Andriy Golovnya

Hi weasleyma, I do not have an access to such a exotic HW like mentioned A100 SMX GPU though I can't test CUDA-Z for this kind of modules ;) I might try to update CUDA SDK to a newer version and hope it will work better with your SMX GPU. No promises though. Best regards, AG

You are right. The direction of data transfer does not metter. If you map GPU memory block to CPU memory you will have the abstraction code to translate your calls in backgroud. I do not know if this is the only limiting factor here, but it shurely a significant one. If you use OpenGL, the use of cudaMemcpy() in your code will be most probably useless. OpenGL and CUDA are two non-crossing subsystems, as I know. You have to find something like this in OpenGL itself.

Hi! You cannot copy data from CPU memory to GPU memory with normal memcpy() routine directly. The reason for this is that beween CPU and GPU you have a PCIe bus. You have to use a special bus transfer routines like cudaMemcpy(). If you ARE able to use normal memcpy() routine to transfer data to GPU memory, you are clearly dealing with some emulation layer which detects change you made in the CPU memory page and translates them (probably not efficient) to bus transfer calls. Try to use native bus...

CUDA-Z not opening

No CUDA Devices found (sudo required) Mac

Hi! I can't offer my support here because I have no real mac (and no hackintosh) in my possession for now. I'll try to fix CUDA-Z for mac or I'll drop it completly. WBR, Ag

Build error

Build error