Miner cudaerror

1. Cudaerror in CUDA_ T this category can record CUDA errors. Almost all CUDA library functions return a cudaerror_ t
using handle in GPU high performance programming_ Error macro function to process cudaerror
static void handleerror (cudaerror) returned by each function_ t err,const char *file,int line ) {
if (err != cudaSuccess) {
printf( "% s in %s at line %d\ n", cudaGetErrorString( err ), file, line );< br />exit( EXIT_ FAILURE );< br />}
}

3. Today, as soon as the server was set up, the program couldn't run. At that time, I almost scared myself to pee
error type: CUDA_ ERROE_ OUT_ OF_ MEMORY
E tensorflow/stream_ executor/cuda/cuda_ driver.cc:924] failed to alloc 17179869184 bytes on host: CUDA_ ERROR_ OUT_ OF_ MEMORY
W ./tensorflow/core/common_ runtime/gpu/pool_ Allocator. H: 195] could not allocate pinned host memory of size: 17179869184
killed

in fact, it is very easy to understand, which roughly means that the GPU size of the server is m
tensorflow, only n (n & lt; M)
in other words, tensorflow tells you that you can't apply for all the resources of GPU, and then quit.
solution:
find the session in the code
Add

config = tf.configproto (allow) before the session definition_ soft_ Placement = true)
# accounting for 70% of GPU resources at most
GPU_ options = tf.GPUOptions(per_ process_ gpu_ memory_ (fraction = 0.7)
# instead of giving tensorflow all GPU resources at first, it will increase
config.gpu on demand_ options.allow_ Growth = true
sess = TF. Session (config = config)
this is no problem

in fact, tensorflow is a greedy tool
even if you use device_ ID specifies that the GPU will also occupy the video memory resources of other GPUs. You must
execute export CUDA before executing the program_ VISIBLE_ Devices = n (n is the visible server number)
to execute Python code. Py will not occupy other GPU resources
recently, before tensorflow, it was Cafe
this week, for three consecutive days, people in the laboratory reported that they occupied too much server resources. It's really tiring. Just use the above method
that is, to execute export CUDA before executing code_ VISIBLE_ Devices = n
only one or a few GPUs can be seen, and other GPUs can't be seen

4. I used the case of CUDA by example to compile. I also encountered this problem and changed the following header file & quot; book.h" It passed. By the way, let's see if the CUDA file attribute of CUDA is CUDA C / C++

6. This error is because your runtime is not complete. There are many runtime libraries needed for the game, such as vc2005 +, etc. the integration package of the next game runtime is very common. Rangers, 3DM or some other websites have it. If you install one, you won't be short of something and cause running errors

7. The r6034 error may be that you have installed anti spyware software such as spyware doctor. After installing such software, it will appear in C: &# 92; windows\ system32\ A file named msvcr80.dll appears in the directory. When the software is unloaded, msvcr80.dll will stay in the directory. You can try C: &; windows\ system32\ Directory of msvcr80.dll to delete, if equipped with 360 security guard such software, delete may appear a prompt, be sure to select allow, otherwise cannot delete. Then click the start menu in the lower left corner of the desktop, click Run, enter CMD, and then enter for% 1 in (% windir% & # 92; system32\*. DLL) do Regsvr32 / S% 1, if you are afraid of wrong input, you can it, and then click enter CMD to pop up the upper left corner of the window C: &# 92;, Click Edit paste, enter and wait for the screen to run. After that, wait for a while until the hard disk indicator no longer flashes, and then restart

8. It has collapsed. Don't believe in the game of capital