This CUDA program queries and displays detailed properties of all CUDA-capable GPUs in your system. It's useful for understanding your hardware capabilities and determining optimal kernel configuration parameters.
The program uses CUDA's device query API to retrieve comprehensive information about each GPU, including memory specifications, compute capabilities, and hardware features. This information is essential for:
- Understanding hardware limitations
- Optimizing kernel configurations
- Ensuring compatibility across different GPU models
- Debugging CUDA applications
Compile and run the program to see your GPU properties:
nvcc kernel.cu -o device_properties
./device_propertiesExample output from an NVIDIA GeForce GTX 1070:
Device Index: 0
Device Name: NVIDIA GeForce GTX 1070
Total Global Memory: 8589737984 bytes
Shared Memory Per Block: 49152 bytes
Registers Per Block: 65536
Warp Size: 32
Memory Pitch: 2147483647 bytes
Max Threads Per Block: 1024
Max Threads Dim: [1024, 1024, 64]
Max Grid Size: [2147483647, 65535, 65535]
Total Constant Memory: 65536 bytes
Major Compute Capability: 6
Minor Compute Capability: 1
Clock Rate: 1645000 kHz
Texture Alignment: 512 bytes
Device Overlap: 1
Multiprocessor Count: 16
Kernel Execution Timeout Enabled: 1
Integrated GPU: 0
Can Map Host Memory: 1
Compute Mode: 0
Max Texture 1D: 131072
Max Texture 2D: [131072, 65536]
Max Texture 3D: [16384, 16384, 16384]
Concurrent Kernels: 1
- Total Global Memory: Available GPU memory for allocations
- Compute Capability: GPU architecture version (major.minor)
- Multiprocessor Count: Number of streaming multiprocessors (SMs)
- Max Threads Per Block: Maximum threads in a single block
- Warp Size: Number of threads executed together (typically 32)
- Shared Memory Per Block: Fast on-chip memory per block