# error when running tensorflow-gpu (CuDNN incompatible with CUDA)

**URL:** <https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380>\
**Category:** CUDA Setup and Installation\
**Created:** [April 18, 2018, 3:46am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380 "2018-04-18T03:46:20Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![hurwitz.david](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@hurwitz.david](https://forums.developer.nvidia.com/u/hurwitz.david)\
**Post date:** [April 18, 2018, 3:46am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380/1 "2018-04-18T03:46:20Z")

</div>

Can someone translate this for me:

> Loaded runtime CuDNN library: 7103 (compatibility version 7100) but source was compiled with 7003 (compatibility version 7000). If using a binary install, upgrade your CuDNN library to match…

I get this error when I try running tensorflow-gpu. I have installed CUDA 9.0, and CuDNN 7.0. I gather the message is telling me I need to install and point to a different CuDNN to be compatible with the CUDA version I have installed, but specifically, which one?

I am running on a PC that has a “GeForce 940MX major: 5 minor: 0” GPU.

Thanks.

Dave

---

<div class="post-metadata">

**Author:** ![Robert\_Crovella](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/robert_crovella/32/14043_2.png) [@Robert\_Crovella](https://forums.developer.nvidia.com/u/Robert_Crovella)\
**Post date:** [April 18, 2018, 4:07am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380/2 "2018-04-18T04:07:05Z")

</div>

source was compiled with 7003, compatibility version 7000

cudnn 7.0.x

The fact that it found 7103 means you have 7.1 loaded (somewhere)

[url][python - What does the error: `Loaded runtime CuDNN library: 5005 but source was compiled with 5103` mean? - Stack Overflow](https://stackoverflow.com/questions/41005249/what-does-the-error-loaded-runtime-cudnn-library-5005-but-source-was-compiled/41008150#41008150%5B/url%5D)

---

<div class="post-metadata">

**Author:** ![hurwitz.david](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@hurwitz.david](https://forums.developer.nvidia.com/u/hurwitz.david)\
**Post date:** [April 18, 2018, 4:14am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380/3 "2018-04-18T04:14:26Z")

</div>

Thanks!

Separately, I figured it out too. Which, for me is quite amazing. Anyway, yes, you are right. When I downloaded and pointed to cudnn 7.0.5 instead of the default one (7.1.3) it worked. Yippee.

Just as an aside, I see my laptop is running tensorflow about 10x faster with the gpu than without. Nice. I know this is a low-power GPU, but still, nice to get that bump.

Dave

---

<div class="post-metadata">

**Author:** ![parkdonghun90](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@parkdonghun90](https://forums.developer.nvidia.com/u/parkdonghun90)\
**Post date:** [July 31, 2018, 4:27am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380/4 "2018-07-31T04:27:06Z")

</div>

@hurwitz.david

How do you point 7.0.5 instead of 7.1.0.\*

I had a same problem~!

Thanks ahead.

---

<div class="post-metadata">

**Author:** ![hurwitz.david](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@hurwitz.david](https://forums.developer.nvidia.com/u/hurwitz.david)\
**Post date:** [July 31, 2018, 4:52am UTC](https://forums.developer.nvidia.com/t/error-when-running-tensorflow-gpu-cudnn-incompatible-with-cuda/60380/5 "2018-07-31T04:52:20Z")

</div>

As best I recall, the CuDNN driver has to be exactly the one that matches the version of CUDA you’re using. When I look in the CuDNN archive, I see “Download cuDNN v7.0.5 (Dec 5, 2017), for CUDA 9.0” which is the one I needed. Here’s the page that has the archive of drivers:

[https://developer.nvidia.com/rdp/cudnn-archive](https://developer.nvidia.com/rdp/cudnn-archive)

Good luck. I found this part a bit tricky.