Check failed: status == CUDNN_STATUS_SUCCESS (4 vs. 0) CUDNN_STATUS_INTERNAL

Andreas ten Pas

unread,

Sep 4, 2015, 11:53:07 AM9/4/15

to Caffe Users

I'm getting this error since this morning: Check failed: status == CUDNN_STATUS_SUCCESS (4 vs. 0) CUDNN_STATUS_INTERNAL_ERROR

Here's my NVCC version info:

nvcc: NVIDIA (R) Cuda compiler driver

Built on Mon_Feb_16_22:59:02_CST_2015

Cuda compilation tools, release 7.0, V7.0.27

My graphics card is: NVIDIA Corporation GK106 [GeForce GTX 660].

Can Poyrazoğlu

unread,

Jan 5, 2016, 7:06:46 AM1/5/16

to Caffe Users

Hello,

Did you ever solve this?

Can Poyrazoğlu

unread,

Jan 5, 2016, 7:30:13 AM1/5/16

to Caffe Users

Nevermind, I was out of GPU memory. Closed some apps (e.g. Parallels) and it worked.

On Friday, September 4, 2015 at 6:53:07 PM UTC+3, Andreas ten Pas wrote:

Alfredo Nava

unread,

Apr 23, 2016, 6:12:01 PM4/23/16

to Caffe Users

I have this same error, but I am sure memory is not an issue because my card has 3 GB and the process only uses about 500 MB, plus my video card on idle is only using about 20% memory (600 MB)

Can Poyrazoğlu

unread,

Apr 24, 2016, 4:25:32 AM4/24/16

to Caffe Users

Maybe your OS is preallocating some VRAM for its own use? Or maybe you don't have contiguous video memory for that allocation.

강동현

unread,

Jul 17, 2017, 6:12:39 AM7/17/17

to Caffe Users

Hello,

I found out what the problem was.

If you've got that problem over the "cudnnCreate()", It is related to limit on the number of cudnn handles.

you should create fewer cudnn handles on your caffe code.

2015년 9월 5일 토요일 오전 12시 53분 7초 UTC+9, Andreas ten Pas 님의 말:

Reply all

Reply to author

Forward