When I compile and run this executes fine: nvcc copy.cu -o mat & nvprof. Example code:ĬudaFuncSetAttribute(copy1, cudaFuncAttributePreferredSharedMemor圜arveout, cudaSharedmemCarveoutMaxShared) However, when I actually try and do this I seem to be unable to reconfigure it properly. According to the docs (table 15 here), I should be able to configure this later using cudaFuncSetAttribute() to as much as 64kB per block. The total amount of shared memory is listed as 49kB per block.
#MEMORY DIM3 DRIVER#
Supports MultiDevice Co-op Kernel Launch: Yesĭevice PCI Domain ID / Bus ID / location ID: 0 / 10 / 0ĭeviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.2, CUDA Runtime Version = 10.2, NumDevs = 1 Support host page-locked memory mapping: Yesĭevice supports Unified Addressing (UVA): Yes Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)Ĭoncurrent copy and kernel execution: Yes with 3 copy engine(s) Max dimension size of a thread block (x,y,z): (1024, 1024, 64) Maximum number of threads per block: 1024 Maximum number of threads per multiprocessor: 1024 Total number of registers available per block: 65536 **Total amount of shared memory per block: 49152 bytes** Total amount of constant memory: 65536 bytes Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers (68) Multiprocessors, ( 64) CUDA Cores/MP: 4352 CUDA Cores Total amount of global memory: 11017 MBytes (11552096256 bytes) When I run `./deviceQuery`, this is the output I get:ĬUDA Device Query (Runtime API) version (CUDART static linking)ĬUDA Driver Version / Runtime Version 10.2 / 10.2ĬUDA Capability Major/Minor version number: 7.5
I am trying to configure my RTX 2080 TI to use 64kB of shared memory per block, which I have read in the docs should be possible, as my device is cc7.5.
0 kommentar(er)
