Darknet: cuDNN Error: CUDNN_STATUS_BAD_PARAM: No error
when I run "darknet detector test data/coco.data yolov3.cfg yolov3.weight dog.jpg" in cmd,it's show that
cuDNN status Error in: file: ....\src\convolutional_layer.c : cudnn_convolutional_setup() : line: 293 : build time: Feb 20 2019 - 20:58:09
cuDNN Error: CUDNN_STATUS_BAD_PARAM
cuDNN Error: CUDNN_STATUS_BAD_PARAM: No error
but "darknet_no_gpu detector test data/coco.data yolov3.cfg yolov3.weight dog.jpg",everything is gonna be fine. Why? MY GPU is GT 635M ,compute_capability =2.1,Is its performance too low?
z19940829i
All 13 comments
cuDNN supports only CC >= 3.0
So try to build Darknet without cuDNN.
on Linux: CUDNN=0 in the Makefile
on Windows: If you want to build without CUDNN then: open \darknet.sln -> (right click on project) -> properties -> C/C++ -> Preprocessor -> Preprocessor Definitions, and remove this: CUDNN;
AlexeyAB
on 20 Feb 2019
Thank you so much!
z19940829i
on 21 Feb 2019
Hi @AlexeyAB ,
My GPU is GT 730 (which CC is 3.5) but I am facing the same error as @z19940829i . What can I do?
Thank you very much.
PD: Morever, I have tried to build Darknet without cuDNN but when I enter the image path it says: "CUDA Error: no kernel image is available for execution on the device"
martisaju
on 3 Jun 2019
@martisaju Do you use make or cmake for compiling? Did you compile Darknet for CC 3.5?
AlexeyAB
on 3 Jun 2019
@AlexeyAB I have compiled Darknet following the steps on 'How to compile on Windows (legacy way)' but using CUDA 9.1 instead of 10.0 and cuDNN 7.1. I have used VS 2017. How can I compile Darknet for CC 3.5?
Thank you
martisaju
on 3 Jun 2019
Open in MSVS \darknet.sln -> (right click on project) -> properties -> CUDA C/C++ -> Device
and set there compute_35,sm_35 instead of compute_30,sm_30;compute_75,sm_75
AlexeyAB
on 3 Jun 2019
@AlexeyAB I have just rebuilt it considering this change and I get the same error...
martisaju
on 3 Jun 2019
Hi@AlexeyAB , My GPU is p100, compute_60, CUDA-9.0, cudnn7.0.5(also tried cudnn7.1.4). However i met the same error "cuDNN Error: CUDNN_STATUS_BAD_PARAM".
"0,1,2,3
yolov3-voc
layer filters size/strd(dil) input output
0
cuDNN status Error in: file: ./src/convolutional_layer.c : () : line: 301 : build time: Jul 30 2019 - 16:13:30
cuDNN Error: CUDNN_STATUS_BAD_PARAM
cuDNN Error: CUDNN_STATUS_BAD_PARAM: File exists
darknet: ./src/utils.c:293: error: Assertion `0' failed.
train.sh: line 1: 32480 Aborted ./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74 -gpus 0,1,2,3
"
when i use the same cuda and cudnn, pjreddie/darknet runs well.
Thank you!
phexic
on 30 Jul 2019
cuDNN supports only CC >= 3.0
So try to build Darknet without cuDNN.
on Linux:
CUDNN=0in the Makefileon Windows: If you want to build without CUDNN then: open \darknet.sln -> (right click on project) -> properties -> C/C++ -> Preprocessor -> Preprocessor Definitions, and remove this:
CUDNN;
Alexy What are the demerits you get by removing CUDNN? Thanks
aql315
on 12 Nov 2019
@aql315 Large models works slower without cuDNN.
AlexeyAB
on 12 Nov 2019
@aql315 Large models works slower without cuDNN.
@AlexeyAB thanks for the reply!
But does not using cuDNN effect the accuracy? just only the speed?
aql315
on 13 Nov 2019
@aql315 only speed.
AlexeyAB
on 13 Nov 2019
@aql315 only speed.
Thanks @AlexeyAB
by any chance do you Know how many times it would be slower?
aql315
on 13 Nov 2019
Related issues
Yumin-Sun-00
路
3Comments
bit-scientist
路
3Comments
jasleen137
路
3Comments
Greta-A
路
3Comments
zihaozhang9
路
3Comments
Most helpful comment
cuDNN supports only CC >= 3.0
So try to build Darknet without cuDNN.
on Linux:
CUDNN=0in the Makefileon Windows: If you want to build without CUDNN then: open \darknet.sln -> (right click on project) -> properties -> C/C++ -> Preprocessor -> Preprocessor Definitions, and remove this:
CUDNN;