cuDNN Error ARCH_MISMATCH on a Kepler GPU
I'm just running the Alea TK samples. As long as I don't activate cuDNN things work will except for the CNN, since it uses Pooling2D which relies on cuDNN.
If I now activate cuDNN (via app.config) I get this error [cuDNN Error] ARCH_MISMATCH as soon as I call Context.GpuContext(0).
If I run Context.GpuContext(0) without cuDNN I can clearly see, it's my 'GT 730' which should have compute 3.0 capabilities.
Any idea, what's going wrong here?
There are two different versions of the GT-730, and they don't have the same CC :
GeForce GT 730 : 3.5
GeForce GT 730 DDR3, 128bit : 2.1
Source : CUDA GPUs, select "CUDA-Enabled GeForce Products" category.
If you have, as said by talonmies, the DDR3 & 128bit version, your CC is not sufficient for what you are trying to achieve.
You can either change your requirements (using other software, writing your own...) or change your GeForce graphics card.
- Makefile variable substitution sometimes ignored
- Cupy with dynamically allocated shared memory
- What is the difference when compiling a CUDA program with or without option -g -G
- CMake does not properly find CUDA library
- Does PTX (8.4) not cover smaller-shape WMMA instructions?
- CUDA 4.2 nvcc (cicc) error while compiling with sm_21
- Why does CUDA.rules have two identical command lines
- Why (x / y)[i] faster than x[i] / y[i]?
- Template excessive recursion at instantiation cuda
- Is it "worth it" to reuse events in CUDA?
- Clarifying memory transactions in CUDA
- Cuda C++ Accessing struct from device global memory in kernel causes illegal memory access
- nvcc version remains the same after updating the CUDA Toolkit version
- How does one compile a CUDA Toolkit 4.0 RC2 program under VS2010 or VS2008?
- How to compile CUDA C files and Nvidia OptiX files inside the same Visual Studio project
- Why is CUDA function cudaLaunchKernel passed a function pointer to host-code function?
- Resetting GPU and driver after CUDA error
- How is stack frame managed within a thread in Cuda?
- How can I convert an integer to CUDA's __half FP16 type, in a constexpr fashion?
- Dump/inspect NVIDIA GPU global memory contents corresponding to arbitrary (but not invalid) addresses
- Poetry PyTorch dependency exclude cuda as I want to use the system cuda
- Cuda nvJitLink error because fatbin does not contains the correct function
- Vectorized Memory Stores Reduce Load Instructions
- How to properly free a Cuda context?
- How can I flush GPU memory using CUDA (physical reset is unavailable)
- Can CUDA unified memory be written to by another CPU thread?
- CUDA image upsampling with FFT method
- how does cv::cuda::GpuMat turn into cv::cuda::PtrStepSz when passed to a kernel?
- Cuda with multiple GPUs: host and device report different device numbers
- Weird mistake while compilation