GTX 1080Ti with CUDA 9.0 Toolkit

Hello this is sunny!

I used gtx 1080Ti and Visual Studio with CUDA 9.0 Toolkit.

I want to process images on GPU and the size of one image is 23.4MB

I want to ask How many images can I load on GPU?? (GTX 1080 Ti)

The standard memory config in spec Sheet is 11GB GDDR5X.

I attach my code in page~~

for (i = 0; i < cnt; i++) {
GPUDATA* m_gpudata = new GPUDATA();
m_gpudata->imgidx = m_rawImgList[i].imgidx;

HANDLE_ERROR(cudaMalloc((void**)&m_gpudata->gpu_rawimg, size));
HANDLE_ERROR(cudaMalloc((void**)&m_gpudata->gpu_zmapimg, size));
HANDLE_ERROR(cudaMalloc((void**)&m_gpudata->gpu_output, size));



for (i = 0; i < cnt; i++) {
HANDLE_ERROR(cudaMemcpy(m_gpu_ImgList[i]->gpu_rawimg, m_rawImgList[i].rawimg, size, cudaMemcpyHostToDevice));
HANDLE_ERROR(cudaMemcpy(m_gpu_ImgList[i]->gpu_zmapimg, m_rawImgList[i].zonemapimg, size, cudaMemcpyHostToDevice));

What’s your OS platform? If you are on Windows 10 using the WDDM 2.x driver, GPU memory available to CUDA programs will be at most about 81% of the physical memory.

That’s a function of Microsoft’s WDDM 2.x driver architecture best anyone has ever been able to figure out.

My OS is Windows 7 and wddm 1.1 version driver.
And then …
How many images can I load on GPU?? (GTX 1080 Ti)

If I remember correctly, with a WDDM 1.x driver you can utilize about 95% of the physical memory of the GPU. Note that the GPU heap may become fragmented during use, so the total amount of memory you find you can actually allocate may depend somewhat on the granularity of your memory allocations.

It should be simple enough to set up a test scaffold that exercises the same sequence of allocations you plan to make in your actual application; that way you will know for sure.

I tested on my computer by loading the many images on gpu.
However, I can only loaded 3.5GB on GPU (GTX 1080Ti)
Over 3.5GB, The Error Message Poped up => “GPU Out of Memory”
What’s the problem…??
physical memory is 11GB but I can’t use 70% physical memory of GTX 1080Ti…
How can I use most memory of GPU?

Impossible to tell remotely. Among the many possibilities:

(1) Maybe there is a bug in the program you are using to test the amount of memory you are able to allocate.
(2) Maybe there is another application running on the GPU that is using the remaining memory.
(3) Maybe you have multiple GPUs in you system and your application is actually running on the other GPU, not on the GTX 1080 Ti
(4) Maybe you are running a 32-bit version of the Windows operating systems and are therefore limited to a 4 GB address space.
(5) Maybe your GTX 1080 Ti is counterfeit (we have had a few instances in these forums over the years where we were able to collect pretty conclusive evidence that an older or lower-end GPU had been remarked as a newer and more capable one).