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a>>><br>
> > Date: 9/5/18 4:40 PM (GMT-05:00)<br>
> > To: Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a><mailto:<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>>><br>
> > Cc: Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a><mailto:<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>>>, Yichong Xu <<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a><mailto:<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a>>>, Emre Yolcu <<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a><mailto:<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>>>, <a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a><mailto:<a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a>><br>
> > Subject: Re: PyTorch problem<br>
> ><br>
> > I just rebooted GPU8. All packages are up to date. NVidia driver appears to be working properly and I can do GPU computations from MATLAB. Let's try now to get pytorch working on GPU8.<br>
> ><br>
> > Predrag<br>
> ><br>
> > On Wed, Sep 5, 2018 at 12:19 AM, Biswajit Paria <<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a><mailto:<a href="mailto:bparia@cs.cmu.edu" target="_blank" rel="noreferrer">bparia@cs.cmu.edu</a>>> wrote:<br>
> > I am facing a similar error on all GPU machines. Did someone find a solution yet?<br>
> ><br>
> ><br>
> > 2018-09-05 00:27:41.546064: E tensorflow/stream_executor/cuda/cuda_blas.cc:459] failed to create cublas handle: CUBLAS_STATUS_NOT_INITIALIZED<br>
> ><br>
> > On Tue, Sep 4, 2018 at 10:03 PM Manzil Zaheer <<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a><mailto:<a href="mailto:manzil@cmu.edu" target="_blank" rel="noreferrer">manzil@cmu.edu</a>>> wrote:<br>
> > Hi Yichong<br>
> ><br>
> > Yes I am able to run TF and PyTorch on these machines. Recently someone else also had similar issue, but it got fixed by reinstalling some local packages.<br>
> ><br>
> > Thanks,<br>
> > Manzil<br>
> ><br>
> ><br>
> > -------- Original message --------<br>
> > From: Yichong Xu <<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a><mailto:<a href="mailto:yichongx@cs.cmu.edu" target="_blank" rel="noreferrer">yichongx@cs.cmu.edu</a>>><br>
> > Date: 9/4/18 9:58 PM (GMT-05:00)<br>
> > To: Emre Yolcu <<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a><mailto:<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>>>, Predrag Punosevac <<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a><mailto:<a href="mailto:predragp@andrew.cmu.edu" target="_blank" rel="noreferrer">predragp@andrew.cmu.edu</a>>><br>
> > Cc: <a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a><mailto:<a href="mailto:users@autonlab.org" target="_blank" rel="noreferrer">users@autonlab.org</a>><br>
> > Subject: Re: PyTorch problem<br>
> ><br>
> > Just wondering - can Tensorflow run well on these machines? I hope someone to confirm about this so that we can isolate the problem.<br>
> > OK so here?s a further test: I tried running the cuda examples from the cuda installation (in /usr/local/cuda/sample), on gpu2 in my scratch directory. Simple jobs like deviceQuery succeeds, but simpleCUBLAS failed:<br>
> > yichongx@gpu2$ cd /home/scratch/yichongx/<br>
> > yichongx@gpu2$ cd<br>
> > 0_Simple/ 2_Graphics/ 4_Finance/ 6_Advanced/ bin/ conda/<br>
> > 1_Utilities/ 3_Imaging/ 5_Simulations/ 7_CUDALibraries/ common/ miniconda3/<br>
> > yichongx@gpu2$ cd 7_CUDALibraries/<br>
> > yichongx@gpu2$ cd simpleCUBLAS<br>
> > yichongx@gpu2$ CUDA_VISIBLE_DEVICES=3 ./simpleCUBLAS<br>
> > GPU Device 0: "TITAN X (Pascal)" with compute capability 6.1<br>
> ><br>
> > simpleCUBLAS test running..<br>
> > !!!! CUBLAS initialization error<br>
> > yichongx@gpu2$<br>
> ><br>
> ><br>
> > This is also consistent with our previous errors from pytorch, which say cublas library not initialized.<br>
> ><br>
> > So this means at least there is some problem with CUBLAS on gpu2. This post suggests that using sudo can resolve this problem, and this is probably because of some permission problems on CUBLAS libraries:<br>
> > <a href="https://devtalk.nvidia.com/default/topic/1027602/cuda-setup-and-installation/cublas-libraries-with-incorrect-permissions/" rel="noreferrer noreferrer" target="_blank">https://devtalk.nvidia.com/default/topic/1027602/cuda-setup-and-installation/cublas-libraries-with-incorrect-permissions/</a><br>
> > @Predrag: Can you try running the simpleCUBLAS example from the CUDA library, with and without root privilege? I think that might be something that you are more familiar with. Thank you very much!<br>
> ><br>
> ><br>
> > Thanks,<br>
> > Yichong<br>
> ><br>
><br>
> > On Sep 4, 2018, at 3:18 PM, Emre Yolcu <<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a><mailto:<a href="mailto:eyolcu@cs.cmu.edu" target="_blank" rel="noreferrer">eyolcu@cs.cmu.edu</a>>> wrote:<br>
> ><br>
> > Hi,<br>
> ><br>
> > We are trying to troubleshoot the PyTorch issue with Predrag and were wondering:<br>
> ><br>
> > Is anybody able to run PyTorch GPU models on gpu1-9? If you can, we would appreciate if you can respond.<br>
> ><br>
> > Also, is it a problem for anyone if gpu8 is rebooted today?<br>
> ><br>
> > Thanks,<br>
> ><br>
> > Emre<br>
> ><br>
> ><br>
> ><br>
> > --<br>
> > Biswajit Paria<br>
> > PhD in ML @ CMU<br>
> ><br>
> ><br>
><br>
><br>
><br>
><br>
><br>
><br>
> --<br>
><br>
> Biswajit Paria<br>
><br>
> PhD in ML @ CMU<br>
><br>
><br>
><br>
><br>
><br>
><br>
> --<br>
><br>
> Biswajit Paria<br>
><br>
> PhD in ML @ CMU<br>
<br>
</blockquote></div>