3 ms·
addressable is a weird choice of words here. CUDA has had managed memory for a long time now. You absolutely can address the entire host memory from your GPU.
by KeplerBoy 2y ago
addressable is a weird choice of words here.
CUDA has had managed memory for a long time now. You absolutely can address the entire host memory from your GPU. It will fetch it, if it's needed. Not fast, but addressable.
- p_ing 2y agoWindows has been doing this since what... the AGP era? Though this is a function of the ISA rather than the OS.