显卡的相关性能参数含义(struct cudaDeviceProp)

中文译注(英文见下文)：

struct cudaDeviceProp {
char name[256];         //器件的名字
size_t totalGlobalMem;    //Global Memory 的byte大小
size_t sharedMemPerBlock;   //线程块可以使用的共用记忆体的最大值。byte为单位，多处理器上的所有线程块可以同时共用这些记忆体
int regsPerBlock;                 //线程块可以使用的32位寄存器的最大值，多处理器上的所有线程快可以同时实用这些寄存器
int warpSize;                    //按线程计算的wrap块大小
size_t memPitch;        //做内存复制是可以容许的最大间距，允许通过cudaMallocPitch（）为包含记忆体区域的记忆提复制函数的最大间距，以byte为单位。
int maxThreadsPerBlock;   //每个块中最大线程数
int maxThreadsDim[3];       //块各维度的最大值
int maxGridSize[3];             //Grid各维度的最大值
size_t totalConstMem;  //常量内存的大小
int major;            //计算能力的主代号
int minor;            //计算能力的次要代号
int clockRate;     //时钟频率
size_t textureAlignment; //纹理的对齐要求
int deviceOverlap;    //器件是否能同时执行cudaMemcpy()和器件的核心代码
int multiProcessorCount; //设备上多处理器的数量
int kernelExecTimeoutEnabled; //是否可以给核心代码的执行时间设置限制
int integrated;                  //这个GPU是否是集成的
int canMapHostMemory; //这个GPU是否可以讲主CPU上的存储映射到GPU器件的地址空间
int computeMode;           //计算模式
int maxTexture1D;          //一维Textures的最大维度  
int maxTexture2D[2];      //二维Textures的最大维度
int maxTexture3D[3];      //三维Textures的最大维度
int maxTexture2DArray[3];     //二维Textures阵列的最大维度
int concurrentKernels;           //GPU是否支持同时执行多个核心程序
}

English Version：

struct __device_builtin__ cudaDeviceProp
{
    char   name[256];                  /**< ASCII string identifying device */
    size_t totalGlobalMem;             /**< Global memory available on device in bytes */
    size_t sharedMemPerBlock;          /**< Shared memory available per block in bytes */
    int    regsPerBlock;               /**< 32-bit registers available per block */
    int    warpSize;                   /**< Warp size in threads */
    size_t memPitch;                   /**< Maximum pitch in bytes allowed by memory copies */
    int    maxThreadsPerBlock;         /**< Maximum number of threads per block */
    int    maxThreadsDim[3];           /**< Maximum size of each dimension of a block */
    int    maxGridSize[3];             /**< Maximum size of each dimension of a grid */
    int    clockRate;                  /**< Clock frequency in kilohertz */
    size_t totalConstMem;              /**< Constant memory available on device in bytes */
    int    major;                      /**< Major compute capability */
    int    minor;                      /**< Minor compute capability */
    size_t textureAlignment;           /**< Alignment requirement for textures */
    size_t texturePitchAlignment;      /**< Pitch alignment requirement for texture references bound to pitched memory */
    int    deviceOverlap;              /**< Device can concurrently copy memory and execute a kernel. Deprecated. Use instead asyncEngineCount. */
    int    multiProcessorCount;        /**< Number of multiprocessors on device */
    int    kernelExecTimeoutEnabled;   /**< Specified whether there is a run time limit on kernels */
    int    integrated;                 /**< Device is integrated as opposed to discrete */
    int    canMapHostMemory;           /**< Device can map host memory with cudaHostAlloc/cudaHostGetDevicePointer */
    int    computeMode;                /**< Compute mode (See ::cudaComputeMode) */
    int    maxTexture1D;               /**< Maximum 1D texture size */
    int    maxTexture1DMipmap;         /**< Maximum 1D mipmapped texture size */
    int    maxTexture1DLinear;         /**< Maximum size for 1D textures bound to linear memory */
    int    maxTexture2D[2];            /**< Maximum 2D texture dimensions */
    int    maxTexture2DMipmap[2];      /**< Maximum 2D mipmapped texture dimensions */
    int    maxTexture2DLinear[3];      /**< Maximum dimensions (width, height, pitch) for 2D textures bound to pitched memory */
    int    maxTexture2DGather[2];      /**< Maximum 2D texture dimensions if texture gather operations have to be performed */
    int    maxTexture3D[3];            /**< Maximum 3D texture dimensions */
    int    maxTexture3DAlt[3];         /**< Maximum alternate 3D texture dimensions */
    int    maxTextureCubemap;          /**< Maximum Cubemap texture dimensions */
    int    maxTexture1DLayered[2];     /**< Maximum 1D layered texture dimensions */
    int    maxTexture2DLayered[3];     /**< Maximum 2D layered texture dimensions */
    int    maxTextureCubemapLayered[2];/**< Maximum Cubemap layered texture dimensions */
    int    maxSurface1D;               /**< Maximum 1D surface size */
    int    maxSurface2D[2];            /**< Maximum 2D surface dimensions */
    int    maxSurface3D[3];            /**< Maximum 3D surface dimensions */
    int    maxSurface1DLayered[2];     /**< Maximum 1D layered surface dimensions */
    int    maxSurface2DLayered[3];     /**< Maximum 2D layered surface dimensions */
    int    maxSurfaceCubemap;          /**< Maximum Cubemap surface dimensions */
    int    maxSurfaceCubemapLayered[2];/**< Maximum Cubemap layered surface dimensions */
    size_t surfaceAlignment;           /**< Alignment requirements for surfaces */
    int    concurrentKernels;          /**< Device can possibly execute multiple kernels concurrently */
    int    ECCEnabled;                 /**< Device has ECC support enabled */
    int    pciBusID;                   /**< PCI bus ID of the device */
    int    pciDeviceID;                /**< PCI device ID of the device */
    int    pciDomainID;                /**< PCI domain ID of the device */
    int    tccDriver;                  /**< 1 if device is a Tesla device using TCC driver, 0 otherwise */
    int    asyncEngineCount;           /**< Number of asynchronous engines */
    int    unifiedAddressing;          /**< Device shares a unified address space with the host */
    int    memoryClockRate;            /**< Peak memory clock frequency in kilohertz */
    int    memoryBusWidth;             /**< Global memory bus width in bits */
    int    l2CacheSize;                /**< Size of L2 cache in bytes */
    int    maxThreadsPerMultiProcessor;/**< Maximum resident threads per multiprocessor */
    int    streamPrioritiesSupported;  /**< Device supports stream priorities */
    int    globalL1CacheSupported;     /**< Device supports caching globals in L1 */
    int    localL1CacheSupported;      /**< Device supports caching locals in L1 */
    size_t sharedMemPerMultiprocessor; /**< Shared memory available per multiprocessor in bytes */
    int    regsPerMultiprocessor;      /**< 32-bit registers available per multiprocessor */
    int    managedMemory;              /**< Device supports allocating managed memory on this system */
    int    isMultiGpuBoard;            /**< Device is on a multi-GPU board */
    int    multiGpuBoardGroupID;       /**< Unique identifier for a group of devices on the same multi-GPU board */
};