With OpenCL getDeviceInfo, you can get the number of compute units available (CL_DEVICE_MAX_COMPUTE_UNITS). On my nVidia Geforce 8600GTS, I have 4 processing units with 8 cores per unit. With getDeviceInfo (... CL_DEVICE_MAX_COMPUTE_UNITS ...) I get 4 as an answer for computational units. But how can I get information about the number of cores per unit of calculation?
There are no allusions to this subject in the OpenCL spec. Does anyone know how to get the number of cores per computational unit in a standard way?
, - API- CUDA . OpenCL, , SIMD API , .
NVIDIA cl_nv_device_attribute_query, CUDA . :
1.0, 1.1, 1.2, 1.3: 8 cores per execution unit 2.0: 32 cores per execution unit 2.1: 48 cores per execution unit
. NVIDIA NVIDIA OpenCL, .