デバイス情報の取得

デバイスプロパティの取得†

デバイスの情報は cudaGetDeviceProperties命令でデバイスプロパティを格納する構造体cudaDeviceProp を介して取得できます．

int n;	//デバイス数
	cutilSafeCall(cudaGetDeviceCount(&n));

	for(int i = 0; i < n; ++i){
		cudaDeviceProp dev;

		// デバイスプロパティ取得
		cutilSafeCall(cudaGetDeviceProperties(&dev, i));

		printf("device %d\n", i);
		printf(" device name : %s\n", dev.name);
		printf(" total global memory : %d (MB)\n", dev.totalGlobalMem/1024/1024);
		printf(" shared memory / block : %d (KB)\n", dev.sharedMemPerBlock/1024);
		printf(" register / block : %d\n", dev.regsPerBlock);
		printf(" warp size : %d\n", dev.warpSize);
		printf(" max pitch : %d (B)\n", dev.memPitch);
		printf(" max threads / block : %d\n", dev.maxThreadsPerBlock);
		printf(" max size of each dim. of block : (%d, %d, %d)\n", dev.maxThreadsDim[0], dev.maxThreadsDim[1], dev.maxThreadsDim[2]);
		printf(" max size of each dim. of grid  : (%d, %d, %d)\n", dev.maxGridSize[0], dev.maxGridSize[1], dev.maxGridSize[2]);
		printf(" clock rate : %d (MHz)\n", dev.clockRate/1000);
		printf(" total constant memory : %d (KB)\n", dev.totalConstMem/1024);
		printf(" compute capability : %d.%d\n", dev.major, dev.minor);
		printf(" alignment requirement for texture : %d\n", dev.textureAlignment);
		printf(" device overlap : %s\n", (dev.deviceOverlap ? "ok" : "not"));
		printf(" num. of multiprocessors : %d\n", dev.multiProcessorCount);
		printf(" kernel execution timeout : %s\n", (dev.kernelExecTimeoutEnabled ? "on" : "off"));
		printf(" integrated : %s\n", (dev.integrated ? "on" : "off"));
		printf(" host memory mapping : %s\n", (dev.canMapHostMemory ? "on" : "off"));

		printf(" compute mode : ");
		if(dev.computeMode == cudaComputeModeDefault) printf("default mode (multiple threads can use) \n");
		else if(dev.computeMode == cudaComputeModeExclusive) printf("exclusive mode (only one thread will be able to use)\n");
		else if(dev.computeMode == cudaComputeModeProhibited) printf("prohibited mode (no threads can use)\n");
		
	}

GeForce GTX 285を搭載したビデオカード上での実行例†

device 0
 device name : GeForce GTX 285
 total global memory : 1024 (MB)
 shared memory / block : 16 (KB)
 register / block : 16384
 warp size : 32
 max pitch : 262144 (B)
 max threads / block : 512
 max size of each dim. of block : (512, 512, 64)
 max size of each dim. of grid  : (65535, 65535, 1)
 clock rate : 1476 (MHz)
 total constant memory : 64 (KB)
 compute capability : 1.3
 alignment requirement for texture : 256
 device overlap : ok
 num. of multiprocessors : 30
 kernel execution timeout : off
 integrated : off
 host memory mapping : off
 compute mode : default mode (multiple threads can use)

GeForce GTX 580を搭載したビデオカード上での実行例†

device 0
 device name : GeForce GTX 580
 total global memory : 1503 (MB)
 shared memory / block : 48 (KB)
 register / block : 32768
 warp size : 32
 max pitch : 2147483647 (B)
 max threads / block : 1024
 max size of each dim. of block : (1024, 1024, 64)
 max size of each dim. of grid  : (65535, 65535, 65535)
 clock rate : 1544 (MHz)
 total constant memory : 64 (KB)
 compute capability : 2.0
 alignment requirement for texture : 512
 device overlap : ok
 num. of multiprocessors : 16
 kernel execution timeout : on
 integrated : off
 host memory mapping : on
 compute mode : default mode (multiple threads can use)

GeForce GTX TITANを搭載したビデオカード上での実行例†

device 0
 device name : GeForce GTX TITAN
 total global memory : 4095 (MB)
 shared memory / block : 48 (KB)
 register / block : 65536
 warp size : 32
 max pitch : 2147483647 (B)
 max threads / block : 1024
 max size of each dim. of block : (1024, 1024, 64)
 max size of each dim. of grid  : (2147483647, 65535, 65535)
 clock rate : 875 (MHz)
 total constant memory : 64 (KB)
 compute capability : 3.5
 alignment requirement for texture : 512
 device overlap : ok
 num. of multiprocessors : 14
 kernel execution timeout : on
 integrated : off
 host memory mapping : on
 compute mode : default mode (multiple threads can use)