I have a tensorflow model with multiple inputs and several layers, and a final softmax layer. The model is trained in Python (using the Keras framework), then saved and inference is done using a C++ program that facilitates a CMake build of TensorFlow (following basically those instructions: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/cmake).
In python (tensorflow-gpu) all ops use the GPU (using log_device_placement):
out/MatMul: (MatMul): /job:localhost/replica:0/task:0/gpu:0
2017-12-04 14:07:38.005837: I C:\tf_jenkins\home\workspace\rel-in\M\windows-gpu\PY\35\tensorflow\core\common_runtime\simple_placer.cc:872] out/MatMul: (MatMul)/job:localhost/replica:0/task:0/gpu:0
out/BiasAdd: (BiasAdd): /job:localhost/replica:0/task:0/gpu:0
2017-12-04 14:07:38.006201: I C:\tf_jenkins\home\workspace\rel-win\M\windows-gpu\PY\35\tensorflow\core\common_runtime\simple_placer.cc:872]
out/BiasAdd: (BiasAdd)/job:localhost/replica:0/task:0/gpu:0
out/Softmax: (Softmax): /job:localhost/replica:0/task:0/gpu:0
2017-12-04 14:07:38.006535: I C:\tf_jenkins\home\workspace\rel-win\M\windows-gpu\PY\35\tensorflow\core\common_runtime\simple_placer.cc:872] out/Softmax: (Softmax)/job:localhost/replica:0/task:0/gpu:0
To save the graph, the freeze_graph script is used (the script producing the log above loads again the freezed graph in .pb format).
When I use the C++ program and load the freezed graph (following closely the LoadGraph() function in https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/label_image/main.cc - ReadBinaryProto() and session->Create()), and log again the device placements, I find that the Softmax is placed on CPU (all others ops are on GPU):
dense_6/MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
dense_6/BiasAdd: (BiasAdd): /job:localhost/replica:0/task:0/device:GPU:0
dense_6/Relu: (Relu): /job:localhost/replica:0/task:0/device:GPU:0
out/MatMul: (MatMul): /job:localhost/replica:0/task:0/device:GPU:0
out/BiasAdd: (BiasAdd): /job:localhost/replica:0/task:0/device:GPU:0
out/Softmax: (Softmax): /job:localhost/replica:0/task:0/device:CPU:0
This placement is also confirmed by high CPU/low GPU utilization, and also apparent from profiling the application. The data type of the out layer is float32 (out/Softmax -> (<tf.Tensor 'out/Softmax:0' shape=(?, 1418) dtype=float32>,)).
Further investigation revealed:
Creating the softmax-op in C++ and placing it on GPU explicitly throws this error message:
Cannot assign a device for operation 'tsoftmax': Could not satisfy explicit device specification '/device:GPU:0' because no supported kernel for GPU devices is available.
A call to tensorflow::LogAllRegisteredKernels() showed also that Softmax is only available for CPU!
The build directory contains many files related to "softmax" (e.g. `tf_core_gpu_kernels_generated_softmax_op_gpu.cu.cc.obj.Release.cmake). Don't know how to check every compilation step, though.
when I look into the "tf_core_gpu_kernels.lib" (one can open a .lib with 7Z ;)), there are files like "tf_core_gpu_kernels_generated_softmax_op_gpu.cu.cc.lib" - so I believe there is nothing wrong with compiling the kernels itself
But: inspecting the "tensorflow.dll" (Dependency Walker) shows that only CPU kernels for Softmax are included (there are functions like const tensorflow::SoftmaxOp<struct Eigen::ThreadPoolDevice,double>, but no functions with GPU such as const tensorflow::SoftplusGradOp<struct Eigen::GpuDevice,float>).
Setup: Tensorflow 1.3.0, Windows 10, GPU: NVidia GTX 1070 (8GB RAM, memory utilization also very low).
I found a workaround - the workaround is to include the tf_core_gpu_kernels.lib in some of the steps (create_def_file.py). More details here: GitHub Issue 15254
Related
__global__ void functionA()
{
printf("functionA");
}
int main()
{
printf("main1");
functionA<<<1,1>>>();
printf("main2");
}
I'm trying to run a simple test with the above. But the program only outputs "main1". The program should output "functionA" and "main2" too.
This seems to have two reasons:
First of all you need to add
cudaDeviceSynchronize();
after the CUDA routine in order to block the main until the device has completed all tasks.
Furthermore this might happen if you set the wrong GPU architecture/compute capability XX when compiling the code
$ nvcc -gencode=arch=compute_XX,code=sm_XX -o my_app my_app.cu
In this case only the host code is run while the parts on the accelerator will be omitted it seems. You can find an overview of the corresponding number XX for the different hardware generations over here. The K20m you are running is 35. So it should be
$ nvcc -gencode=arch=compute_35,code=sm_35 -o my_app my_app.cu
in your case.
This might also occur if you have multiple graphic accelerators in your system and the code is executed on the wrong one. Each graphics card/accelerator is assigned a particular device id. The device with number 0 should be assigned automatically to the most powerful device and will be used by default. Therefore the first time I compiled the code on my system containing a powerful Tesla K80 (architecture 37) and a low power Quadro P620 (architecture 60) I selected 37 and had the same error as you have while when selecting 60 the code would run. I then used then the Querying Device Properties example to give me a list of the CUDA-capable devices and their corresponding device id, just to find out that on my system the Tesla K80 is set as 1 and 2 while the simple Quadro P620 graphics card is set as 0. I assume this is the case as the K80 is deprecated in CUDA 11!
You can select the device inside your code with cudaSetDevice or change it when launching the program with
$ CUDA_VISIBLE_DEVICES="1" ./my_app
where 1 has to be replaced by the device id you wish to use. Doing so should make your code run without any problems.
You can also test if this really is the issue this by cloning the Github repository of "Learn CUDA Programming", then browsing Chapter01/01_cuda_introduction/01_hello_world/, compile the make file with $ make and finally run it with $ ./hello_world. It automatically compiles for multiple architectures/compute capabilities and should therefore run without any issue!
I have develop two applications based on tf in c++ language, these applications are served as libraries. In the caller execuable program, library1 is called then library2. In library1 initialization, gpu memory fraction is set to 0.5, run some inference, and session closed. then library2 is called, gpu memory fraction is set to 0.8, but the setting can not work, gpu memory allocation did not change. Both library have the same initialization code but differnet fraction value
int XXXLib::init(double per_process_gpu_memory_fraction)
{
SessionOptions options;
ConfigProto* config = &options.config;
GPUOptions* gpu_options = config->mutable_gpu_options();
// for library1, fraction = 0.5; for library2, fraction = 0.8
gpu_options->set_per_process_gpu_memory_fraction(per_process_gpu_memory_fraction);
Status status = NewSession(options, &_session);
}
It seems that when set_per_process_gpu_memory_fraction() is called, the gpu memory in this process is fixed, even new another Newsession(), the original fraction value is used.
Should different app(library) use different session ?
gpu memory fraction is related to session or to process ?
How to change the fraction in different session but the same process?
Some env info:
Have I written custom code? NO
OS Platform and Distribution? Win10 Pro
TensorFlow installed from? Source code
TensorFlow version? 1.9
CUDA/cuDNN version? CUDA9.0, cudnn 7.05
GPU model and memory? GTX1080 with 8GB memory
It is unfortunate, but in the current TensorFlow (1.11) the GPU memory allocator is created once (per GPU device) - the first time a session is created in the process. Changing per_process_gpu_memory_fraction in the following sessions will not have any effect.
In regards to your library, I would suggest not creating sessions inside it. Ask the user to provide you a session that they configure as they wish. Alternatively, you can just create a graph and return the operations to run. The user can then run them as they see fit.
So here it goes: I wanted to use TensorFlow with GPU on AWS - p2.xlarge plan. Unfortunately, something must have gone wrong and I continue to get:
InvalidArgumentError (see above for traceback): Cannot assign a device to node 'Variable_1': Could not satisfy explicit device specification '/device:GPU:0' because no devices matching that specification are registered in this process; available devices: /job:localhost/replica:0/task:0/cpu:0
I checked both CUDA and cuDNN:
nvcc -V
cat /usr/local/cuda/include/cudnn.h
and got 8.0 and 5.1, respectively.
I call gpu like this:
with tf.device('/gpu:0'):
a = tf.Variable(tf.truncated_normal([100, 100]))
b = tf.Variable(tf.truncated_normal([100, 1000]))
with tf.Session() as sess:
sess.run(tf.matmul(a,b))
happy to post more details if necessary - don't know what will be useful yet.
I suppose you're trying to set up an EC2 instance from scratch? That can be difficult.
Instead, I'd strongly recommend using the Deep Learning AMI (https://aws.amazon.com/machine-learning/amis/). It comes preinstalled with everything you need (drivers, popular DL libraries, etc.). It's also free to use, you just pay for the instance itself.
I am getting the following error when running the default generated kernel when creating a CUDA project in VS Community:
addKernel launch failed: invalid device function
addWithCuda failed!
I searched for how to solve it, and found out that have to change the Project->Properties->CUDA C/C++->Device->Code Generation(default values for [architecture, code] are compute_20,sm_20), but I couldn't find the values needed for my graphic card (GeForce 8400 GS)
Is there any list on the net for the [architecture, code] or is it possible to get them by any command?
The numeric value in compute_XX and sm_XX are the Compute Capability (CC) for your CUDA device.
You can lookup this link http://en.wikipedia.org/wiki/CUDA#Supported_GPUs for a (maybe not complete) list of GPUs and there corresponding CC.
Your quite old 8400 GS (when I remember correctly) hosts a G86 chip which supports CC 1.1.
So you have to change to compute_11,sm_11
`
I recently starting working with OpenCV with the intent of stitching large amounts of images together to create massive panoramas. To begin my experimentation, I looked into the sample programs that come with the OpenCV files to get an idea about how to implement the OpenCV libraries. Since I was interested in image stitching, I went straight for the "stitching_detailed.cpp." The code can be found at:
https://code.ros.org/trac/opencv/browser/trunk/opencv/samples/cpp/stitching_detailed.cpp?rev=6856
Now, this program does most of what I need it to do, but I ran into something interesting. I found that for 9 out of 15 of the optional projection warpers, I receive the following error when I try to run the program:
Insufficient memory (Failed to allocate XXXXXXXXXX bytes) in unknown function,
file C:\slave\winInstallerMegaPack\src\opencv\modules\core\src\alloc.cpp,
line 52
where the "X's" mark integer that change between the different types of projection (as though different methods require different amounts of space). The full source code for "alloc.cpp" can be found at the following website:
https://code.ros.org/trac/opencv/browser/trunk/opencv/modules/core/src/alloc.cpp?rev=3060
However, the line of code that emits this error in alloc.cpp is:
static void* OutOfMemoryError(size_t size)
{
--HERE--> CV_Error_(CV_StsNoMem, ("Failed to allocate %lu bytes", (unsigned long)size));
return 0;
}
So, I am simply lost as to the possible reasons that this error may be occurring. I realize that this error would normally occur if the system is out of memory, but I when running this program with my test images I am never using more that ~3.5GB of RAM, according to my Task Manager.
Also, since the program was written as an sample of the OpenCV stitching capabilities BY OpenCV developers I find it hard to believe that there is a drastic memory error present within the source code.
Finally, the program works fine if I use some of the warping methods:
- spherical
- fisheye
- transverseMercator
- compressedPlanePortraitA2B1
- paniniPortraitA2B1
- paniniPortraitA1.5B1)
but as ask the program to use any of the others (through the command line tag
--warp [PROJECTION_NAME]):
- plane
- cylindrical
- stereographic
- compressedPlaneA2B1
- mercator
- compressedPlaneA1.5B1
- compressedPlanePortraitA1.5B1
- paniniA2B1
- paniniA1.5B1
I get the error mentioned above. I get pretty good results from the transverseMercator project warper, but I would like to test the stereographic in particular. Can anyone help me figure this out?
The pictures that I am trying to process are 1360 x 1024 in resolution and my computer has the following stats:
Model: HP Z800 Workstation
Operating System: Windows 7 enterprise 64-bit OPS
Processor: Intel Xeon 2.40GHz (12 cores)
Memory: 14GB RAM
Hard Drive: 1TB Hitachi
Video Card: ATI FirePro V4800
Any help would be greatly appreciated, thanks!
When I run OpenCV's APP traincascade, i get just the same error as you:
Insufficient memory (Failed to allocate XXXXXXXXXX bytes) in unknown function,
file C:\slave\winInstallerMegaPack\src\opencv\modules\core\src\alloc.cpp,
line 52
at the time, only about 70% pecent of my RAM(6G) was occupied. And when runnig trainscascade step by step, I found that the error would be thrown.when it use about more than 1.5G RAM space.
then, I found the are two arguments which can control how many memory should be used:
-precalcValBufSize
-precalcIdxBufSize
so i tried to set these two to 128, it run. I hope my experience can help you.
I thought this problem is nothing about memory leak, it is just relate to how many memory the OS limits a application occupy. I expect someone can check my guess.
I've recently had a similar issue with OpenCV image stitching. I've used create method to create stitcher instance and provided 5 images in vertical order to stitch method, but I've received insufficient memory error.
Panorama was successfully created after setting:
setWaveCorrection(false)
This solution will not be applicable if you need wave correction.
This may be related to the sequence of the stitching, I split a big picture into 3*3, and firstly I stitch them row by row and there is no problem, when I stitch them column by column, there is the problem same as you.